Documentation
Kafka 3.6 Documentation
Prior releases: 0.7.x, 0.8.0, 0.8.1.X, 0.8.2.X, 0.9.0.X, 0.10.0.X, 0.10.1.X, 0.10.2.X, 0.11.0.X, 1.0.X, 1.1.X, 2.0.X, 2.1.X, 2.2.X, 2.3.X, 2.4.X, 2.5.X, 2.6.X, 2.7.X, 2.8.X, 3.0.X, 3.1.X, 3.2.X, 3.3.X, 3.4.X, 3.5.X.1. Getting Started
1.1 Introduction
1.2 Use Cases
Here is a description of a few of the popular use cases for Apache Kafka®. For an overview of a number of these areas in action, see this blog post.
Messaging
Kafka works well as a replacement for a more traditional message broker. Message brokers are used for a variety of reasons (to decouple processing from data producers, to buffer unprocessed messages, etc). In comparison to most messaging systems Kafka has better throughput, built-in partitioning, replication, and fault-tolerance which makes it a good solution for large scale message processing applications.In our experience messaging uses are often comparatively low-throughput, but may require low end-to-end latency and often depend on the strong durability guarantees Kafka provides.
In this domain Kafka is comparable to traditional messaging systems such as ActiveMQ or RabbitMQ.
Website Activity Tracking
The original use case for Kafka was to be able to rebuild a user activity tracking pipeline as a set of real-time publish-subscribe feeds. This means site activity (page views, searches, or other actions users may take) is published to central topics with one topic per activity type. These feeds are available for subscription for a range of use cases including real-time processing, real-time monitoring, and loading into Hadoop or offline data warehousing systems for offline processing and reporting.Activity tracking is often very high volume as many activity messages are generated for each user page view.
Metrics
Kafka is often used for operational monitoring data. This involves aggregating statistics from distributed applications to produce centralized feeds of operational data.Log Aggregation
Many people use Kafka as a replacement for a log aggregation solution. Log aggregation typically collects physical log files off servers and puts them in a central place (a file server or HDFS perhaps) for processing. Kafka abstracts away the details of files and gives a cleaner abstraction of log or event data as a stream of messages. This allows for lower-latency processing and easier support for multiple data sources and distributed data consumption. In comparison to log-centric systems like Scribe or Flume, Kafka offers equally good performance, stronger durability guarantees due to replication, and much lower end-to-end latency.Stream Processing
Many users of Kafka process data in processing pipelines consisting of multiple stages, where raw input data is consumed from Kafka topics and then aggregated, enriched, or otherwise transformed into new topics for further consumption or follow-up processing. For example, a processing pipeline for recommending news articles might crawl article content from RSS feeds and publish it to an "articles" topic; further processing might normalize or deduplicate this content and publish the cleansed article content to a new topic; a final processing stage might attempt to recommend this content to users. Such processing pipelines create graphs of real-time data flows based on the individual topics. Starting in 0.10.0.0, a light-weight but powerful stream processing library called Kafka Streams is available in Apache Kafka to perform such data processing as described above. Apart from Kafka Streams, alternative open source stream processing tools include Apache Storm and Apache Samza.Event Sourcing
Event sourcing is a style of application design where state changes are logged as a time-ordered sequence of records. Kafka's support for very large stored log data makes it an excellent backend for an application built in this style.Commit Log
Kafka can serve as a kind of external commit-log for a distributed system. The log helps replicate data between nodes and acts as a re-syncing mechanism for failed nodes to restore their data. The log compaction feature in Kafka helps support this usage. In this usage Kafka is similar to Apache BookKeeper project.1.3 Quick Start
1.4 Ecosystem
There are a plethora of tools that integrate with Kafka outside the main distribution. The ecosystem page lists many of these, including stream processing systems, Hadoop integration, monitoring, and deployment tools.1.5 Upgrading From Previous Versions
2. APIs
3. Configuration
4. Design
5. Implementation
6. Operations
Here is some information on actually running Kafka as a production system based on usage and experience at LinkedIn. Please send us any additional tips you know of.6.1 Basic Kafka Operations
This section will review the most common operations you will perform on your Kafka cluster. All of the tools reviewed in this section are available under thebin/ directory of the Kafka distribution and each tool will print details on all possible commandline options if it is run with no arguments.
Adding and removing topics
You have the option of either adding topics manually or having them be created automatically when data is first published to a non-existent topic. If topics are auto-created then you may want to tune the default topic configurations used for auto-created topics.Topics are added and modified using the topic tool:
> bin/kafka-topics.sh --bootstrap-server broker_host:port --create --topic my_topic_name \
--partitions 20 --replication-factor 3 --config x=yThe partition count controls how many logs the topic will be sharded into. There are several impacts of the partition count. First each partition must fit entirely on a single server. So if you have 20 partitions the full data set (and read and write load) will be handled by no more than 20 servers (not counting replicas). Finally the partition count impacts the maximum parallelism of your consumers. This is discussed in greater detail in the concepts section.
Each sharded partition log is placed into its own folder under the Kafka log directory. The name of such folders consists of the topic name, appended by a dash (-) and the partition id. Since a typical folder name can not be over 255 characters long, there will be a limitation on the length of topic names. We assume the number of partitions will not ever be above 100,000. Therefore, topic names cannot be longer than 249 characters. This leaves just enough room in the folder name for a dash and a potentially 5 digit long partition id.
The configurations added on the command line override the default settings the server has for things like the length of time data should be retained. The complete set of per-topic configurations is documented here.
Modifying topics
You can change the configuration or partitioning of a topic using the same topic tool.To add partitions you can do
> bin/kafka-topics.sh --bootstrap-server broker_host:port --alter --topic my_topic_name \
--partitions 40hash(key) % number_of_partitions then this partitioning will potentially be shuffled by adding partitions but Kafka will not attempt to automatically redistribute data in any way.
To add configs:
> bin/kafka-configs.sh --bootstrap-server broker_host:port --entity-type topics --entity-name my_topic_name --alter --add-config x=y > bin/kafka-configs.sh --bootstrap-server broker_host:port --entity-type topics --entity-name my_topic_name --alter --delete-config x > bin/kafka-topics.sh --bootstrap-server broker_host:port --delete --topic my_topic_nameKafka does not currently support reducing the number of partitions for a topic.
Instructions for changing the replication factor of a topic can be found here.
Graceful shutdown
The Kafka cluster will automatically detect any broker shutdown or failure and elect new leaders for the partitions on that machine. This will occur whether a server fails or it is brought down intentionally for maintenance or configuration changes. For the latter cases Kafka supports a more graceful mechanism for stopping a server than just killing it. When a server is stopped gracefully it has two optimizations it will take advantage of:- It will sync all its logs to disk to avoid needing to do any log recovery when it restarts (i.e. validating the checksum for all messages in the tail of the log). Log recovery takes time so this speeds up intentional restarts.
- It will migrate any partitions the server is the leader for to other replicas prior to shutting down. This will make the leadership transfer faster and minimize the time each partition is unavailable to a few milliseconds.
controlled.shutdown.enable=trueBalancing leadership
Whenever a broker stops or crashes, leadership for that broker's partitions transfers to other replicas. When the broker is restarted it will only be a follower for all its partitions, meaning it will not be used for client reads and writes.To avoid this imbalance, Kafka has a notion of preferred replicas. If the list of replicas for a partition is 1,5,9 then node 1 is preferred as the leader to either node 5 or 9 because it is earlier in the replica list. By default the Kafka cluster will try to restore leadership to the preferred replicas. This behaviour is configured with:
auto.leader.rebalance.enable=true > bin/kafka-leader-election.sh --bootstrap-server broker_host:port --election-type preferred --all-topic-partitionsBalancing Replicas Across Racks
The rack awareness feature spreads replicas of the same partition across different racks. This extends the guarantees Kafka provides for broker-failure to cover rack-failure, limiting the risk of data loss should all the brokers on a rack fail at once. The feature can also be applied to other broker groupings such as availability zones in EC2. You can specify that a broker belongs to a particular rack by adding a property to the broker config: broker.rack=my-rack-idMirroring data between clusters & Geo-replication
Kafka administrators can define data flows that cross the boundaries of individual Kafka clusters, data centers, or geographical regions. Please refer to the section on Geo-Replication for further information.
Checking consumer position
Sometimes it's useful to see the position of your consumers. We have a tool that will show the position of all consumers in a consumer group as well as how far behind the end of the log they are. To run this tool on a consumer group named my-group consuming a topic named my-topic would look like this: > bin/kafka-consumer-groups.sh --bootstrap-server localhost:9092 --describe --group my-group
TOPIC PARTITION CURRENT-OFFSET LOG-END-OFFSET LAG CONSUMER-ID HOST CLIENT-ID
my-topic 0 2 4 2 consumer-1-029af89c-873c-4751-a720-cefd41a669d6 /127.0.0.1 consumer-1
my-topic 1 2 3 1 consumer-1-029af89c-873c-4751-a720-cefd41a669d6 /127.0.0.1 consumer-1
my-topic 2 2 3 1 consumer-2-42c1abd4-e3b2-425d-a8bb-e1ea49b29bb2 /127.0.0.1 consumer-2Managing Consumer Groups
With the ConsumerGroupCommand tool, we can list, describe, or delete the consumer groups. The consumer group can be deleted manually, or automatically when the last committed offset for that group expires. Manual deletion works only if the group does not have any active members. For example, to list all consumer groups across all topics: > bin/kafka-consumer-groups.sh --bootstrap-server localhost:9092 --list
test-consumer-group > bin/kafka-consumer-groups.sh --bootstrap-server localhost:9092 --describe --group my-group
TOPIC PARTITION CURRENT-OFFSET LOG-END-OFFSET LAG CONSUMER-ID HOST CLIENT-ID
topic3 0 241019 395308 154289 consumer2-e76ea8c3-5d30-4299-9005-47eb41f3d3c4 /127.0.0.1 consumer2
topic2 1 520678 803288 282610 consumer2-e76ea8c3-5d30-4299-9005-47eb41f3d3c4 /127.0.0.1 consumer2
topic3 1 241018 398817 157799 consumer2-e76ea8c3-5d30-4299-9005-47eb41f3d3c4 /127.0.0.1 consumer2
topic1 0 854144 855809 1665 consumer1-3fc8d6f1-581a-4472-bdf3-3515b4aee8c1 /127.0.0.1 consumer1
topic2 0 460537 803290 342753 consumer1-3fc8d6f1-581a-4472-bdf3-3515b4aee8c1 /127.0.0.1 consumer1
topic3 2 243655 398812 155157 consumer4-117fe4d3-c6c1-4178-8ee9-eb4a3954bee0 /127.0.0.1 consumer4- --members: This option provides the list of all active members in the consumer group.
> bin/kafka-consumer-groups.sh --bootstrap-server localhost:9092 --describe --group my-group --members CONSUMER-ID HOST CLIENT-ID #PARTITIONS consumer1-3fc8d6f1-581a-4472-bdf3-3515b4aee8c1 /127.0.0.1 consumer1 2 consumer4-117fe4d3-c6c1-4178-8ee9-eb4a3954bee0 /127.0.0.1 consumer4 1 consumer2-e76ea8c3-5d30-4299-9005-47eb41f3d3c4 /127.0.0.1 consumer2 3 consumer3-ecea43e4-1f01-479f-8349-f9130b75d8ee /127.0.0.1 consumer3 0 - --members --verbose: On top of the information reported by the "--members" options above, this option also provides the partitions assigned to each member.
> bin/kafka-consumer-groups.sh --bootstrap-server localhost:9092 --describe --group my-group --members --verbose CONSUMER-ID HOST CLIENT-ID #PARTITIONS ASSIGNMENT consumer1-3fc8d6f1-581a-4472-bdf3-3515b4aee8c1 /127.0.0.1 consumer1 2 topic1(0), topic2(0) consumer4-117fe4d3-c6c1-4178-8ee9-eb4a3954bee0 /127.0.0.1 consumer4 1 topic3(2) consumer2-e76ea8c3-5d30-4299-9005-47eb41f3d3c4 /127.0.0.1 consumer2 3 topic2(1), topic3(0,1) consumer3-ecea43e4-1f01-479f-8349-f9130b75d8ee /127.0.0.1 consumer3 0 - - --offsets: This is the default describe option and provides the same output as the "--describe" option.
- --state: This option provides useful group-level information.
> bin/kafka-consumer-groups.sh --bootstrap-server localhost:9092 --describe --group my-group --state COORDINATOR (ID) ASSIGNMENT-STRATEGY STATE #MEMBERS localhost:9092 (0) range Stable 4
> bin/kafka-consumer-groups.sh --bootstrap-server localhost:9092 --delete --group my-group --group my-other-group
Deletion of requested consumer groups ('my-group', 'my-other-group') was successful.To reset offsets of a consumer group, "--reset-offsets" option can be used. This option supports one consumer group at the time. It requires defining following scopes: --all-topics or --topic. One scope must be selected, unless you use '--from-file' scenario. Also, first make sure that the consumer instances are inactive. See KIP-122 for more details.
It has 3 execution options:
- (default) to display which offsets to reset.
- --execute : to execute --reset-offsets process.
- --export : to export the results to a CSV format.
--reset-offsets also has following scenarios to choose from (at least one scenario must be selected):
- --to-datetime <String: datetime> : Reset offsets to offsets from datetime. Format: 'YYYY-MM-DDTHH:mm:SS.sss'
- --to-earliest : Reset offsets to earliest offset.
- --to-latest : Reset offsets to latest offset.
- --shift-by <Long: number-of-offsets> : Reset offsets shifting current offset by 'n', where 'n' can be positive or negative.
- --from-file : Reset offsets to values defined in CSV file.
- --to-current : Resets offsets to current offset.
- --by-duration <String: duration> : Reset offsets to offset by duration from current timestamp. Format: 'PnDTnHnMnS'
- --to-offset : Reset offsets to a specific offset.
For example, to reset offsets of a consumer group to the latest offset:
> bin/kafka-consumer-groups.sh --bootstrap-server localhost:9092 --reset-offsets --group consumergroup1 --topic topic1 --to-latest
TOPIC PARTITION NEW-OFFSET
topic1 0 0
If you are using the old high-level consumer and storing the group metadata in ZooKeeper (i.e. offsets.storage=zookeeper), pass
--zookeeper instead of --bootstrap-server:
> bin/kafka-consumer-groups.sh --zookeeper localhost:2181 --listExpanding your cluster
Adding servers to a Kafka cluster is easy, just assign them a unique broker id and start up Kafka on your new servers. However these new servers will not automatically be assigned any data partitions, so unless partitions are moved to them they won't be doing any work until new topics are created. So usually when you add machines to your cluster you will want to migrate some existing data to these machines.The process of migrating data is manually initiated but fully automated. Under the covers what happens is that Kafka will add the new server as a follower of the partition it is migrating and allow it to fully replicate the existing data in that partition. When the new server has fully replicated the contents of this partition and joined the in-sync replica one of the existing replicas will delete their partition's data.
The partition reassignment tool can be used to move partitions across brokers. An ideal partition distribution would ensure even data load and partition sizes across all brokers. The partition reassignment tool does not have the capability to automatically study the data distribution in a Kafka cluster and move partitions around to attain an even load distribution. As such, the admin has to figure out which topics or partitions should be moved around.
The partition reassignment tool can run in 3 mutually exclusive modes:
- --generate: In this mode, given a list of topics and a list of brokers, the tool generates a candidate reassignment to move all partitions of the specified topics to the new brokers. This option merely provides a convenient way to generate a partition reassignment plan given a list of topics and target brokers.
- --execute: In this mode, the tool kicks off the reassignment of partitions based on the user provided reassignment plan. (using the --reassignment-json-file option). This can either be a custom reassignment plan hand crafted by the admin or provided by using the --generate option
- --verify: In this mode, the tool verifies the status of the reassignment for all partitions listed during the last --execute. The status can be either of successfully completed, failed or in progress
Automatically migrating data to new machines
The partition reassignment tool can be used to move some topics off of the current set of brokers to the newly added brokers. This is typically useful while expanding an existing cluster since it is easier to move entire topics to the new set of brokers, than moving one partition at a time. When used to do this, the user should provide a list of topics that should be moved to the new set of brokers and a target list of new brokers. The tool then evenly distributes all partitions for the given list of topics across the new set of brokers. During this move, the replication factor of the topic is kept constant. Effectively the replicas for all partitions for the input list of topics are moved from the old set of brokers to the newly added brokers.For instance, the following example will move all partitions for topics foo1,foo2 to the new set of brokers 5,6. At the end of this move, all partitions for topics foo1 and foo2 will only exist on brokers 5,6.
Since the tool accepts the input list of topics as a json file, you first need to identify the topics you want to move and create the json file as follows:
> cat topics-to-move.json
{"topics": [{"topic": "foo1"},
{"topic": "foo2"}],
"version":1
} > bin/kafka-reassign-partitions.sh --bootstrap-server localhost:9092 --topics-to-move-json-file topics-to-move.json --broker-list "5,6" --generate
Current partition replica assignment
{"version":1,
"partitions":[{"topic":"foo1","partition":0,"replicas":[2,1]},
{"topic":"foo1","partition":1,"replicas":[1,3]},
{"topic":"foo1","partition":2,"replicas":[3,4]},
{"topic":"foo2","partition":0,"replicas":[4,2]},
{"topic":"foo2","partition":1,"replicas":[2,1]},
{"topic":"foo2","partition":2,"replicas":[1,3]}]
}
Proposed partition reassignment configuration
{"version":1,
"partitions":[{"topic":"foo1","partition":0,"replicas":[6,5]},
{"topic":"foo1","partition":1,"replicas":[5,6]},
{"topic":"foo1","partition":2,"replicas":[6,5]},
{"topic":"foo2","partition":0,"replicas":[5,6]},
{"topic":"foo2","partition":1,"replicas":[6,5]},
{"topic":"foo2","partition":2,"replicas":[5,6]}]
}The tool generates a candidate assignment that will move all partitions from topics foo1,foo2 to brokers 5,6. Note, however, that at this point, the partition movement has not started, it merely tells you the current assignment and the proposed new assignment. The current assignment should be saved in case you want to rollback to it. The new assignment should be saved in a json file (e.g. expand-cluster-reassignment.json) to be input to the tool with the --execute option as follows:
> bin/kafka-reassign-partitions.sh --bootstrap-server localhost:9092 --reassignment-json-file expand-cluster-reassignment.json --execute
Current partition replica assignment
{"version":1,
"partitions":[{"topic":"foo1","partition":0,"replicas":[2,1]},
{"topic":"foo1","partition":1,"replicas":[1,3]},
{"topic":"foo1","partition":2,"replicas":[3,4]},
{"topic":"foo2","partition":0,"replicas":[4,2]},
{"topic":"foo2","partition":1,"replicas":[2,1]},
{"topic":"foo2","partition":2,"replicas":[1,3]}]
}
Save this to use as the --reassignment-json-file option during rollback
Successfully started partition reassignments for foo1-0,foo1-1,foo1-2,foo2-0,foo2-1,foo2-2
Finally, the --verify option can be used with the tool to check the status of the partition reassignment. Note that the same expand-cluster-reassignment.json (used with the --execute option) should be used with the --verify option:
> bin/kafka-reassign-partitions.sh --bootstrap-server localhost:9092 --reassignment-json-file expand-cluster-reassignment.json --verify
Status of partition reassignment:
Reassignment of partition [foo1,0] is completed
Reassignment of partition [foo1,1] is still in progress
Reassignment of partition [foo1,2] is still in progress
Reassignment of partition [foo2,0] is completed
Reassignment of partition [foo2,1] is completed
Reassignment of partition [foo2,2] is completedCustom partition assignment and migration
The partition reassignment tool can also be used to selectively move replicas of a partition to a specific set of brokers. When used in this manner, it is assumed that the user knows the reassignment plan and does not require the tool to generate a candidate reassignment, effectively skipping the --generate step and moving straight to the --execute stepFor instance, the following example moves partition 0 of topic foo1 to brokers 5,6 and partition 1 of topic foo2 to brokers 2,3:
The first step is to hand craft the custom reassignment plan in a json file:
> cat custom-reassignment.json
{"version":1,"partitions":[{"topic":"foo1","partition":0,"replicas":[5,6]},{"topic":"foo2","partition":1,"replicas":[2,3]}]} > bin/kafka-reassign-partitions.sh --bootstrap-server localhost:9092 --reassignment-json-file custom-reassignment.json --execute
Current partition replica assignment
{"version":1,
"partitions":[{"topic":"foo1","partition":0,"replicas":[1,2]},
{"topic":"foo2","partition":1,"replicas":[3,4]}]
}
Save this to use as the --reassignment-json-file option during rollback
Successfully started partition reassignments for foo1-0,foo2-1
The --verify option can be used with the tool to check the status of the partition reassignment. Note that the same custom-reassignment.json (used with the --execute option) should be used with the --verify option:
> bin/kafka-reassign-partitions.sh --bootstrap-server localhost:9092 --reassignment-json-file custom-reassignment.json --verify
Status of partition reassignment:
Reassignment of partition [foo1,0] is completed
Reassignment of partition [foo2,1] is completedDecommissioning brokers
The partition reassignment tool does not have the ability to automatically generate a reassignment plan for decommissioning brokers yet. As such, the admin has to come up with a reassignment plan to move the replica for all partitions hosted on the broker to be decommissioned, to the rest of the brokers. This can be relatively tedious as the reassignment needs to ensure that all the replicas are not moved from the decommissioned broker to only one other broker. To make this process effortless, we plan to add tooling support for decommissioning brokers in the future.Increasing replication factor
Increasing the replication factor of an existing partition is easy. Just specify the extra replicas in the custom reassignment json file and use it with the --execute option to increase the replication factor of the specified partitions.For instance, the following example increases the replication factor of partition 0 of topic foo from 1 to 3. Before increasing the replication factor, the partition's only replica existed on broker 5. As part of increasing the replication factor, we will add more replicas on brokers 6 and 7.
The first step is to hand craft the custom reassignment plan in a json file:
> cat increase-replication-factor.json
{"version":1,
"partitions":[{"topic":"foo","partition":0,"replicas":[5,6,7]}]} > bin/kafka-reassign-partitions.sh --bootstrap-server localhost:9092 --reassignment-json-file increase-replication-factor.json --execute
Current partition replica assignment
{"version":1,
"partitions":[{"topic":"foo","partition":0,"replicas":[5]}]}
Save this to use as the --reassignment-json-file option during rollback
Successfully started partition reassignment for foo-0The --verify option can be used with the tool to check the status of the partition reassignment. Note that the same increase-replication-factor.json (used with the --execute option) should be used with the --verify option:
> bin/kafka-reassign-partitions.sh --bootstrap-server localhost:9092 --reassignment-json-file increase-replication-factor.json --verify
Status of partition reassignment:
Reassignment of partition [foo,0] is completed > bin/kafka-topics.sh --bootstrap-server localhost:9092 --topic foo --describe
Topic:foo PartitionCount:1 ReplicationFactor:3 Configs:
Topic: foo Partition: 0 Leader: 5 Replicas: 5,6,7 Isr: 5,6,7Limiting Bandwidth Usage during Data Migration
Kafka lets you apply a throttle to replication traffic, setting an upper bound on the bandwidth used to move replicas from machine to machine. This is useful when rebalancing a cluster, bootstrapping a new broker or adding or removing brokers, as it limits the impact these data-intensive operations will have on users. There are two interfaces that can be used to engage a throttle. The simplest, and safest, is to apply a throttle when invoking the kafka-reassign-partitions.sh, but kafka-configs.sh can also be used to view and alter the throttle values directly. So for example, if you were to execute a rebalance, with the below command, it would move partitions at no more than 50MB/s.$ bin/kafka-reassign-partitions.sh --bootstrap-server localhost:9092 --execute --reassignment-json-file bigger-cluster.json --throttle 50000000When you execute this script you will see the throttle engage:
The inter-broker throttle limit was set to 50000000 B/s
Successfully started partition reassignment for foo1-0Should you wish to alter the throttle, during a rebalance, say to increase the throughput so it completes quicker, you can do this by re-running the execute command with the --additional option passing the same reassignment-json-file:
$ bin/kafka-reassign-partitions.sh --bootstrap-server localhost:9092 --additional --execute --reassignment-json-file bigger-cluster.json --throttle 700000000 The inter-broker throttle limit was set to 700000000 B/s
Once the rebalance completes the administrator can check the status of the rebalance using the --verify option. If the rebalance has completed, the throttle will be removed via the --verify command. It is important that administrators remove the throttle in a timely manner once rebalancing completes by running the command with the --verify option. Failure to do so could cause regular replication traffic to be throttled.
When the --verify option is executed, and the reassignment has completed, the script will confirm that the throttle was removed:
> bin/kafka-reassign-partitions.sh --bootstrap-server localhost:9092 --verify --reassignment-json-file bigger-cluster.json
Status of partition reassignment:
Reassignment of partition [my-topic,1] is completed
Reassignment of partition [my-topic,0] is completed
Clearing broker-level throttles on brokers 1,2,3
Clearing topic-level throttles on topic my-topicThe administrator can also validate the assigned configs using the kafka-configs.sh. There are two pairs of throttle configuration used to manage the throttling process. First pair refers to the throttle value itself. This is configured, at a broker level, using the dynamic properties:
leader.replication.throttled.rate
follower.replication.throttled.rateThen there is the configuration pair of enumerated sets of throttled replicas:
leader.replication.throttled.replicas
follower.replication.throttled.replicasWhich are configured per topic.
All four config values are automatically assigned by kafka-reassign-partitions.sh (discussed below).
To view the throttle limit configuration:
> bin/kafka-configs.sh --describe --bootstrap-server localhost:9092 --entity-type brokers
Configs for brokers '2' are leader.replication.throttled.rate=700000000,follower.replication.throttled.rate=700000000
Configs for brokers '1' are leader.replication.throttled.rate=700000000,follower.replication.throttled.rate=700000000This shows the throttle applied to both leader and follower side of the replication protocol. By default both sides are assigned the same throttled throughput value.
To view the list of throttled replicas:
> bin/kafka-configs.sh --describe --bootstrap-server localhost:9092 --entity-type topics
Configs for topic 'my-topic' are leader.replication.throttled.replicas=1:102,0:101,
follower.replication.throttled.replicas=1:101,0:102Here we see the leader throttle is applied to partition 1 on broker 102 and partition 0 on broker 101. Likewise the follower throttle is applied to partition 1 on broker 101 and partition 0 on broker 102.
By default kafka-reassign-partitions.sh will apply the leader throttle to all replicas that exist before the rebalance, any one of which might be leader. It will apply the follower throttle to all move destinations. So if there is a partition with replicas on brokers 101,102, being reassigned to 102,103, a leader throttle, for that partition, would be applied to 101,102 and a follower throttle would be applied to 103 only.
If required, you can also use the --alter switch on kafka-configs.sh to alter the throttle configurations manually.
Safe usage of throttled replication
Some care should be taken when using throttled replication. In particular:
(1) Throttle Removal:
The throttle should be removed in a timely manner once reassignment completes (by running kafka-reassign-partitions.sh --verify).(2) Ensuring Progress:
If the throttle is set too low, in comparison to the incoming write rate, it is possible for replication to not make progress. This occurs when:
max(BytesInPerSec) > throttle
Where BytesInPerSec is the metric that monitors the write throughput of producers into each broker.
The administrator can monitor whether replication is making progress, during the rebalance, using the metric:
kafka.server:type=FetcherLagMetrics,name=ConsumerLag,clientId=([-.\w]+),topic=([-.\w]+),partition=([0-9]+)
The lag should constantly decrease during replication. If the metric does not decrease the administrator should increase the throttle throughput as described above.
Setting quotas
Quotas overrides and defaults may be configured at (user, client-id), user or client-id levels as described here. By default, clients receive an unlimited quota. It is possible to set custom quotas for each (user, client-id), user or client-id group.Configure custom quota for (user=user1, client-id=clientA):
> bin/kafka-configs.sh --bootstrap-server localhost:9092 --alter --add-config 'producer_byte_rate=1024,consumer_byte_rate=2048,request_percentage=200' --entity-type users --entity-name user1 --entity-type clients --entity-name clientA
Updated config for entity: user-principal 'user1', client-id 'clientA'. > bin/kafka-configs.sh --bootstrap-server localhost:9092 --alter --add-config 'producer_byte_rate=1024,consumer_byte_rate=2048,request_percentage=200' --entity-type users --entity-name user1
Updated config for entity: user-principal 'user1'. > bin/kafka-configs.sh --bootstrap-server localhost:9092 --alter --add-config 'producer_byte_rate=1024,consumer_byte_rate=2048,request_percentage=200' --entity-type clients --entity-name clientA
Updated config for entity: client-id 'clientA'.Configure default client-id quota for user=userA:
> bin/kafka-configs.sh --bootstrap-server localhost:9092 --alter --add-config 'producer_byte_rate=1024,consumer_byte_rate=2048,request_percentage=200' --entity-type users --entity-name user1 --entity-type clients --entity-default
Updated config for entity: user-principal 'user1', default client-id. > bin/kafka-configs.sh --bootstrap-server localhost:9092 --alter --add-config 'producer_byte_rate=1024,consumer_byte_rate=2048,request_percentage=200' --entity-type users --entity-default
Updated config for entity: default user-principal. > bin/kafka-configs.sh --bootstrap-server localhost:9092 --alter --add-config 'producer_byte_rate=1024,consumer_byte_rate=2048,request_percentage=200' --entity-type clients --entity-default
Updated config for entity: default client-id. > bin/kafka-configs.sh --bootstrap-server localhost:9092 --describe --entity-type users --entity-name user1 --entity-type clients --entity-name clientA
Configs for user-principal 'user1', client-id 'clientA' are producer_byte_rate=1024,consumer_byte_rate=2048,request_percentage=200 > bin/kafka-configs.sh --bootstrap-server localhost:9092 --describe --entity-type users --entity-name user1
Configs for user-principal 'user1' are producer_byte_rate=1024,consumer_byte_rate=2048,request_percentage=200 > bin/kafka-configs.sh --bootstrap-server localhost:9092 --describe --entity-type clients --entity-name clientA
Configs for client-id 'clientA' are producer_byte_rate=1024,consumer_byte_rate=2048,request_percentage=200 > bin/kafka-configs.sh --bootstrap-server localhost:9092 --describe --entity-type users
Configs for user-principal 'user1' are producer_byte_rate=1024,consumer_byte_rate=2048,request_percentage=200
Configs for default user-principal are producer_byte_rate=1024,consumer_byte_rate=2048,request_percentage=200 > bin/kafka-configs.sh --bootstrap-server localhost:9092 --describe --entity-type users --entity-type clients
Configs for user-principal 'user1', default client-id are producer_byte_rate=1024,consumer_byte_rate=2048,request_percentage=200
Configs for user-principal 'user1', client-id 'clientA' are producer_byte_rate=1024,consumer_byte_rate=2048,request_percentage=2006.2 Datacenters
Some deployments will need to manage a data pipeline that spans multiple datacenters. Our recommended approach to this is to deploy a local Kafka cluster in each datacenter, with application instances in each datacenter interacting only with their local cluster and mirroring data between clusters (see the documentation on Geo-Replication for how to do this).This deployment pattern allows datacenters to act as independent entities and allows us to manage and tune inter-datacenter replication centrally. This allows each facility to stand alone and operate even if the inter-datacenter links are unavailable: when this occurs the mirroring falls behind until the link is restored at which time it catches up.
For applications that need a global view of all data you can use mirroring to provide clusters which have aggregate data mirrored from the local clusters in all datacenters. These aggregate clusters are used for reads by applications that require the full data set.
This is not the only possible deployment pattern. It is possible to read from or write to a remote Kafka cluster over the WAN, though obviously this will add whatever latency is required to get the cluster.
Kafka naturally batches data in both the producer and consumer so it can achieve high-throughput even over a high-latency connection. To allow this though it may be necessary to increase the TCP socket buffer sizes for the producer, consumer, and broker using the socket.send.buffer.bytes and socket.receive.buffer.bytes configurations. The appropriate way to set this is documented here.
It is generally not advisable to run a single Kafka cluster that spans multiple datacenters over a high-latency link. This will incur very high replication latency both for Kafka writes and ZooKeeper writes, and neither Kafka nor ZooKeeper will remain available in all locations if the network between locations is unavailable.
6.3 Geo-Replication (Cross-Cluster Data Mirroring)
Geo-Replication Overview
Kafka administrators can define data flows that cross the boundaries of individual Kafka clusters, data centers, or geo-regions. Such event streaming setups are often needed for organizational, technical, or legal requirements. Common scenarios include:
- Geo-replication
- Disaster recovery
- Feeding edge clusters into a central, aggregate cluster
- Physical isolation of clusters (such as production vs. testing)
- Cloud migration or hybrid cloud deployments
- Legal and compliance requirements
Administrators can set up such inter-cluster data flows with Kafka's MirrorMaker (version 2), a tool to replicate data between different Kafka environments in a streaming manner. MirrorMaker is built on top of the Kafka Connect framework and supports features such as:
- Replicates topics (data plus configurations)
- Replicates consumer groups including offsets to migrate applications between clusters
- Replicates ACLs
- Preserves partitioning
- Automatically detects new topics and partitions
- Provides a wide range of metrics, such as end-to-end replication latency across multiple data centers/clusters
- Fault-tolerant and horizontally scalable operations
Note: Geo-replication with MirrorMaker replicates data across Kafka clusters. This inter-cluster replication is different from Kafka's intra-cluster replication, which replicates data within the same Kafka cluster.
What Are Replication Flows
With MirrorMaker, Kafka administrators can replicate topics, topic configurations, consumer groups and their offsets, and ACLs from one or more source Kafka clusters to one or more target Kafka clusters, i.e., across cluster environments. In a nutshell, MirrorMaker uses Connectors to consume from source clusters and produce to target clusters.
These directional flows from source to target clusters are called replication flows. They are defined with the format {source_cluster}->{target_cluster} in the MirrorMaker configuration file as described later. Administrators can create complex replication topologies based on these flows.
Here are some example patterns:
- Active/Active high availability deployments:
A->B, B->A - Active/Passive or Active/Standby high availability deployments:
A->B - Aggregation (e.g., from many clusters to one):
A->K, B->K, C->K - Fan-out (e.g., from one to many clusters):
K->A, K->B, K->C - Forwarding:
A->B, B->C, C->D
By default, a flow replicates all topics and consumer groups (except excluded ones). However, each replication flow can be configured independently. For instance, you can define that only specific topics or consumer groups are replicated from the source cluster to the target cluster.
Here is a first example on how to configure data replication from a primary cluster to a secondary cluster (an active/passive setup):
# Basic settings
clusters = primary, secondary
primary.bootstrap.servers = broker3-primary:9092
secondary.bootstrap.servers = broker5-secondary:9092
# Define replication flows
primary->secondary.enabled = true
primary->secondary.topics = foobar-topic, quux-.*
Configuring Geo-Replication
The following sections describe how to configure and run a dedicated MirrorMaker cluster. If you want to run MirrorMaker within an existing Kafka Connect cluster or other supported deployment setups, please refer to KIP-382: MirrorMaker 2.0 and be aware that the names of configuration settings may vary between deployment modes.
Beyond what's covered in the following sections, further examples and information on configuration settings are available at:
- MirrorMakerConfig, MirrorConnectorConfig
- DefaultTopicFilter for topics, DefaultGroupFilter for consumer groups
- Example configuration settings in connect-mirror-maker.properties, KIP-382: MirrorMaker 2.0
Configuration File Syntax
The MirrorMaker configuration file is typically named connect-mirror-maker.properties. You can configure a variety of components in this file:
- MirrorMaker settings: global settings including cluster definitions (aliases), plus custom settings per replication flow
- Kafka Connect and connector settings
- Kafka producer, consumer, and admin client settings
Example: Define MirrorMaker settings (explained in more detail later).
# Global settings
clusters = us-west, us-east # defines cluster aliases
us-west.bootstrap.servers = broker3-west:9092
us-east.bootstrap.servers = broker5-east:9092
topics = .* # all topics to be replicated by default
# Specific replication flow settings (here: flow from us-west to us-east)
us-west->us-east.enabled = true
us-west->us.east.topics = foo.*, bar.* # override the default above
MirrorMaker is based on the Kafka Connect framework. Any Kafka Connect, source connector, and sink connector settings as described in the documentation chapter on Kafka Connect can be used directly in the MirrorMaker configuration, without having to change or prefix the name of the configuration setting.
Example: Define custom Kafka Connect settings to be used by MirrorMaker.
# Setting Kafka Connect defaults for MirrorMaker
tasks.max = 5
Most of the default Kafka Connect settings work well for MirrorMaker out-of-the-box, with the exception of tasks.max. In order to evenly distribute the workload across more than one MirrorMaker process, it is recommended to set tasks.max to at least 2 (preferably higher) depending on the available hardware resources and the total number of topic-partitions to be replicated.
You can further customize MirrorMaker's Kafka Connect settings per source or target cluster (more precisely, you can specify Kafka Connect worker-level configuration settings "per connector"). Use the format of {cluster}.{config_name} in the MirrorMaker configuration file.
Example: Define custom connector settings for the us-west cluster.
# us-west custom settings
us-west.offset.storage.topic = my-mirrormaker-offsets
MirrorMaker internally uses the Kafka producer, consumer, and admin clients. Custom settings for these clients are often needed. To override the defaults, use the following format in the MirrorMaker configuration file:
{source}.consumer.{consumer_config_name}{target}.producer.{producer_config_name}{source_or_target}.admin.{admin_config_name}
Example: Define custom producer, consumer, admin client settings.
# us-west cluster (from which to consume)
us-west.consumer.isolation.level = read_committed
us-west.admin.bootstrap.servers = broker57-primary:9092
# us-east cluster (to which to produce)
us-east.producer.compression.type = gzip
us-east.producer.buffer.memory = 32768
us-east.admin.bootstrap.servers = broker8-secondary:9092
Exactly once
Exactly-once semantics are supported for dedicated MirrorMaker clusters as of version 3.5.0.
For new MirrorMaker clusters, set the exactly.once.source.support property to enabled for all targeted Kafka clusters that should be written to with exactly-once semantics. For example, to enable exactly-once for writes to cluster us-east, the following configuration can be used:
us-east.exactly.once.source.support = enabled
For existing MirrorMaker clusters, a two-step upgrade is necessary. Instead of immediately setting the exactly.once.source.support property to enabled, first set it to preparing on all nodes in the cluster. Once this is complete, it can be set to enabled on all nodes in the cluster, in a second round of restarts.
In either case, it is also necessary to enable intra-cluster communication between the MirrorMaker nodes, as described in KIP-710. To do this, the dedicated.mode.enable.internal.rest property must be set to true. In addition, many of the REST-related configuration properties available for Kafka Connect can be specified the MirrorMaker config. For example, to enable intra-cluster communication in MirrorMaker cluster with each node listening on port 8080 of their local machine, the following should be added to the MirrorMaker config file:
dedicated.mode.enable.internal.rest = true
listeners = http://localhost:8080
Note that, if intra-cluster communication is enabled in production environments, it is highly recommended to secure the REST servers brought up by each MirrorMaker node. See the configuration properties for Kafka Connect for information on how this can be accomplished.
It is also recommended to filter records from aborted transactions out from replicated data when running MirrorMaker. To do this, ensure that the consumer used to read from source clusters is configured with isolation.level set to read_committed. If replicating data from cluster us-west, this can be done for all replication flows that read from that cluster by adding the following to the MirrorMaker config file:
us-west.consumer.isolation.level = read_committed
As a final note, under the hood, MirrorMaker uses Kafka Connect source connectors to replicate data. For more information on exactly-once support for these kinds of connectors, see the relevant docs page.
Creating and Enabling Replication Flows
To define a replication flow, you must first define the respective source and target Kafka clusters in the MirrorMaker configuration file.
clusters(required): comma-separated list of Kafka cluster "aliases"{clusterAlias}.bootstrap.servers(required): connection information for the specific cluster; comma-separated list of "bootstrap" Kafka brokers
Example: Define two cluster aliases primary and secondary, including their connection information.
clusters = primary, secondary
primary.bootstrap.servers = broker10-primary:9092,broker-11-primary:9092
secondary.bootstrap.servers = broker5-secondary:9092,broker6-secondary:9092
Secondly, you must explicitly enable individual replication flows with {source}->{target}.enabled = true as needed. Remember that flows are directional: if you need two-way (bidirectional) replication, you must enable flows in both directions.
# Enable replication from primary to secondary
primary->secondary.enabled = true
By default, a replication flow will replicate all but a few special topics and consumer groups from the source cluster to the target cluster, and automatically detect any newly created topics and groups. The names of replicated topics in the target cluster will be prefixed with the name of the source cluster (see section further below). For example, the topic foo in the source cluster us-west would be replicated to a topic named us-west.foo in the target cluster us-east.
The subsequent sections explain how to customize this basic setup according to your needs.
Configuring Replication Flows
The configuration of a replication flow is a combination of top-level default settings (e.g., topics), on top of which flow-specific settings, if any, are applied (e.g., us-west->us-east.topics). To change the top-level defaults, add the respective top-level setting to the MirrorMaker configuration file. To override the defaults for a specific replication flow only, use the syntax format {source}->{target}.{config.name}.
The most important settings are:
topics: list of topics or a regular expression that defines which topics in the source cluster to replicate (default:topics = .*)topics.exclude: list of topics or a regular expression to subsequently exclude topics that were matched by thetopicssetting (default:topics.exclude = .*[\-\.]internal, .*\.replica, __.*)groups: list of topics or regular expression that defines which consumer groups in the source cluster to replicate (default:groups = .*)groups.exclude: list of topics or a regular expression to subsequently exclude consumer groups that were matched by thegroupssetting (default:groups.exclude = console-consumer-.*, connect-.*, __.*){source}->{target}.enable: set totrueto enable the replication flow (default:false)
Example:
# Custom top-level defaults that apply to all replication flows
topics = .*
groups = consumer-group1, consumer-group2
# Don't forget to enable a flow!
us-west->us-east.enabled = true
# Custom settings for specific replication flows
us-west->us-east.topics = foo.*
us-west->us-east.groups = bar.*
us-west->us-east.emit.heartbeats = false
Additional configuration settings are supported which can be left with their default values in most cases. See MirrorMaker Configs.
Securing Replication Flows
MirrorMaker supports the same security settings as Kafka Connect, so please refer to the linked section for further information.
Example: Encrypt communication between MirrorMaker and the us-east cluster.
us-east.security.protocol=SSL
us-east.ssl.truststore.location=/path/to/truststore.jks
us-east.ssl.truststore.password=my-secret-password
us-east.ssl.keystore.location=/path/to/keystore.jks
us-east.ssl.keystore.password=my-secret-password
us-east.ssl.key.password=my-secret-password
Custom Naming of Replicated Topics in Target Clusters
Replicated topics in a target cluster—sometimes called remote topics—are renamed according to a replication policy. MirrorMaker uses this policy to ensure that events (aka records, messages) from different clusters are not written to the same topic-partition. By default as per DefaultReplicationPolicy, the names of replicated topics in the target clusters have the format {source}.{source_topic_name}:
us-west us-east
========= =================
bar-topic
foo-topic --> us-west.foo-topic
You can customize the separator (default: .) with the replication.policy.separator setting:
# Defining a custom separator
us-west->us-east.replication.policy.separator = _
If you need further control over how replicated topics are named, you can implement a custom ReplicationPolicy and override replication.policy.class (default is DefaultReplicationPolicy) in the MirrorMaker configuration.
Preventing Configuration Conflicts
MirrorMaker processes share configuration via their target Kafka clusters. This behavior may cause conflicts when configurations differ among MirrorMaker processes that operate against the same target cluster.
For example, the following two MirrorMaker processes would be racy:
# Configuration of process 1
A->B.enabled = true
A->B.topics = foo
# Configuration of process 2
A->B.enabled = true
A->B.topics = bar
In this case, the two processes will share configuration via cluster B, which causes a conflict. Depending on which of the two processes is the elected "leader", the result will be that either the topic foo or the topic bar is replicated, but not both.
It is therefore important to keep the MirrorMaker configuration consistent across replication flows to the same target cluster. This can be achieved, for example, through automation tooling or by using a single, shared MirrorMaker configuration file for your entire organization.
Best Practice: Consume from Remote, Produce to Local
To minimize latency ("producer lag"), it is recommended to locate MirrorMaker processes as close as possible to their target clusters, i.e., the clusters that it produces data to. That's because Kafka producers typically struggle more with unreliable or high-latency network connections than Kafka consumers.
First DC Second DC
========== =========================
primary --------- MirrorMaker --> secondary
(remote) (local)
To run such a "consume from remote, produce to local" setup, run the MirrorMaker processes close to and preferably in the same location as the target clusters, and explicitly set these "local" clusters in the --clusters command line parameter (blank-separated list of cluster aliases):
# Run in secondary's data center, reading from the remote `primary` cluster
$ ./bin/connect-mirror-maker.sh connect-mirror-maker.properties --clusters secondary
--clusters secondary tells the MirrorMaker process that the given cluster(s) are nearby, and prevents it from replicating data or sending configuration to clusters at other, remote locations.
Example: Active/Passive High Availability Deployment
The following example shows the basic settings to replicate topics from a primary to a secondary Kafka environment, but not from the secondary back to the primary. Please be aware that most production setups will need further configuration, such as security settings.
# Unidirectional flow (one-way) from primary to secondary cluster
primary.bootstrap.servers = broker1-primary:9092
secondary.bootstrap.servers = broker2-secondary:9092
primary->secondary.enabled = true
secondary->primary.enabled = false
primary->secondary.topics = foo.* # only replicate some topics
Example: Active/Active High Availability Deployment
The following example shows the basic settings to replicate topics between two clusters in both ways. Please be aware that most production setups will need further configuration, such as security settings.
# Bidirectional flow (two-way) between us-west and us-east clusters
clusters = us-west, us-east
us-west.bootstrap.servers = broker1-west:9092,broker2-west:9092
Us-east.bootstrap.servers = broker3-east:9092,broker4-east:9092
us-west->us-east.enabled = true
us-east->us-west.enabled = true
Note on preventing replication "loops" (where topics will be originally replicated from A to B, then the replicated topics will be replicated yet again from B to A, and so forth): As long as you define the above flows in the same MirrorMaker configuration file, you do not need to explicitly add topics.exclude settings to prevent replication loops between the two clusters.
Example: Multi-Cluster Geo-Replication
Let's put all the information from the previous sections together in a larger example. Imagine there are three data centers (west, east, north), with two Kafka clusters in each data center (e.g., west-1, west-2). The example in this section shows how to configure MirrorMaker (1) for Active/Active replication within each data center, as well as (2) for Cross Data Center Replication (XDCR).
First, define the source and target clusters along with their replication flows in the configuration:
# Basic settings
clusters: west-1, west-2, east-1, east-2, north-1, north-2
west-1.bootstrap.servers = ...
west-2.bootstrap.servers = ...
east-1.bootstrap.servers = ...
east-2.bootstrap.servers = ...
north-1.bootstrap.servers = ...
north-2.bootstrap.servers = ...
# Replication flows for Active/Active in West DC
west-1->west-2.enabled = true
west-2->west-1.enabled = true
# Replication flows for Active/Active in East DC
east-1->east-2.enabled = true
east-2->east-1.enabled = true
# Replication flows for Active/Active in North DC
north-1->north-2.enabled = true
north-2->north-1.enabled = true
# Replication flows for XDCR via west-1, east-1, north-1
west-1->east-1.enabled = true
west-1->north-1.enabled = true
east-1->west-1.enabled = true
east-1->north-1.enabled = true
north-1->west-1.enabled = true
north-1->east-1.enabled = true
Then, in each data center, launch one or more MirrorMaker as follows:
# In West DC:
$ ./bin/connect-mirror-maker.sh connect-mirror-maker.properties --clusters west-1 west-2
# In East DC:
$ ./bin/connect-mirror-maker.sh connect-mirror-maker.properties --clusters east-1 east-2
# In North DC:
$ ./bin/connect-mirror-maker.sh connect-mirror-maker.properties --clusters north-1 north-2
With this configuration, records produced to any cluster will be replicated within the data center, as well as across to other data centers. By providing the --clusters parameter, we ensure that each MirrorMaker process produces data to nearby clusters only.
Note: The --clusters parameter is, technically, not required here. MirrorMaker will work fine without it. However, throughput may suffer from "producer lag" between data centers, and you may incur unnecessary data transfer costs.
Starting Geo-Replication
You can run as few or as many MirrorMaker processes (think: nodes, servers) as needed. Because MirrorMaker is based on Kafka Connect, MirrorMaker processes that are configured to replicate the same Kafka clusters run in a distributed setup: They will find each other, share configuration (see section below), load balance their work, and so on. If, for example, you want to increase the throughput of replication flows, one option is to run additional MirrorMaker processes in parallel.
To start a MirrorMaker process, run the command:
$ ./bin/connect-mirror-maker.sh connect-mirror-maker.properties
After startup, it may take a few minutes until a MirrorMaker process first begins to replicate data.
Optionally, as described previously, you can set the parameter --clusters to ensure that the MirrorMaker process produces data to nearby clusters only.
# Note: The cluster alias us-west must be defined in the configuration file
$ ./bin/connect-mirror-maker.sh connect-mirror-maker.properties \
--clusters us-west
Note when testing replication of consumer groups: By default, MirrorMaker does not replicate consumer groups created by the kafka-console-consumer.sh tool, which you might use to test your MirrorMaker setup on the command line. If you do want to replicate these consumer groups as well, set the groups.exclude configuration accordingly (default: groups.exclude = console-consumer-.*, connect-.*, __.*). Remember to update the configuration again once you completed your testing.
Stopping Geo-Replication
You can stop a running MirrorMaker process by sending a SIGTERM signal with the command:
$ kill <MirrorMaker pid>
Applying Configuration Changes
To make configuration changes take effect, the MirrorMaker process(es) must be restarted.
Monitoring Geo-Replication
It is recommended to monitor MirrorMaker processes to ensure all defined replication flows are up and running correctly. MirrorMaker is built on the Connect framework and inherits all of Connect's metrics, such source-record-poll-rate. In addition, MirrorMaker produces its own metrics under the kafka.connect.mirror metric group. Metrics are tagged with the following properties:
source: alias of source cluster (e.g.,primary)target: alias of target cluster (e.g.,secondary)topic: replicated topic on target clusterpartition: partition being replicated
Metrics are tracked for each replicated topic. The source cluster can be inferred from the topic name. For example, replicating topic1 from primary->secondary will yield metrics like:
target=secondarytopic=primary.topic1partition=1
The following metrics are emitted:
# MBean: kafka.connect.mirror:type=MirrorSourceConnector,target=([-.w]+),topic=([-.w]+),partition=([0-9]+)
record-count # number of records replicated source -> target
record-age-ms # age of records when they are replicated
record-age-ms-min
record-age-ms-max
record-age-ms-avg
replication-latency-ms # time it takes records to propagate source->target
replication-latency-ms-min
replication-latency-ms-max
replication-latency-ms-avg
byte-rate # average number of bytes/sec in replicated records
# MBean: kafka.connect.mirror:type=MirrorCheckpointConnector,source=([-.w]+),target=([-.w]+)
checkpoint-latency-ms # time it takes to replicate consumer offsets
checkpoint-latency-ms-min
checkpoint-latency-ms-max
checkpoint-latency-ms-avg
These metrics do not differentiate between created-at and log-append timestamps.
6.4 Multi-Tenancy
Multi-Tenancy Overview
As a highly scalable event streaming platform, Kafka is used by many users as their central nervous system, connecting in real-time a wide range of different systems and applications from various teams and lines of businesses. Such multi-tenant cluster environments command proper control and management to ensure the peaceful coexistence of these different needs. This section highlights features and best practices to set up such shared environments, which should help you operate clusters that meet SLAs/OLAs and that minimize potential collateral damage caused by "noisy neighbors".
Multi-tenancy is a many-sided subject, including but not limited to:
- Creating user spaces for tenants (sometimes called namespaces)
- Configuring topics with data retention policies and more
- Securing topics and clusters with encryption, authentication, and authorization
- Isolating tenants with quotas and rate limits
- Monitoring and metering
- Inter-cluster data sharing (cf. geo-replication)
Creating User Spaces (Namespaces) For Tenants With Topic Naming
Kafka administrators operating a multi-tenant cluster typically need to define user spaces for each tenant. For the purpose of this section, "user spaces" are a collection of topics, which are grouped together under the management of a single entity or user.
In Kafka, the main unit of data is the topic. Users can create and name each topic. They can also delete them, but it is not possible to rename a topic directly. Instead, to rename a topic, the user must create a new topic, move the messages from the original topic to the new, and then delete the original. With this in mind, it is recommended to define logical spaces, based on an hierarchical topic naming structure. This setup can then be combined with security features, such as prefixed ACLs, to isolate different spaces and tenants, while also minimizing the administrative overhead for securing the data in the cluster.
These logical user spaces can be grouped in different ways, and the concrete choice depends on how your organization prefers to use your Kafka clusters. The most common groupings are as follows.
By team or organizational unit: Here, the team is the main aggregator. In an organization where teams are the main user of the Kafka infrastructure, this might be the best grouping.
Example topic naming structure:
<organization>.<team>.<dataset>.<event-name>
(e.g., "acme.infosec.telemetry.logins")
By project or product: Here, a team manages more than one project. Their credentials will be different for each project, so all the controls and settings will always be project related.
Example topic naming structure:
<project>.<product>.<event-name>
(e.g., "mobility.payments.suspicious")
Certain information should normally not be put in a topic name, such as information that is likely to change over time (e.g., the name of the intended consumer) or that is a technical detail or metadata that is available elsewhere (e.g., the topic's partition count and other configuration settings).
To enforce a topic naming structure, several options are available:
- Use prefix ACLs (cf. KIP-290) to enforce a common prefix for topic names. For example, team A may only be permitted to create topics whose names start with
payments.teamA.. - Define a custom
CreateTopicPolicy(cf. KIP-108 and the setting create.topic.policy.class.name) to enforce strict naming patterns. These policies provide the most flexibility and can cover complex patterns and rules to match an organization's needs. - Disable topic creation for normal users by denying it with an ACL, and then rely on an external process to create topics on behalf of users (e.g., scripting or your favorite automation toolkit).
- It may also be useful to disable the Kafka feature to auto-create topics on demand by setting
auto.create.topics.enable=falsein the broker configuration. Note that you should not rely solely on this option.
Configuring Topics: Data Retention And More
Kafka's configuration is very flexible due to its fine granularity, and it supports a plethora of per-topic configuration settings to help administrators set up multi-tenant clusters. For example, administrators often need to define data retention policies to control how much and/or for how long data will be stored in a topic, with settings such as retention.bytes (size) and retention.ms (time). This limits storage consumption within the cluster, and helps complying with legal requirements such as GDPR.
Securing Clusters and Topics: Authentication, Authorization, Encryption
Because the documentation has a dedicated chapter on security that applies to any Kafka deployment, this section focuses on additional considerations for multi-tenant environments.
Security settings for Kafka fall into three main categories, which are similar to how administrators would secure other client-server data systems, like relational databases and traditional messaging systems.
- Encryption of data transferred between Kafka brokers and Kafka clients, between brokers, between brokers and ZooKeeper nodes, and between brokers and other, optional tools.
- Authentication of connections from Kafka clients and applications to Kafka brokers, as well as connections from Kafka brokers to ZooKeeper nodes.
- Authorization of client operations such as creating, deleting, and altering the configuration of topics; writing events to or reading events from a topic; creating and deleting ACLs. Administrators can also define custom policies to put in place additional restrictions, such as a
CreateTopicPolicyandAlterConfigPolicy(see KIP-108 and the settings create.topic.policy.class.name, alter.config.policy.class.name).
When securing a multi-tenant Kafka environment, the most common administrative task is the third category (authorization), i.e., managing the user/client permissions that grant or deny access to certain topics and thus to the data stored by users within a cluster. This task is performed predominantly through the setting of access control lists (ACLs). Here, administrators of multi-tenant environments in particular benefit from putting a hierarchical topic naming structure in place as described in a previous section, because they can conveniently control access to topics through prefixed ACLs (--resource-pattern-type Prefixed). This significantly minimizes the administrative overhead of securing topics in multi-tenant environments: administrators can make their own trade-offs between higher developer convenience (more lenient permissions, using fewer and broader ACLs) vs. tighter security (more stringent permissions, using more and narrower ACLs).
In the following example, user Alice—a new member of ACME corporation's InfoSec team—is granted write permissions to all topics whose names start with "acme.infosec.", such as "acme.infosec.telemetry.logins" and "acme.infosec.syslogs.events".
# Grant permissions to user Alice
$ bin/kafka-acls.sh \
--bootstrap-server broker1:9092 \
--add --allow-principal User:Alice \
--producer \
--resource-pattern-type prefixed --topic acme.infosec.
You can similarly use this approach to isolate different customers on the same shared cluster.
Isolating Tenants: Quotas, Rate Limiting, Throttling
Multi-tenant clusters should generally be configured with quotas, which protect against users (tenants) eating up too many cluster resources, such as when they attempt to write or read very high volumes of data, or create requests to brokers at an excessively high rate. This may cause network saturation, monopolize broker resources, and impact other clients—all of which you want to avoid in a shared environment.
Client quotas: Kafka supports different types of (per-user principal) client quotas. Because a client's quotas apply irrespective of which topics the client is writing to or reading from, they are a convenient and effective tool to allocate resources in a multi-tenant cluster. Request rate quotas, for example, help to limit a user's impact on broker CPU usage by limiting the time a broker spends on the request handling path for that user, after which throttling kicks in. In many situations, isolating users with request rate quotas has a bigger impact in multi-tenant clusters than setting incoming/outgoing network bandwidth quotas, because excessive broker CPU usage for processing requests reduces the effective bandwidth the broker can serve. Furthermore, administrators can also define quotas on topic operations—such as create, delete, and alter—to prevent Kafka clusters from being overwhelmed by highly concurrent topic operations (see KIP-599 and the quota type controller_mutation_rate).
Server quotas: Kafka also supports different types of broker-side quotas. For example, administrators can set a limit on the rate with which the broker accepts new connections, set the maximum number of connections per broker, or set the maximum number of connections allowed from a specific IP address.
For more information, please refer to the quota overview and how to set quotas.
Monitoring and Metering
Monitoring is a broader subject that is covered elsewhere in the documentation. Administrators of any Kafka environment, but especially multi-tenant ones, should set up monitoring according to these instructions. Kafka supports a wide range of metrics, such as the rate of failed authentication attempts, request latency, consumer lag, total number of consumer groups, metrics on the quotas described in the previous section, and many more.
For example, monitoring can be configured to track the size of topic-partitions (with the JMX metric kafka.log.Log.Size.<TOPIC-NAME>), and thus the total size of data stored in a topic. You can then define alerts when tenants on shared clusters are getting close to using too much storage space.
Multi-Tenancy and Geo-Replication
Kafka lets you share data across different clusters, which may be located in different geographical regions, data centers, and so on. Apart from use cases such as disaster recovery, this functionality is useful when a multi-tenant setup requires inter-cluster data sharing. See the section Geo-Replication (Cross-Cluster Data Mirroring) for more information.
Further considerations
Data contracts: You may need to define data contracts between the producers and the consumers of data in a cluster, using event schemas. This ensures that events written to Kafka can always be read properly again, and prevents malformed or corrupt events being written. The best way to achieve this is to deploy a so-called schema registry alongside the cluster. (Kafka does not include a schema registry, but there are third-party implementations available.) A schema registry manages the event schemas and maps the schemas to topics, so that producers know which topics are accepting which types (schemas) of events, and consumers know how to read and parse events in a topic. Some registry implementations provide further functionality, such as schema evolution, storing a history of all schemas, and schema compatibility settings.
6.5 Kafka Configuration
Important Client Configurations
The most important producer configurations are:- acks
- compression
- batch size
All configurations are documented in the configuration section.
A Production Server Config
Here is an example production server configuration: # ZooKeeper
zookeeper.connect=[list of ZooKeeper servers]
# Log configuration
num.partitions=8
default.replication.factor=3
log.dir=[List of directories. Kafka should have its own dedicated disk(s) or SSD(s).]
# Other configurations
broker.id=[An integer. Start with 0 and increment by 1 for each new broker.]
listeners=[list of listeners]
auto.create.topics.enable=false
min.insync.replicas=2
queued.max.requests=[number of concurrent requests]6.6 Java Version
Java 8, Java 11, and Java 17 are supported. Note that Java 8 support has been deprecated since Apache Kafka 3.0 and will be removed in Apache Kafka 4.0. Java 11 and later versions perform significantly better if TLS is enabled, so they are highly recommended (they also include a number of other performance improvements: G1GC, CRC32C, Compact Strings, Thread-Local Handshakes and more). From a security perspective, we recommend the latest released patch version as older freely available versions have disclosed security vulnerabilities. Typical arguments for running Kafka with OpenJDK-based Java implementations (including Oracle JDK) are: -Xmx6g -Xms6g -XX:MetaspaceSize=96m -XX:+UseG1GC
-XX:MaxGCPauseMillis=20 -XX:InitiatingHeapOccupancyPercent=35 -XX:G1HeapRegionSize=16M
-XX:MinMetaspaceFreeRatio=50 -XX:MaxMetaspaceFreeRatio=80 -XX:+ExplicitGCInvokesConcurrent- 60 brokers
- 50k partitions (replication factor 2)
- 800k messages/sec in
- 300 MB/sec inbound, 1 GB/sec+ outbound
6.7 Hardware and OS
We are using dual quad-core Intel Xeon machines with 24GB of memory.You need sufficient memory to buffer active readers and writers. You can do a back-of-the-envelope estimate of memory needs by assuming you want to be able to buffer for 30 seconds and compute your memory need as write_throughput*30.
The disk throughput is important. We have 8x7200 rpm SATA drives. In general disk throughput is the performance bottleneck, and more disks is better. Depending on how you configure flush behavior you may or may not benefit from more expensive disks (if you force flush often then higher RPM SAS drives may be better).
OS
Kafka should run well on any unix system and has been tested on Linux and Solaris.We have seen a few issues running on Windows and Windows is not currently a well supported platform though we would be happy to change that.
It is unlikely to require much OS-level tuning, but there are three potentially important OS-level configurations:
- File descriptor limits: Kafka uses file descriptors for log segments and open connections. If a broker hosts many partitions, consider that the broker needs at least (number_of_partitions)*(partition_size/segment_size) to track all log segments in addition to the number of connections the broker makes. We recommend at least 100000 allowed file descriptors for the broker processes as a starting point. Note: The mmap() function adds an extra reference to the file associated with the file descriptor fildes which is not removed by a subsequent close() on that file descriptor. This reference is removed when there are no more mappings to the file.
- Max socket buffer size: can be increased to enable high-performance data transfer between data centers as described here.
- Maximum number of memory map areas a process may have (aka vm.max_map_count). See the Linux kernel documentation. You should keep an eye at this OS-level property when considering the maximum number of partitions a broker may have. By default, on a number of Linux systems, the value of vm.max_map_count is somewhere around 65535. Each log segment, allocated per partition, requires a pair of index/timeindex files, and each of these files consumes 1 map area. In other words, each log segment uses 2 map areas. Thus, each partition requires minimum 2 map areas, as long as it hosts a single log segment. That is to say, creating 50000 partitions on a broker will result allocation of 100000 map areas and likely cause broker crash with OutOfMemoryError (Map failed) on a system with default vm.max_map_count. Keep in mind that the number of log segments per partition varies depending on the segment size, load intensity, retention policy and, generally, tends to be more than one.
Disks and Filesystem
We recommend using multiple drives to get good throughput and not sharing the same drives used for Kafka data with application logs or other OS filesystem activity to ensure good latency. You can either RAID these drives together into a single volume or format and mount each drive as its own directory. Since Kafka has replication the redundancy provided by RAID can also be provided at the application level. This choice has several tradeoffs.If you configure multiple data directories partitions will be assigned round-robin to data directories. Each partition will be entirely in one of the data directories. If data is not well balanced among partitions this can lead to load imbalance between disks.
RAID can potentially do better at balancing load between disks (although it doesn't always seem to) because it balances load at a lower level. The primary downside of RAID is that it is usually a big performance hit for write throughput and reduces the available disk space.
Another potential benefit of RAID is the ability to tolerate disk failures. However our experience has been that rebuilding the RAID array is so I/O intensive that it effectively disables the server, so this does not provide much real availability improvement.
Application vs. OS Flush Management
Kafka always immediately writes all data to the filesystem and supports the ability to configure the flush policy that controls when data is forced out of the OS cache and onto disk using the flush. This flush policy can be controlled to force data to disk after a period of time or after a certain number of messages has been written. There are several choices in this configuration.Kafka must eventually call fsync to know that data was flushed. When recovering from a crash for any log segment not known to be fsync'd Kafka will check the integrity of each message by checking its CRC and also rebuild the accompanying offset index file as part of the recovery process executed on startup.
Note that durability in Kafka does not require syncing data to disk, as a failed node will always recover from its replicas.
We recommend using the default flush settings which disable application fsync entirely. This means relying on the background flush done by the OS and Kafka's own background flush. This provides the best of all worlds for most uses: no knobs to tune, great throughput and latency, and full recovery guarantees. We generally feel that the guarantees provided by replication are stronger than sync to local disk, however the paranoid still may prefer having both and application level fsync policies are still supported.
The drawback of using application level flush settings is that it is less efficient in its disk usage pattern (it gives the OS less leeway to re-order writes) and it can introduce latency as fsync in most Linux filesystems blocks writes to the file whereas the background flushing does much more granular page-level locking.
In general you don't need to do any low-level tuning of the filesystem, but in the next few sections we will go over some of this in case it is useful.
Understanding Linux OS Flush Behavior
In Linux, data written to the filesystem is maintained in pagecache until it must be written out to disk (due to an application-level fsync or the OS's own flush policy). The flushing of data is done by a set of background threads called pdflush (or in post 2.6.32 kernels "flusher threads").Pdflush has a configurable policy that controls how much dirty data can be maintained in cache and for how long before it must be written back to disk. This policy is described here. When Pdflush cannot keep up with the rate of data being written it will eventually cause the writing process to block incurring latency in the writes to slow down the accumulation of data.
You can see the current state of OS memory usage by doing
> cat /proc/meminfoThe meaning of these values are described in the link above.
Using pagecache has several advantages over an in-process cache for storing data that will be written out to disk:
- The I/O scheduler will batch together consecutive small writes into bigger physical writes which improves throughput.
- The I/O scheduler will attempt to re-sequence writes to minimize movement of the disk head which improves throughput.
- It automatically uses all the free memory on the machine
Filesystem Selection
Kafka uses regular files on disk, and as such it has no hard dependency on a specific filesystem. The two filesystems which have the most usage, however, are EXT4 and XFS. Historically, EXT4 has had more usage, but recent improvements to the XFS filesystem have shown it to have better performance characteristics for Kafka's workload with no compromise in stability.
Comparison testing was performed on a cluster with significant message loads, using a variety of filesystem creation and mount options. The primary metric in Kafka that was monitored was the "Request Local Time", indicating the amount of time append operations were taking. XFS resulted in much better local times (160ms vs. 250ms+ for the best EXT4 configuration), as well as lower average wait times. The XFS performance also showed less variability in disk performance.
General Filesystem Notes
For any filesystem used for data directories, on Linux systems, the following options are recommended to be used at mount time:- noatime: This option disables updating of a file's atime (last access time) attribute when the file is read. This can eliminate a significant number of filesystem writes, especially in the case of bootstrapping consumers. Kafka does not rely on the atime attributes at all, so it is safe to disable this.
XFS Notes
The XFS filesystem has a significant amount of auto-tuning in place, so it does not require any change in the default settings, either at filesystem creation time or at mount. The only tuning parameters worth considering are:- largeio: This affects the preferred I/O size reported by the stat call. While this can allow for higher performance on larger disk writes, in practice it had minimal or no effect on performance.
- nobarrier: For underlying devices that have battery-backed cache, this option can provide a little more performance by disabling periodic write flushes. However, if the underlying device is well-behaved, it will report to the filesystem that it does not require flushes, and this option will have no effect.
EXT4 Notes
EXT4 is a serviceable choice of filesystem for the Kafka data directories, however getting the most performance out of it will require adjusting several mount options. In addition, these options are generally unsafe in a failure scenario, and will result in much more data loss and corruption. For a single broker failure, this is not much of a concern as the disk can be wiped and the replicas rebuilt from the cluster. In a multiple-failure scenario, such as a power outage, this can mean underlying filesystem (and therefore data) corruption that is not easily recoverable. The following options can be adjusted:- data=writeback: Ext4 defaults to data=ordered which puts a strong order on some writes. Kafka does not require this ordering as it does very paranoid data recovery on all unflushed log. This setting removes the ordering constraint and seems to significantly reduce latency.
- Disabling journaling: Journaling is a tradeoff: it makes reboots faster after server crashes but it introduces a great deal of additional locking which adds variance to write performance. Those who don't care about reboot time and want to reduce a major source of write latency spikes can turn off journaling entirely.
- commit=num_secs: This tunes the frequency with which ext4 commits to its metadata journal. Setting this to a lower value reduces the loss of unflushed data during a crash. Setting this to a higher value will improve throughput.
- nobh: This setting controls additional ordering guarantees when using data=writeback mode. This should be safe with Kafka as we do not depend on write ordering and improves throughput and latency.
- delalloc: Delayed allocation means that the filesystem avoid allocating any blocks until the physical write occurs. This allows ext4 to allocate a large extent instead of smaller pages and helps ensure the data is written sequentially. This feature is great for throughput. It does seem to involve some locking in the filesystem which adds a bit of latency variance.
Replace KRaft Controller Disk
When Kafka is configured to use KRaft, the controllers store the cluster metadata in the directory specified in metadata.log.dir -- or the first log directory, if metadata.log.dir is not configured. See the documentation for metadata.log.dir for details.
If the data in the cluster metadata directory is lost either because of hardware failure or the hardware needs to be replaced, care should be taken when provisioning the new controller node. The new controller node should not be formatted and started until the majority of the controllers have all of the committed data. To determine if the majority of the controllers have the committed data, run the kafka-metadata-quorum.sh tool to describe the replication status:
> bin/kafka-metadata-quorum.sh --bootstrap-server broker_host:port describe --replication
NodeId LogEndOffset Lag LastFetchTimestamp LastCaughtUpTimestamp Status
1 25806 0 1662500992757 1662500992757 Leader
... ... ... ... ... ...
Check and wait until the Lag is small for a majority of the controllers. If the leader's end offset is not increasing, you can wait until the lag is 0 for a majority; otherwise, you can pick the latest leader end offset and wait until all replicas have reached it. Check and wait until the LastFetchTimestamp and LastCaughtUpTimestamp are close to each other for the majority of the controllers. At this point it is safer to format the controller's metadata log directory. This can be done by running the kafka-storage.sh command.
> bin/kafka-storage.sh format --cluster-id uuid --config server_propertiesIt is possible for the bin/kafka-storage.sh format command above to fail with a message like Log directory ... is already formatted. This can happen when combined mode is used and only the metadata log directory was lost but not the others. In that case and only in that case, can you run the kafka-storage.sh format command with the --ignore-formatted option.
Start the KRaft controller after formatting the log directories.
> /bin/kafka-server-start.sh server_properties6.8 Monitoring
Kafka uses Yammer Metrics for metrics reporting in the server. The Java clients use Kafka Metrics, a built-in metrics registry that minimizes transitive dependencies pulled into client applications. Both expose metrics via JMX and can be configured to report stats using pluggable stats reporters to hook up to your monitoring system.
All Kafka rate metrics have a corresponding cumulative count metric with suffix -total. For example,
records-consumed-rate has a corresponding metric named records-consumed-total.
The easiest way to see the available metrics is to fire up jconsole and point it at a running kafka client or server; this will allow browsing all metrics with JMX.
Security Considerations for Remote Monitoring using JMX
Apache Kafka disables remote JMX by default. You can enable remote monitoring using JMX by setting the environment variableJMX_PORT for processes started using the CLI or standard Java system properties to enable remote JMX programmatically.
You must enable security when enabling remote JMX in production scenarios to ensure that unauthorized users cannot monitor or
control your broker or application as well as the platform on which these are running. Note that authentication is disabled for
JMX by default in Kafka and security configs must be overridden for production deployments by setting the environment variable
KAFKA_JMX_OPTS for processes started using the CLI or by setting appropriate Java system properties. See
Monitoring and Management Using JMX Technology
for details on securing JMX.
We do graphing and alerting on the following metrics:
| Description | Mbean name | Normal value |
|---|---|---|
| Message in rate | kafka.server:type=BrokerTopicMetrics,name=MessagesInPerSec,topic=([-.\w]+) | Incoming message rate per topic. Omitting 'topic=(...)' will yield the all-topic rate. |
| Byte in rate from clients | kafka.server:type=BrokerTopicMetrics,name=BytesInPerSec,topic=([-.\w]+) | Byte in (from the clients) rate per topic. Omitting 'topic=(...)' will yield the all-topic rate. |
| Byte in rate from other brokers | kafka.server:type=BrokerTopicMetrics,name=ReplicationBytesInPerSec | Byte in (from the other brokers) rate across all topics. |
| Controller Request rate from Broker | kafka.controller:type=ControllerChannelManager,name=RequestRateAndQueueTimeMs,brokerId=([0-9]+) | The rate (requests per second) at which the ControllerChannelManager takes requests from the queue of the given broker. And the time it takes for a request to stay in this queue before it is taken from the queue. |
| Controller Event queue size | kafka.controller:type=ControllerEventManager,name=EventQueueSize | Size of the ControllerEventManager's queue. |
| Controller Event queue time | kafka.controller:type=ControllerEventManager,name=EventQueueTimeMs | Time that takes for any event (except the Idle event) to wait in the ControllerEventManager's queue before being processed |
| Request rate | kafka.network:type=RequestMetrics,name=RequestsPerSec,request={Produce|FetchConsumer|FetchFollower},version=([0-9]+) | |
| Error rate | kafka.network:type=RequestMetrics,name=ErrorsPerSec,request=([-.\w]+),error=([-.\w]+) | Number of errors in responses counted per-request-type, per-error-code. If a response contains multiple errors, all are counted. error=NONE indicates successful responses. |
| Produce request rate | kafka.server:type=BrokerTopicMetrics,name=TotalProduceRequestsPerSec,topic=([-.\w]+) | Produce request rate per topic. Omitting 'topic=(...)' will yield the all-topic rate. |
| Fetch request rate | kafka.server:type=BrokerTopicMetrics,name=TotalFetchRequestsPerSec,topic=([-.\w]+) | Fetch request (from clients or followers) rate per topic. Omitting 'topic=(...)' will yield the all-topic rate. |
| Failed produce request rate | kafka.server:type=BrokerTopicMetrics,name=FailedProduceRequestsPerSec,topic=([-.\w]+) | Failed Produce request rate per topic. Omitting 'topic=(...)' will yield the all-topic rate. |
| Failed fetch request rate | kafka.server:type=BrokerTopicMetrics,name=FailedFetchRequestsPerSec,topic=([-.\w]+) | Failed Fetch request (from clients or followers) rate per topic. Omitting 'topic=(...)' will yield the all-topic rate. |
| Request size in bytes | kafka.network:type=RequestMetrics,name=RequestBytes,request=([-.\w]+) | Size of requests for each request type. |
| Temporary memory size in bytes | kafka.network:type=RequestMetrics,name=TemporaryMemoryBytes,request={Produce|Fetch} | Temporary memory used for message format conversions and decompression. |
| Message conversion time | kafka.network:type=RequestMetrics,name=MessageConversionsTimeMs,request={Produce|Fetch} | Time in milliseconds spent on message format conversions. |
| Message conversion rate | kafka.server:type=BrokerTopicMetrics,name={Produce|Fetch}MessageConversionsPerSec,topic=([-.\w]+) | Message format conversion rate, for Produce or Fetch requests, per topic. Omitting 'topic=(...)' will yield the all-topic rate. |
| Request Queue Size | kafka.network:type=RequestChannel,name=RequestQueueSize | Size of the request queue. |
| Byte out rate to clients | kafka.server:type=BrokerTopicMetrics,name=BytesOutPerSec,topic=([-.\w]+) | Byte out (to the clients) rate per topic. Omitting 'topic=(...)' will yield the all-topic rate. |
| Byte out rate to other brokers | kafka.server:type=BrokerTopicMetrics,name=ReplicationBytesOutPerSec | Byte out (to the other brokers) rate across all topics |
| Rejected byte rate | kafka.server:type=BrokerTopicMetrics,name=BytesRejectedPerSec,topic=([-.\w]+) | Rejected byte rate per topic, due to the record batch size being greater than max.message.bytes configuration. Omitting 'topic=(...)' will yield the all-topic rate. |
| Message validation failure rate due to no key specified for compacted topic | kafka.server:type=BrokerTopicMetrics,name=NoKeyCompactedTopicRecordsPerSec | 0 |
| Message validation failure rate due to invalid magic number | kafka.server:type=BrokerTopicMetrics,name=InvalidMagicNumberRecordsPerSec | 0 |
| Message validation failure rate due to incorrect crc checksum | kafka.server:type=BrokerTopicMetrics,name=InvalidMessageCrcRecordsPerSec | 0 |
| Message validation failure rate due to non-continuous offset or sequence number in batch | kafka.server:type=BrokerTopicMetrics,name=InvalidOffsetOrSequenceRecordsPerSec | 0 |
| Log flush rate and time | kafka.log:type=LogFlushStats,name=LogFlushRateAndTimeMs | |
| # of offline log directories | kafka.log:type=LogManager,name=OfflineLogDirectoryCount | 0 |
| Leader election rate | kafka.controller:type=ControllerStats,name=LeaderElectionRateAndTimeMs | non-zero when there are broker failures |
| Unclean leader election rate | kafka.controller:type=ControllerStats,name=UncleanLeaderElectionsPerSec | 0 |
| Is controller active on broker | kafka.controller:type=KafkaController,name=ActiveControllerCount | only one broker in the cluster should have 1 |
| Pending topic deletes | kafka.controller:type=KafkaController,name=TopicsToDeleteCount | |
| Pending replica deletes | kafka.controller:type=KafkaController,name=ReplicasToDeleteCount | |
| Ineligible pending topic deletes | kafka.controller:type=KafkaController,name=TopicsIneligibleToDeleteCount | |
| Ineligible pending replica deletes | kafka.controller:type=KafkaController,name=ReplicasIneligibleToDeleteCount | |
| # of under replicated partitions (|ISR| < |all replicas|) | kafka.server:type=ReplicaManager,name=UnderReplicatedPartitions | 0 |
| # of under minIsr partitions (|ISR| < min.insync.replicas) | kafka.server:type=ReplicaManager,name=UnderMinIsrPartitionCount | 0 |
| # of at minIsr partitions (|ISR| = min.insync.replicas) | kafka.server:type=ReplicaManager,name=AtMinIsrPartitionCount | 0 |
| Producer Id counts | kafka.server:type=ReplicaManager,name=ProducerIdCount | Count of all producer ids created by transactional and idempotent producers in each replica on the broker |
| Partition counts | kafka.server:type=ReplicaManager,name=PartitionCount | mostly even across brokers |
| Offline Replica counts | kafka.server:type=ReplicaManager,name=OfflineReplicaCount | 0 |
| Leader replica counts | kafka.server:type=ReplicaManager,name=LeaderCount | mostly even across brokers |
| ISR shrink rate | kafka.server:type=ReplicaManager,name=IsrShrinksPerSec | If a broker goes down, ISR for some of the partitions will shrink. When that broker is up again, ISR will be expanded once the replicas are fully caught up. Other than that, the expected value for both ISR shrink rate and expansion rate is 0. |
| ISR expansion rate | kafka.server:type=ReplicaManager,name=IsrExpandsPerSec | See above |
| Failed ISR update rate | kafka.server:type=ReplicaManager,name=FailedIsrUpdatesPerSec | 0 |
| Max lag in messages btw follower and leader replicas | kafka.server:type=ReplicaFetcherManager,name=MaxLag,clientId=Replica | lag should be proportional to the maximum batch size of a produce request. |
| Lag in messages per follower replica | kafka.server:type=FetcherLagMetrics,name=ConsumerLag,clientId=([-.\w]+),topic=([-.\w]+),partition=([0-9]+) | lag should be proportional to the maximum batch size of a produce request. |
| Requests waiting in the producer purgatory | kafka.server:type=DelayedOperationPurgatory,name=PurgatorySize,delayedOperation=Produce | non-zero if ack=-1 is used |
| Requests waiting in the fetch purgatory | kafka.server:type=DelayedOperationPurgatory,name=PurgatorySize,delayedOperation=Fetch | size depends on fetch.wait.max.ms in the consumer |
| Request total time | kafka.network:type=RequestMetrics,name=TotalTimeMs,request={Produce|FetchConsumer|FetchFollower} | broken into queue, local, remote and response send time |
| Time the request waits in the request queue | kafka.network:type=RequestMetrics,name=RequestQueueTimeMs,request={Produce|FetchConsumer|FetchFollower} | |
| Time the request is processed at the leader | kafka.network:type=RequestMetrics,name=LocalTimeMs,request={Produce|FetchConsumer|FetchFollower} | |
| Time the request waits for the follower | kafka.network:type=RequestMetrics,name=RemoteTimeMs,request={Produce|FetchConsumer|FetchFollower} | non-zero for produce requests when ack=-1 |
| Time the request waits in the response queue | kafka.network:type=RequestMetrics,name=ResponseQueueTimeMs,request={Produce|FetchConsumer|FetchFollower} | |
| Time to send the response | kafka.network:type=RequestMetrics,name=ResponseSendTimeMs,request={Produce|FetchConsumer|FetchFollower} | |
| Number of messages the consumer lags behind the producer by. Published by the consumer, not broker. | kafka.consumer:type=consumer-fetch-manager-metrics,client-id={client-id} Attribute: records-lag-max | |
| The average fraction of time the network processors are idle | kafka.network:type=SocketServer,name=NetworkProcessorAvgIdlePercent | between 0 and 1, ideally > 0.3 |
| The number of connections disconnected on a processor due to a client not re-authenticating and then using the connection beyond its expiration time for anything other than re-authentication | kafka.server:type=socket-server-metrics,listener=[SASL_PLAINTEXT|SASL_SSL],networkProcessor=<#>,name=expired-connections-killed-count | ideally 0 when re-authentication is enabled, implying there are no longer any older, pre-2.2.0 clients connecting to this (listener, processor) combination |
| The total number of connections disconnected, across all processors, due to a client not re-authenticating and then using the connection beyond its expiration time for anything other than re-authentication | kafka.network:type=SocketServer,name=ExpiredConnectionsKilledCount | ideally 0 when re-authentication is enabled, implying there are no longer any older, pre-2.2.0 clients connecting to this broker |
| The average fraction of time the request handler threads are idle | kafka.server:type=KafkaRequestHandlerPool,name=RequestHandlerAvgIdlePercent | between 0 and 1, ideally > 0.3 |
| Bandwidth quota metrics per (user, client-id), user or client-id | kafka.server:type={Produce|Fetch},user=([-.\w]+),client-id=([-.\w]+) | Two attributes. throttle-time indicates the amount of time in ms the client was throttled. Ideally = 0. byte-rate indicates the data produce/consume rate of the client in bytes/sec. For (user, client-id) quotas, both user and client-id are specified. If per-client-id quota is applied to the client, user is not specified. If per-user quota is applied, client-id is not specified. |
| Request quota metrics per (user, client-id), user or client-id | kafka.server:type=Request,user=([-.\w]+),client-id=([-.\w]+) | Two attributes. throttle-time indicates the amount of time in ms the client was throttled. Ideally = 0. request-time indicates the percentage of time spent in broker network and I/O threads to process requests from client group. For (user, client-id) quotas, both user and client-id are specified. If per-client-id quota is applied to the client, user is not specified. If per-user quota is applied, client-id is not specified. |
| Requests exempt from throttling | kafka.server:type=Request | exempt-throttle-time indicates the percentage of time spent in broker network and I/O threads to process requests that are exempt from throttling. |
| ZooKeeper client request latency | kafka.server:type=ZooKeeperClientMetrics,name=ZooKeeperRequestLatencyMs | Latency in milliseconds for ZooKeeper requests from broker. |
| ZooKeeper connection status | kafka.server:type=SessionExpireListener,name=SessionState | Connection status of broker's ZooKeeper session which may be one of Disconnected|SyncConnected|AuthFailed|ConnectedReadOnly|SaslAuthenticated|Expired. |
| Max time to load group metadata | kafka.server:type=group-coordinator-metrics,name=partition-load-time-max | maximum time, in milliseconds, it took to load offsets and group metadata from the consumer offset partitions loaded in the last 30 seconds (including time spent waiting for the loading task to be scheduled) |
| Avg time to load group metadata | kafka.server:type=group-coordinator-metrics,name=partition-load-time-avg | average time, in milliseconds, it took to load offsets and group metadata from the consumer offset partitions loaded in the last 30 seconds (including time spent waiting for the loading task to be scheduled) |
| Max time to load transaction metadata | kafka.server:type=transaction-coordinator-metrics,name=partition-load-time-max | maximum time, in milliseconds, it took to load transaction metadata from the consumer offset partitions loaded in the last 30 seconds (including time spent waiting for the loading task to be scheduled) |
| Avg time to load transaction metadata | kafka.server:type=transaction-coordinator-metrics,name=partition-load-time-avg | average time, in milliseconds, it took to load transaction metadata from the consumer offset partitions loaded in the last 30 seconds (including time spent waiting for the loading task to be scheduled) |
| Rate of transactional verification errors | kafka.server:type=AddPartitionsToTxnManager,name=VerificationFailureRate | Rate of verifications that returned in failure either from the AddPartitionsToTxn API response or through errors in the AddPartitionsToTxnManager. In steady state 0, but transient errors are expected during rolls and reassignments of the transactional state partition. |
| Time to verify a transactional request | kafka.server:type=AddPartitionsToTxnManager,name=VerificationTimeMs | The amount of time queueing while a possible previous request is in-flight plus the round trip to the transaction coordinator to verify (or not verify) |
| Consumer Group Offset Count | kafka.server:type=GroupMetadataManager,name=NumOffsets | Total number of committed offsets for Consumer Groups |
| Consumer Group Count | kafka.server:type=GroupMetadataManager,name=NumGroups | Total number of Consumer Groups |
| Consumer Group Count, per State | kafka.server:type=GroupMetadataManager,name=NumGroups[PreparingRebalance,CompletingRebalance,Empty,Stable,Dead] | The number of Consumer Groups in each state: PreparingRebalance, CompletingRebalance, Empty, Stable, Dead |
| Number of reassigning partitions | kafka.server:type=ReplicaManager,name=ReassigningPartitions | The number of reassigning leader partitions on a broker. |
| Outgoing byte rate of reassignment traffic | kafka.server:type=BrokerTopicMetrics,name=ReassignmentBytesOutPerSec | 0; non-zero when a partition reassignment is in progress. |
| Incoming byte rate of reassignment traffic | kafka.server:type=BrokerTopicMetrics,name=ReassignmentBytesInPerSec | 0; non-zero when a partition reassignment is in progress. |
| Size of a partition on disk (in bytes) | kafka.log:type=Log,name=Size,topic=([-.\w]+),partition=([0-9]+) | The size of a partition on disk, measured in bytes. |
| Number of log segments in a partition | kafka.log:type=Log,name=NumLogSegments,topic=([-.\w]+),partition=([0-9]+) | The number of log segments in a partition. |
| First offset in a partition | kafka.log:type=Log,name=LogStartOffset,topic=([-.\w]+),partition=([0-9]+) | The first offset in a partition. |
| Last offset in a partition | kafka.log:type=Log,name=LogEndOffset,topic=([-.\w]+),partition=([0-9]+) | The last offset in a partition. |
Tiered Storage Monitoring
The following set of metrics are available for monitoring of the tiered storage feature:| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| Remote Fetch Bytes Per Sec | Rate of bytes read from remote storage per topic. Omitting 'topic=(...)' will yield the all-topic rate | kafka.server:type=BrokerTopicMetrics,name=RemoteFetchBytesPerSec,topic=([-.\w]+) |
| Remote Fetch Requests Per Sec | Rate of read requests from remote storage per topic. Omitting 'topic=(...)' will yield the all-topic rate | kafka.server:type=BrokerTopicMetrics,name=RemoteFetchRequestsPerSec,topic=([-.\w]+) |
| Remote Fetch Errors Per Sec | Rate of read errors from remote storage per topic. Omitting 'topic=(...)' will yield the all-topic rate | kafka.server:type=BrokerTopicMetrics,name=RemoteFetchErrorsPerSec,topic=([-.\w]+) |
| Remote Copy Bytes Per Sec | Rate of bytes copied to remote storage per topic. Omitting 'topic=(...)' will yield the all-topic rate | kafka.server:type=BrokerTopicMetrics,name=RemoteCopyBytesPerSec,topic=([-.\w]+) |
| Remote Copy Requests Per Sec | Rate of write requests to remote storage per topic. Omitting 'topic=(...)' will yield the all-topic rate | kafka.server:type=BrokerTopicMetrics,name=RemoteCopyRequestsPerSec,topic=([-.\w]+) |
| Remote Copy Errors Per Sec | Rate of write errors from remote storage per topic. Omitting 'topic=(...)' will yield the all-topic rate | kafka.server:type=BrokerTopicMetrics,name=RemoteCopyErrorsPerSec,topic=([-.\w]+) |
| RemoteLogReader Task Queue Size | Size of the queue holding remote storage read tasks | org.apache.kafka.storage.internals.log:type=RemoteStorageThreadPool,name=RemoteLogReaderTaskQueueSize |
| RemoteLogReader Avg Idle Percent | Average idle percent of thread pool for processing remote storage read tasks | org.apache.kafka.storage.internals.log:type=RemoteStorageThreadPool,name=RemoteLogReaderAvgIdlePercent |
| RemoteLogManager Tasks Avg Idle Percent | Average idle percent of thread pool for copying data to remote storage | kafka.log.remote:type=RemoteLogManager,name=RemoteLogManagerTasksAvgIdlePercent |
KRaft Monitoring Metrics
The set of metrics that allow monitoring of the KRaft quorum and the metadata log.Note that some exposed metrics depend on the role of the node as defined by
process.roles
KRaft Quorum Monitoring Metrics
These metrics are reported on both Controllers and Brokers in a KRaft Cluster| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| Current State | The current state of this member; possible values are leader, candidate, voted, follower, unattached, observer. | kafka.server:type=raft-metrics,name=current-state |
| Current Leader | The current quorum leader's id; -1 indicates unknown. | kafka.server:type=raft-metrics,name=current-leader |
| Current Voted | The current voted leader's id; -1 indicates not voted for anyone. | kafka.server:type=raft-metrics,name=current-vote |
| Current Epoch | The current quorum epoch. | kafka.server:type=raft-metrics,name=current-epoch |
| High Watermark | The high watermark maintained on this member; -1 if it is unknown. | kafka.server:type=raft-metrics,name=high-watermark |
| Log End Offset | The current raft log end offset. | kafka.server:type=raft-metrics,name=log-end-offset |
| Number of Unknown Voter Connections | Number of unknown voters whose connection information is not cached. This value of this metric is always 0. | kafka.server:type=raft-metrics,name=number-unknown-voter-connections |
| Average Commit Latency | The average time in milliseconds to commit an entry in the raft log. | kafka.server:type=raft-metrics,name=commit-latency-avg |
| Maximum Commit Latency | The maximum time in milliseconds to commit an entry in the raft log. | kafka.server:type=raft-metrics,name=commit-latency-max |
| Average Election Latency | The average time in milliseconds spent on electing a new leader. | kafka.server:type=raft-metrics,name=election-latency-avg |
| Maximum Election Latency | The maximum time in milliseconds spent on electing a new leader. | kafka.server:type=raft-metrics,name=election-latency-max |
| Fetch Records Rate | The average number of records fetched from the leader of the raft quorum. | kafka.server:type=raft-metrics,name=fetch-records-rate |
| Append Records Rate | The average number of records appended per sec by the leader of the raft quorum. | kafka.server:type=raft-metrics,name=append-records-rate |
| Average Poll Idle Ratio | The average fraction of time the client's poll() is idle as opposed to waiting for the user code to process records. | kafka.server:type=raft-metrics,name=poll-idle-ratio-avg |
| Current Metadata Version | Outputs the feature level of the current effective metadata version. | kafka.server:type=MetadataLoader,name=CurrentMetadataVersion |
| Metadata Snapshot Load Count | The total number of times we have loaded a KRaft snapshot since the process was started. | kafka.server:type=MetadataLoader,name=HandleLoadSnapshotCount |
| Latest Metadata Snapshot Size | The total size in bytes of the latest snapshot that the node has generated. If none have been generated yet, this is the size of the latest snapshot that was loaded. If no snapshots have been generated or loaded, this is 0. | kafka.server:type=SnapshotEmitter,name=LatestSnapshotGeneratedBytes |
| Latest Metadata Snapshot Age | The interval in milliseconds since the latest snapshot that the node has generated. If none have been generated yet, this is approximately the time delta since the process was started. | kafka.server:type=SnapshotEmitter,name=LatestSnapshotGeneratedAgeMs |
KRaft Controller Monitoring Metrics
| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| Active Controller Count | The number of Active Controllers on this node. Valid values are '0' or '1'. | kafka.controller:type=KafkaController,name=ActiveControllerCount |
| Event Queue Time Ms | A Histogram of the time in milliseconds that requests spent waiting in the Controller Event Queue. | kafka.controller:type=ControllerEventManager,name=EventQueueTimeMs |
| Event Queue Processing Time Ms | A Histogram of the time in milliseconds that requests spent being processed in the Controller Event Queue. | kafka.controller:type=ControllerEventManager,name=EventQueueProcessingTimeMs |
| Fenced Broker Count | The number of fenced brokers as observed by this Controller. | kafka.controller:type=KafkaController,name=FencedBrokerCount |
| Active Broker Count | The number of active brokers as observed by this Controller. | kafka.controller:type=KafkaController,name=ActiveBrokerCount |
| Global Topic Count | The number of global topics as observed by this Controller. | kafka.controller:type=KafkaController,name=GlobalTopicCount |
| Global Partition Count | The number of global partitions as observed by this Controller. | kafka.controller:type=KafkaController,name=GlobalPartitionCount |
| Offline Partition Count | The number of offline topic partitions (non-internal) as observed by this Controller. | kafka.controller:type=KafkaController,name=OfflinePartitionCount |
| Preferred Replica Imbalance Count | The count of topic partitions for which the leader is not the preferred leader. | kafka.controller:type=KafkaController,name=PreferredReplicaImbalanceCount |
| Metadata Error Count | The number of times this controller node has encountered an error during metadata log processing. | kafka.controller:type=KafkaController,name=MetadataErrorCount |
| Last Applied Record Offset | The offset of the last record from the cluster metadata partition that was applied by the Controller. | kafka.controller:type=KafkaController,name=LastAppliedRecordOffset |
| Last Committed Record Offset | The offset of the last record committed to this Controller. | kafka.controller:type=KafkaController,name=LastCommittedRecordOffset |
| Last Applied Record Timestamp | The timestamp of the last record from the cluster metadata partition that was applied by the Controller. | kafka.controller:type=KafkaController,name=LastAppliedRecordTimestamp |
| Last Applied Record Lag Ms | The difference between now and the timestamp of the last record from the cluster metadata partition that was applied by the controller. For active Controllers the value of this lag is always zero. | kafka.controller:type=KafkaController,name=LastAppliedRecordLagMs |
| ZooKeeper Write Behind Lag | The amount of lag in records that ZooKeeper is behind relative to the highest committed record in the metadata log. This metric will only be reported by the active KRaft controller. | kafka.controller:type=KafkaController,name=ZkWriteBehindLag |
| ZooKeeper Metadata Snapshot Write Time | The number of milliseconds the KRaft controller took reconciling a snapshot into ZooKeeper. | kafka.controller:type=KafkaController,name=ZkWriteSnapshotTimeMs |
| ZooKeeper Metadata Delta Write Time | The number of milliseconds the KRaft controller took writing a delta into ZK. | kafka.controller:type=KafkaController,name=ZkWriteDeltaTimeMs |
| Timed-out Broker Heartbeat Count | The number of broker heartbeats that timed out on this controller since the process was started. Note that only active controllers handle heartbeats, so only they will see increases in this metric. | kafka.controller:type=KafkaController,name=TimedOutBrokerHeartbeatCount |
| Number Of Operations Started In Event Queue | The total number of controller event queue operations that were started. This includes deferred operations. | kafka.controller:type=KafkaController,name=EventQueueOperationsStartedCount |
| Number of Operations Timed Out In Event Queue | The total number of controller event queue operations that timed out before they could be performed. | kafka.controller:type=KafkaController,name=EventQueueOperationsTimedOutCount |
| Number Of New Controller Elections | Counts the number of times this node has seen a new controller elected. A transition to the "no leader" state is not counted here. If the same controller as before becomes active, that still counts. | kafka.controller:type=KafkaController,name=NewActiveControllersCount |
KRaft Broker Monitoring Metrics
| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| Last Applied Record Offset | The offset of the last record from the cluster metadata partition that was applied by the broker | kafka.server:type=broker-metadata-metrics,name=last-applied-record-offset |
| Last Applied Record Timestamp | The timestamp of the last record from the cluster metadata partition that was applied by the broker. | kafka.server:type=broker-metadata-metrics,name=last-applied-record-timestamp |
| Last Applied Record Lag Ms | The difference between now and the timestamp of the last record from the cluster metadata partition that was applied by the broker | kafka.server:type=broker-metadata-metrics,name=last-applied-record-lag-ms |
| Metadata Load Error Count | The number of errors encountered by the BrokerMetadataListener while loading the metadata log and generating a new MetadataDelta based on it. | kafka.server:type=broker-metadata-metrics,name=metadata-load-error-count |
| Metadata Apply Error Count | The number of errors encountered by the BrokerMetadataPublisher while applying a new MetadataImage based on the latest MetadataDelta. | kafka.server:type=broker-metadata-metrics,name=metadata-apply-error-count |
Common monitoring metrics for producer/consumer/connect/streams
The following metrics are available on producer/consumer/connector/streams instances. For specific metrics, please see following sections.| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| connection-close-rate | Connections closed per second in the window. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| connection-close-total | Total connections closed in the window. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| connection-creation-rate | New connections established per second in the window. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| connection-creation-total | Total new connections established in the window. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| network-io-rate | The average number of network operations (reads or writes) on all connections per second. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| network-io-total | The total number of network operations (reads or writes) on all connections. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| outgoing-byte-rate | The average number of outgoing bytes sent per second to all servers. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| outgoing-byte-total | The total number of outgoing bytes sent to all servers. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| request-rate | The average number of requests sent per second. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| request-total | The total number of requests sent. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| request-size-avg | The average size of all requests in the window. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| request-size-max | The maximum size of any request sent in the window. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| incoming-byte-rate | Bytes/second read off all sockets. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| incoming-byte-total | Total bytes read off all sockets. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| response-rate | Responses received per second. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| response-total | Total responses received. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| select-rate | Number of times the I/O layer checked for new I/O to perform per second. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| select-total | Total number of times the I/O layer checked for new I/O to perform. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| io-wait-time-ns-avg | The average length of time the I/O thread spent waiting for a socket ready for reads or writes in nanoseconds. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| io-wait-time-ns-total | The total time the I/O thread spent waiting in nanoseconds. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| io-waittime-total | *Deprecated* The total time the I/O thread spent waiting in nanoseconds. Replacement is io-wait-time-ns-total. |
kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| io-wait-ratio | The fraction of time the I/O thread spent waiting. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| io-time-ns-avg | The average length of time for I/O per select call in nanoseconds. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| io-time-ns-total | The total time the I/O thread spent doing I/O in nanoseconds. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| iotime-total | *Deprecated* The total time the I/O thread spent doing I/O in nanoseconds. Replacement is io-time-ns-total. |
kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| io-ratio | The fraction of time the I/O thread spent doing I/O. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| connection-count | The current number of active connections. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| successful-authentication-rate | Connections per second that were successfully authenticated using SASL or SSL. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| successful-authentication-total | Total connections that were successfully authenticated using SASL or SSL. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| failed-authentication-rate | Connections per second that failed authentication. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| failed-authentication-total | Total connections that failed authentication. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| successful-reauthentication-rate | Connections per second that were successfully re-authenticated using SASL. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| successful-reauthentication-total | Total connections that were successfully re-authenticated using SASL. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| reauthentication-latency-max | The maximum latency in ms observed due to re-authentication. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| reauthentication-latency-avg | The average latency in ms observed due to re-authentication. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| failed-reauthentication-rate | Connections per second that failed re-authentication. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| failed-reauthentication-total | Total connections that failed re-authentication. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
| successful-authentication-no-reauth-total | Total connections that were successfully authenticated by older, pre-2.2.0 SASL clients that do not support re-authentication. May only be non-zero. | kafka.[producer|consumer|connect]:type=[producer|consumer|connect]-metrics,client-id=([-.\w]+) |
Common Per-broker metrics for producer/consumer/connect/streams
The following metrics are available on producer/consumer/connector/streams instances. For specific metrics, please see following sections.| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| outgoing-byte-rate | The average number of outgoing bytes sent per second for a node. | kafka.[producer|consumer|connect]:type=[consumer|producer|connect]-node-metrics,client-id=([-.\w]+),node-id=([0-9]+) |
| outgoing-byte-total | The total number of outgoing bytes sent for a node. | kafka.[producer|consumer|connect]:type=[consumer|producer|connect]-node-metrics,client-id=([-.\w]+),node-id=([0-9]+) |
| request-rate | The average number of requests sent per second for a node. | kafka.[producer|consumer|connect]:type=[consumer|producer|connect]-node-metrics,client-id=([-.\w]+),node-id=([0-9]+) |
| request-total | The total number of requests sent for a node. | kafka.[producer|consumer|connect]:type=[consumer|producer|connect]-node-metrics,client-id=([-.\w]+),node-id=([0-9]+) |
| request-size-avg | The average size of all requests in the window for a node. | kafka.[producer|consumer|connect]:type=[consumer|producer|connect]-node-metrics,client-id=([-.\w]+),node-id=([0-9]+) |
| request-size-max | The maximum size of any request sent in the window for a node. | kafka.[producer|consumer|connect]:type=[consumer|producer|connect]-node-metrics,client-id=([-.\w]+),node-id=([0-9]+) |
| incoming-byte-rate | The average number of bytes received per second for a node. | kafka.[producer|consumer|connect]:type=[consumer|producer|connect]-node-metrics,client-id=([-.\w]+),node-id=([0-9]+) |
| incoming-byte-total | The total number of bytes received for a node. | kafka.[producer|consumer|connect]:type=[consumer|producer|connect]-node-metrics,client-id=([-.\w]+),node-id=([0-9]+) |
| request-latency-avg | The average request latency in ms for a node. | kafka.[producer|consumer|connect]:type=[consumer|producer|connect]-node-metrics,client-id=([-.\w]+),node-id=([0-9]+) |
| request-latency-max | The maximum request latency in ms for a node. | kafka.[producer|consumer|connect]:type=[consumer|producer|connect]-node-metrics,client-id=([-.\w]+),node-id=([0-9]+) |
| response-rate | Responses received per second for a node. | kafka.[producer|consumer|connect]:type=[consumer|producer|connect]-node-metrics,client-id=([-.\w]+),node-id=([0-9]+) |
| response-total | Total responses received for a node. | kafka.[producer|consumer|connect]:type=[consumer|producer|connect]-node-metrics,client-id=([-.\w]+),node-id=([0-9]+) |
Producer monitoring
The following metrics are available on producer instances.| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| waiting-threads | The number of user threads blocked waiting for buffer memory to enqueue their records. | kafka.producer:type=producer-metrics,client-id=([-.\w]+) |
| buffer-total-bytes | The maximum amount of buffer memory the client can use (whether or not it is currently used). | kafka.producer:type=producer-metrics,client-id=([-.\w]+) |
| buffer-available-bytes | The total amount of buffer memory that is not being used (either unallocated or in the free list). | kafka.producer:type=producer-metrics,client-id=([-.\w]+) |
| bufferpool-wait-time | The fraction of time an appender waits for space allocation. | kafka.producer:type=producer-metrics,client-id=([-.\w]+) |
| bufferpool-wait-time-total | *Deprecated* The total time an appender waits for space allocation in nanoseconds. Replacement is bufferpool-wait-time-ns-total |
kafka.producer:type=producer-metrics,client-id=([-.\w]+) |
| bufferpool-wait-time-ns-total | The total time an appender waits for space allocation in nanoseconds. | kafka.producer:type=producer-metrics,client-id=([-.\w]+) |
| flush-time-ns-total | The total time the Producer spent in Producer.flush in nanoseconds. | kafka.producer:type=producer-metrics,client-id=([-.\w]+) |
| txn-init-time-ns-total | The total time the Producer spent initializing transactions in nanoseconds (for EOS). | kafka.producer:type=producer-metrics,client-id=([-.\w]+) |
| txn-begin-time-ns-total | The total time the Producer spent in beginTransaction in nanoseconds (for EOS). | kafka.producer:type=producer-metrics,client-id=([-.\w]+) |
| txn-send-offsets-time-ns-total | The total time the Producer spent sending offsets to transactions in nanoseconds (for EOS). | kafka.producer:type=producer-metrics,client-id=([-.\w]+) |
| txn-commit-time-ns-total | The total time the Producer spent committing transactions in nanoseconds (for EOS). | kafka.producer:type=producer-metrics,client-id=([-.\w]+) |
| txn-abort-time-ns-total | The total time the Producer spent aborting transactions in nanoseconds (for EOS). | kafka.producer:type=producer-metrics,client-id=([-.\w]+) |
Producer Sender Metrics
| kafka.producer:type=producer-metrics,client-id="{client-id}" | ||
| Attribute name | Description | |
|---|---|---|
| batch-size-avg | The average number of bytes sent per partition per-request. | |
| batch-size-max | The max number of bytes sent per partition per-request. | |
| batch-split-rate | The average number of batch splits per second | |
| batch-split-total | The total number of batch splits | |
| compression-rate-avg | The average compression rate of record batches, defined as the average ratio of the compressed batch size over the uncompressed size. | |
| metadata-age | The age in seconds of the current producer metadata being used. | |
| produce-throttle-time-avg | The average time in ms a request was throttled by a broker | |
| produce-throttle-time-max | The maximum time in ms a request was throttled by a broker | |
| record-error-rate | The average per-second number of record sends that resulted in errors | |
| record-error-total | The total number of record sends that resulted in errors | |
| record-queue-time-avg | The average time in ms record batches spent in the send buffer. | |
| record-queue-time-max | The maximum time in ms record batches spent in the send buffer. | |
| record-retry-rate | The average per-second number of retried record sends | |
| record-retry-total | The total number of retried record sends | |
| record-send-rate | The average number of records sent per second. | |
| record-send-total | The total number of records sent. | |
| record-size-avg | The average record size | |
| record-size-max | The maximum record size | |
| records-per-request-avg | The average number of records per request. | |
| request-latency-avg | The average request latency in ms | |
| request-latency-max | The maximum request latency in ms | |
| requests-in-flight | The current number of in-flight requests awaiting a response. | |
| kafka.producer:type=producer-topic-metrics,client-id="{client-id}",topic="{topic}" | ||
| Attribute name | Description | |
| byte-rate | The average number of bytes sent per second for a topic. | |
| byte-total | The total number of bytes sent for a topic. | |
| compression-rate | The average compression rate of record batches for a topic, defined as the average ratio of the compressed batch size over the uncompressed size. | |
| record-error-rate | The average per-second number of record sends that resulted in errors for a topic | |
| record-error-total | The total number of record sends that resulted in errors for a topic | |
| record-retry-rate | The average per-second number of retried record sends for a topic | |
| record-retry-total | The total number of retried record sends for a topic | |
| record-send-rate | The average number of records sent per second for a topic. | |
| record-send-total | The total number of records sent for a topic. | |
Consumer monitoring
The following metrics are available on consumer instances.| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| time-between-poll-avg | The average delay between invocations of poll(). | kafka.consumer:type=consumer-metrics,client-id=([-.\w]+) |
| time-between-poll-max | The max delay between invocations of poll(). | kafka.consumer:type=consumer-metrics,client-id=([-.\w]+) |
| last-poll-seconds-ago | The number of seconds since the last poll() invocation. | kafka.consumer:type=consumer-metrics,client-id=([-.\w]+) |
| poll-idle-ratio-avg | The average fraction of time the consumer's poll() is idle as opposed to waiting for the user code to process records. | kafka.consumer:type=consumer-metrics,client-id=([-.\w]+) |
| committed-time-ns-total | The total time the Consumer spent in committed in nanoseconds. | kafka.consumer:type=consumer-metrics,client-id=([-.\w]+) |
| commit-sync-time-ns-total | The total time the Consumer spent committing offsets in nanoseconds (for AOS). | kafka.consumer:type=consumer-metrics,client-id=([-.\w]+) |
Consumer Group Metrics
| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| commit-latency-avg | The average time taken for a commit request | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| commit-latency-max | The max time taken for a commit request | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| commit-rate | The number of commit calls per second | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| commit-total | The total number of commit calls | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| assigned-partitions | The number of partitions currently assigned to this consumer | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| heartbeat-response-time-max | The max time taken to receive a response to a heartbeat request | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| heartbeat-rate | The average number of heartbeats per second | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| heartbeat-total | The total number of heartbeats | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| join-time-avg | The average time taken for a group rejoin | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| join-time-max | The max time taken for a group rejoin | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| join-rate | The number of group joins per second | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| join-total | The total number of group joins | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| sync-time-avg | The average time taken for a group sync | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| sync-time-max | The max time taken for a group sync | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| sync-rate | The number of group syncs per second | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| sync-total | The total number of group syncs | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| rebalance-latency-avg | The average time taken for a group rebalance | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| rebalance-latency-max | The max time taken for a group rebalance | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| rebalance-latency-total | The total time taken for group rebalances so far | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| rebalance-total | The total number of group rebalances participated | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| rebalance-rate-per-hour | The number of group rebalance participated per hour | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| failed-rebalance-total | The total number of failed group rebalances | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| failed-rebalance-rate-per-hour | The number of failed group rebalance event per hour | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| last-rebalance-seconds-ago | The number of seconds since the last rebalance event | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| last-heartbeat-seconds-ago | The number of seconds since the last controller heartbeat | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| partitions-revoked-latency-avg | The average time taken by the on-partitions-revoked rebalance listener callback | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| partitions-revoked-latency-max | The max time taken by the on-partitions-revoked rebalance listener callback | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| partitions-assigned-latency-avg | The average time taken by the on-partitions-assigned rebalance listener callback | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| partitions-assigned-latency-max | The max time taken by the on-partitions-assigned rebalance listener callback | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| partitions-lost-latency-avg | The average time taken by the on-partitions-lost rebalance listener callback | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
| partitions-lost-latency-max | The max time taken by the on-partitions-lost rebalance listener callback | kafka.consumer:type=consumer-coordinator-metrics,client-id=([-.\w]+) |
Consumer Fetch Metrics
| kafka.consumer:type=consumer-fetch-manager-metrics,client-id="{client-id}" | ||
| Attribute name | Description | |
|---|---|---|
| bytes-consumed-rate | The average number of bytes consumed per second | |
| bytes-consumed-total | The total number of bytes consumed | |
| fetch-latency-avg | The average time taken for a fetch request. | |
| fetch-latency-max | The max time taken for any fetch request. | |
| fetch-rate | The number of fetch requests per second. | |
| fetch-size-avg | The average number of bytes fetched per request | |
| fetch-size-max | The maximum number of bytes fetched per request | |
| fetch-throttle-time-avg | The average throttle time in ms | |
| fetch-throttle-time-max | The maximum throttle time in ms | |
| fetch-total | The total number of fetch requests. | |
| records-consumed-rate | The average number of records consumed per second | |
| records-consumed-total | The total number of records consumed | |
| records-lag-max | The maximum lag in terms of number of records for any partition in this window. NOTE: This is based on current offset and not committed offset | |
| records-lead-min | The minimum lead in terms of number of records for any partition in this window | |
| records-per-request-avg | The average number of records in each request | |
| kafka.consumer:type=consumer-fetch-manager-metrics,client-id="{client-id}",topic="{topic}" | ||
| Attribute name | Description | |
| bytes-consumed-rate | The average number of bytes consumed per second for a topic | |
| bytes-consumed-total | The total number of bytes consumed for a topic | |
| fetch-size-avg | The average number of bytes fetched per request for a topic | |
| fetch-size-max | The maximum number of bytes fetched per request for a topic | |
| records-consumed-rate | The average number of records consumed per second for a topic | |
| records-consumed-total | The total number of records consumed for a topic | |
| records-per-request-avg | The average number of records in each request for a topic | |
| kafka.consumer:type=consumer-fetch-manager-metrics,partition="{partition}",topic="{topic}",client-id="{client-id}" | ||
| Attribute name | Description | |
| preferred-read-replica | The current read replica for the partition, or -1 if reading from leader | |
| records-lag | The latest lag of the partition | |
| records-lag-avg | The average lag of the partition | |
| records-lag-max | The max lag of the partition | |
| records-lead | The latest lead of the partition | |
| records-lead-avg | The average lead of the partition | |
| records-lead-min | The min lead of the partition | |
Connect Monitoring
A Connect worker process contains all the producer and consumer metrics as well as metrics specific to Connect. The worker process itself has a number of metrics, while each connector and task have additional metrics. [2023-09-15 00:40:42,725] INFO Metrics scheduler closed (org.apache.kafka.common.metrics.Metrics:693) [2023-09-15 00:40:42,729] INFO Metrics reporters closed (org.apache.kafka.common.metrics.Metrics:703)| kafka.connect:type=connect-worker-metrics | ||
| Attribute name | Description | |
|---|---|---|
| connector-count | The number of connectors run in this worker. | |
| connector-startup-attempts-total | The total number of connector startups that this worker has attempted. | |
| connector-startup-failure-percentage | The average percentage of this worker's connectors starts that failed. | |
| connector-startup-failure-total | The total number of connector starts that failed. | |
| connector-startup-success-percentage | The average percentage of this worker's connectors starts that succeeded. | |
| connector-startup-success-total | The total number of connector starts that succeeded. | |
| task-count | The number of tasks run in this worker. | |
| task-startup-attempts-total | The total number of task startups that this worker has attempted. | |
| task-startup-failure-percentage | The average percentage of this worker's tasks starts that failed. | |
| task-startup-failure-total | The total number of task starts that failed. | |
| task-startup-success-percentage | The average percentage of this worker's tasks starts that succeeded. | |
| task-startup-success-total | The total number of task starts that succeeded. | |
| kafka.connect:type=connect-worker-metrics,connector="{connector}" | ||
| Attribute name | Description | |
| connector-destroyed-task-count | The number of destroyed tasks of the connector on the worker. | |
| connector-failed-task-count | The number of failed tasks of the connector on the worker. | |
| connector-paused-task-count | The number of paused tasks of the connector on the worker. | |
| connector-restarting-task-count | The number of restarting tasks of the connector on the worker. | |
| connector-running-task-count | The number of running tasks of the connector on the worker. | |
| connector-total-task-count | The number of tasks of the connector on the worker. | |
| connector-unassigned-task-count | The number of unassigned tasks of the connector on the worker. | |
| kafka.connect:type=connect-worker-rebalance-metrics | ||
| Attribute name | Description | |
| completed-rebalances-total | The total number of rebalances completed by this worker. | |
| connect-protocol | The Connect protocol used by this cluster | |
| epoch | The epoch or generation number of this worker. | |
| leader-name | The name of the group leader. | |
| rebalance-avg-time-ms | The average time in milliseconds spent by this worker to rebalance. | |
| rebalance-max-time-ms | The maximum time in milliseconds spent by this worker to rebalance. | |
| rebalancing | Whether this worker is currently rebalancing. | |
| time-since-last-rebalance-ms | The time in milliseconds since this worker completed the most recent rebalance. | |
| kafka.connect:type=connector-metrics,connector="{connector}" | ||
| Attribute name | Description | |
| connector-class | The name of the connector class. | |
| connector-type | The type of the connector. One of 'source' or 'sink'. | |
| connector-version | The version of the connector class, as reported by the connector. | |
| status | The status of the connector. One of 'unassigned', 'running', 'paused', 'stopped', 'failed', or 'restarting'. | |
| kafka.connect:type=connector-task-metrics,connector="{connector}",task="{task}" | ||
| Attribute name | Description | |
| batch-size-avg | The average number of records in the batches the task has processed so far. | |
| batch-size-max | The number of records in the largest batch the task has processed so far. | |
| offset-commit-avg-time-ms | The average time in milliseconds taken by this task to commit offsets. | |
| offset-commit-failure-percentage | The average percentage of this task's offset commit attempts that failed. | |
| offset-commit-max-time-ms | The maximum time in milliseconds taken by this task to commit offsets. | |
| offset-commit-success-percentage | The average percentage of this task's offset commit attempts that succeeded. | |
| pause-ratio | The fraction of time this task has spent in the pause state. | |
| running-ratio | The fraction of time this task has spent in the running state. | |
| status | The status of the connector task. One of 'unassigned', 'running', 'paused', 'failed', or 'restarting'. | |
| kafka.connect:type=sink-task-metrics,connector="{connector}",task="{task}" | ||
| Attribute name | Description | |
| offset-commit-completion-rate | The average per-second number of offset commit completions that were completed successfully. | |
| offset-commit-completion-total | The total number of offset commit completions that were completed successfully. | |
| offset-commit-seq-no | The current sequence number for offset commits. | |
| offset-commit-skip-rate | The average per-second number of offset commit completions that were received too late and skipped/ignored. | |
| offset-commit-skip-total | The total number of offset commit completions that were received too late and skipped/ignored. | |
| partition-count | The number of topic partitions assigned to this task belonging to the named sink connector in this worker. | |
| put-batch-avg-time-ms | The average time taken by this task to put a batch of sinks records. | |
| put-batch-max-time-ms | The maximum time taken by this task to put a batch of sinks records. | |
| sink-record-active-count | The number of records that have been read from Kafka but not yet completely committed/flushed/acknowledged by the sink task. | |
| sink-record-active-count-avg | The average number of records that have been read from Kafka but not yet completely committed/flushed/acknowledged by the sink task. | |
| sink-record-active-count-max | The maximum number of records that have been read from Kafka but not yet completely committed/flushed/acknowledged by the sink task. | |
| sink-record-lag-max | The maximum lag in terms of number of records that the sink task is behind the consumer's position for any topic partitions. | |
| sink-record-read-rate | The average per-second number of records read from Kafka for this task belonging to the named sink connector in this worker. This is before transformations are applied. | |
| sink-record-read-total | The total number of records read from Kafka by this task belonging to the named sink connector in this worker, since the task was last restarted. | |
| sink-record-send-rate | The average per-second number of records output from the transformations and sent/put to this task belonging to the named sink connector in this worker. This is after transformations are applied and excludes any records filtered out by the transformations. | |
| sink-record-send-total | The total number of records output from the transformations and sent/put to this task belonging to the named sink connector in this worker, since the task was last restarted. | |
| kafka.connect:type=source-task-metrics,connector="{connector}",task="{task}" | ||
| Attribute name | Description | |
| poll-batch-avg-time-ms | The average time in milliseconds taken by this task to poll for a batch of source records. | |
| poll-batch-max-time-ms | The maximum time in milliseconds taken by this task to poll for a batch of source records. | |
| source-record-active-count | The number of records that have been produced by this task but not yet completely written to Kafka. | |
| source-record-active-count-avg | The average number of records that have been produced by this task but not yet completely written to Kafka. | |
| source-record-active-count-max | The maximum number of records that have been produced by this task but not yet completely written to Kafka. | |
| source-record-poll-rate | The average per-second number of records produced/polled (before transformation) by this task belonging to the named source connector in this worker. | |
| source-record-poll-total | The total number of records produced/polled (before transformation) by this task belonging to the named source connector in this worker. | |
| source-record-write-rate | The average per-second number of records written to Kafka for this task belonging to the named source connector in this worker, since the task was last restarted. This is after transformations are applied, and excludes any records filtered out by the transformations. | |
| source-record-write-total | The number of records output written to Kafka for this task belonging to the named source connector in this worker, since the task was last restarted. This is after transformations are applied, and excludes any records filtered out by the transformations. | |
| transaction-size-avg | The average number of records in the transactions the task has committed so far. | |
| transaction-size-max | The number of records in the largest transaction the task has committed so far. | |
| transaction-size-min | The number of records in the smallest transaction the task has committed so far. | |
| kafka.connect:type=task-error-metrics,connector="{connector}",task="{task}" | ||
| Attribute name | Description | |
| deadletterqueue-produce-failures | The number of failed writes to the dead letter queue. | |
| deadletterqueue-produce-requests | The number of attempted writes to the dead letter queue. | |
| last-error-timestamp | The epoch timestamp when this task last encountered an error. | |
| total-errors-logged | The number of errors that were logged. | |
| total-record-errors | The number of record processing errors in this task. | |
| total-record-failures | The number of record processing failures in this task. | |
| total-records-skipped | The number of records skipped due to errors. | |
| total-retries | The number of operations retried. | |
Streams Monitoring
A Kafka Streams instance contains all the producer and consumer metrics as well as additional metrics specific to Streams. The metrics have three recording levels:info, debug, and trace.
Note that the metrics have a 4-layer hierarchy. At the top level there are client-level metrics for each started Kafka Streams client. Each client has stream threads, with their own metrics. Each stream thread has tasks, with their own metrics. Each task has a number of processor nodes, with their own metrics. Each task also has a number of state stores and record caches, all with their own metrics.
Use the following configuration option to specify which metrics you want collected:metrics.recording.level="info"Client Metrics
All of the following metrics have a recording level ofinfo:
| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| version | The version of the Kafka Streams client. | kafka.streams:type=stream-metrics,client-id=([-.\w]+) |
| commit-id | The version control commit ID of the Kafka Streams client. | kafka.streams:type=stream-metrics,client-id=([-.\w]+) |
| application-id | The application ID of the Kafka Streams client. | kafka.streams:type=stream-metrics,client-id=([-.\w]+) |
| topology-description | The description of the topology executed in the Kafka Streams client. | kafka.streams:type=stream-metrics,client-id=([-.\w]+) |
| state | The state of the Kafka Streams client. | kafka.streams:type=stream-metrics,client-id=([-.\w]+) |
| failed-stream-threads | The number of failed stream threads since the start of the Kafka Streams client. | kafka.streams:type=stream-metrics,client-id=([-.\w]+) |
Thread Metrics
All of the following metrics have a recording level ofinfo:
| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| commit-latency-avg | The average execution time in ms, for committing, across all running tasks of this thread. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| commit-latency-max | The maximum execution time in ms, for committing, across all running tasks of this thread. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| poll-latency-avg | The average execution time in ms, for consumer polling. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| poll-latency-max | The maximum execution time in ms, for consumer polling. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| process-latency-avg | The average execution time in ms, for processing. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| process-latency-max | The maximum execution time in ms, for processing. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| punctuate-latency-avg | The average execution time in ms, for punctuating. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| punctuate-latency-max | The maximum execution time in ms, for punctuating. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| commit-rate | The average number of commits per second. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| commit-total | The total number of commit calls. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| poll-rate | The average number of consumer poll calls per second. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| poll-total | The total number of consumer poll calls. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| process-rate | The average number of processed records per second. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| process-total | The total number of processed records. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| punctuate-rate | The average number of punctuate calls per second. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| punctuate-total | The total number of punctuate calls. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| task-created-rate | The average number of tasks created per second. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| task-created-total | The total number of tasks created. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| task-closed-rate | The average number of tasks closed per second. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| task-closed-total | The total number of tasks closed. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| blocked-time-ns-total | The total time the thread spent blocked on kafka. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
| thread-start-time | The time that the thread was started. | kafka.streams:type=stream-thread-metrics,thread-id=([-.\w]+) |
Task Metrics
All of the following metrics have a recording level ofdebug, except for the dropped-records-* and
active-process-ratio metrics which have a recording level of info:
| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| process-latency-avg | The average execution time in ns, for processing. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| process-latency-max | The maximum execution time in ns, for processing. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| process-rate | The average number of processed records per second across all source processor nodes of this task. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| process-total | The total number of processed records across all source processor nodes of this task. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| commit-latency-avg | The average execution time in ns, for committing. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| commit-latency-max | The maximum execution time in ns, for committing. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| commit-rate | The average number of commit calls per second. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| commit-total | The total number of commit calls. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| record-lateness-avg | The average observed lateness of records (stream time - record timestamp). | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| record-lateness-max | The max observed lateness of records (stream time - record timestamp). | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| enforced-processing-rate | The average number of enforced processings per second. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| enforced-processing-total | The total number enforced processings. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| dropped-records-rate | The average number of records dropped within this task. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| dropped-records-total | The total number of records dropped within this task. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| active-process-ratio | The fraction of time the stream thread spent on processing this task among all assigned active tasks. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| input-buffer-bytes-total | The total number of bytes accumulated by this task, | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
| cache-size-bytes-total | The cache size in bytes accumulated by this task. | kafka.streams:type=stream-task-metrics,thread-id=([-.\w]+),task-id=([-.\w]+) |
Processor Node Metrics
The following metrics are only available on certain types of nodes, i.e., the process-* metrics are only available for source processor nodes, thesuppression-emit-* metrics are only available for suppression operation nodes,
emit-final-* metrics are only available for windowed aggregations nodes, and the
record-e2e-latency-* metrics are only available for source processor nodes and terminal nodes (nodes without successor
nodes).
All of the metrics have a recording level of debug, except for the record-e2e-latency-* metrics which have
a recording level of info:
| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| bytes-consumed-total | The total number of bytes consumed by a source processor node. | kafka.streams:type=stream-topic-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+),topic=([-.\w]+) |
| bytes-produced-total | The total number of bytes produced by a sink processor node. | kafka.streams:type=stream-topic-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+),topic=([-.\w]+) |
| process-rate | The average number of records processed by a source processor node per second. | kafka.streams:type=stream-processor-node-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+) |
| process-total | The total number of records processed by a source processor node per second. | kafka.streams:type=stream-processor-node-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+) |
| suppression-emit-rate | The rate at which records that have been emitted downstream from suppression operation nodes. | kafka.streams:type=stream-processor-node-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+) |
| suppression-emit-total | The total number of records that have been emitted downstream from suppression operation nodes. | kafka.streams:type=stream-processor-node-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+) |
| emit-final-latency-max | The max latency to emit final records when a record could be emitted. | kafka.streams:type=stream-processor-node-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+) |
| emit-final-latency-avg | The avg latency to emit final records when a record could be emitted. | kafka.streams:type=stream-processor-node-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+) |
| emit-final-records-rate | The rate of records emitted when records could be emitted. | kafka.streams:type=stream-processor-node-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+) |
| emit-final-records-total | The total number of records emitted. | kafka.streams:type=stream-processor-node-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+) |
| record-e2e-latency-avg | The average end-to-end latency of a record, measured by comparing the record timestamp with the system time when it has been fully processed by the node. | kafka.streams:type=stream-processor-node-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+) |
| record-e2e-latency-max | The maximum end-to-end latency of a record, measured by comparing the record timestamp with the system time when it has been fully processed by the node. | kafka.streams:type=stream-processor-node-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+) |
| record-e2e-latency-min | The minimum end-to-end latency of a record, measured by comparing the record timestamp with the system time when it has been fully processed by the node. | kafka.streams:type=stream-processor-node-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+) |
| records-consumed-total | The total number of records consumed by a source processor node. | kafka.streams:type=stream-topic-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+),topic=([-.\w]+) |
| records-produced-total | The total number of records produced by a sink processor node. | kafka.streams:type=stream-topic-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),processor-node-id=([-.\w]+),topic=([-.\w]+) |
State Store Metrics
All of the following metrics have a recording level ofdebug, except for the record-e2e-latency-* metrics which have a recording level trace.
Note that the store-scope value is specified in StoreSupplier#metricsScope() for user's customized state stores;
for built-in state stores, currently we have:
in-memory-statein-memory-lru-statein-memory-window-statein-memory-suppression(for suppression buffers)rocksdb-state(for RocksDB backed key-value store)rocksdb-window-state(for RocksDB backed window store)rocksdb-session-state(for RocksDB backed session store)
| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| put-latency-avg | The average put execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| put-latency-max | The maximum put execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| put-if-absent-latency-avg | The average put-if-absent execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| put-if-absent-latency-max | The maximum put-if-absent execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| get-latency-avg | The average get execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| get-latency-max | The maximum get execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| delete-latency-avg | The average delete execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| delete-latency-max | The maximum delete execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| put-all-latency-avg | The average put-all execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| put-all-latency-max | The maximum put-all execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| all-latency-avg | The average all operation execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| all-latency-max | The maximum all operation execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| range-latency-avg | The average range execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| range-latency-max | The maximum range execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| flush-latency-avg | The average flush execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| flush-latency-max | The maximum flush execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| restore-latency-avg | The average restore execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| restore-latency-max | The maximum restore execution time in ns. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| put-rate | The average put rate for this store. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| put-if-absent-rate | The average put-if-absent rate for this store. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| get-rate | The average get rate for this store. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| delete-rate | The average delete rate for this store. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| put-all-rate | The average put-all rate for this store. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| all-rate | The average all operation rate for this store. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| range-rate | The average range rate for this store. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| flush-rate | The average flush rate for this store. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| restore-rate | The average restore rate for this store. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| suppression-buffer-size-avg | The average total size, in bytes, of the buffered data over the sampling window. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),in-memory-suppression-id=([-.\w]+) |
| suppression-buffer-size-max | The maximum total size, in bytes, of the buffered data over the sampling window. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),in-memory-suppression-id=([-.\w]+) |
| suppression-buffer-count-avg | The average number of records buffered over the sampling window. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),in-memory-suppression-id=([-.\w]+) |
| suppression-buffer-count-max | The maximum number of records buffered over the sampling window. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),in-memory-suppression-id=([-.\w]+) |
| record-e2e-latency-avg | The average end-to-end latency of a record, measured by comparing the record timestamp with the system time when it has been fully processed by the node. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| record-e2e-latency-max | The maximum end-to-end latency of a record, measured by comparing the record timestamp with the system time when it has been fully processed by the node. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| record-e2e-latency-min | The minimum end-to-end latency of a record, measured by comparing the record timestamp with the system time when it has been fully processed by the node. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
RocksDB Metrics
RocksDB metrics are grouped into statistics-based metrics and properties-based metrics. The former are recorded from statistics that a RocksDB state store collects whereas the latter are recorded from properties that RocksDB exposes. Statistics collected by RocksDB provide cumulative measurements over time, e.g. bytes written to the state store. Properties exposed by RocksDB provide current measurements, e.g., the amount of memory currently used. Note that thestore-scope for built-in RocksDB state stores are currently the following:
rocksdb-state(for RocksDB backed key-value store)rocksdb-window-state(for RocksDB backed window store)rocksdb-session-state(for RocksDB backed session store)
debug because collecting
statistics in RocksDB
may have an impact on performance.
Statistics-based metrics are collected every minute from the RocksDB state stores.
If a state store consists of multiple RocksDB instances, as is the case for WindowStores and SessionStores,
each metric reports an aggregation over the RocksDB instances of the state store.
| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| bytes-written-rate | The average number of bytes written per second to the RocksDB state store. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| bytes-written-total | The total number of bytes written to the RocksDB state store. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| bytes-read-rate | The average number of bytes read per second from the RocksDB state store. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| bytes-read-total | The total number of bytes read from the RocksDB state store. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| memtable-bytes-flushed-rate | The average number of bytes flushed per second from the memtable to disk. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| memtable-bytes-flushed-total | The total number of bytes flushed from the memtable to disk. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| memtable-hit-ratio | The ratio of memtable hits relative to all lookups to the memtable. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| memtable-flush-time-avg | The average duration of memtable flushes to disc in ms. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| memtable-flush-time-min | The minimum duration of memtable flushes to disc in ms. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| memtable-flush-time-max | The maximum duration of memtable flushes to disc in ms. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| block-cache-data-hit-ratio | The ratio of block cache hits for data blocks relative to all lookups for data blocks to the block cache. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| block-cache-index-hit-ratio | The ratio of block cache hits for index blocks relative to all lookups for index blocks to the block cache. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| block-cache-filter-hit-ratio | The ratio of block cache hits for filter blocks relative to all lookups for filter blocks to the block cache. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| write-stall-duration-avg | The average duration of write stalls in ms. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| write-stall-duration-total | The total duration of write stalls in ms. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| bytes-read-compaction-rate | The average number of bytes read per second during compaction. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| bytes-written-compaction-rate | The average number of bytes written per second during compaction. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| compaction-time-avg | The average duration of disc compactions in ms. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| compaction-time-min | The minimum duration of disc compactions in ms. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| compaction-time-max | The maximum duration of disc compactions in ms. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| number-open-files | The number of current open files. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| number-file-errors-total | The total number of file errors occurred. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
info and are recorded when the
metrics are accessed.
If a state store consists of multiple RocksDB instances, as is the case for WindowStores and SessionStores,
each metric reports the sum over all the RocksDB instances of the state store, except for the block cache metrics
block-cache-*. The block cache metrics report the sum over all RocksDB instances if each instance uses its
own block cache, and they report the recorded value from only one instance if a single block cache is shared
among all instances.
| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| num-immutable-mem-table | The number of immutable memtables that have not yet been flushed. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| cur-size-active-mem-table | The approximate size of the active memtable in bytes. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| cur-size-all-mem-tables | The approximate size of active and unflushed immutable memtables in bytes. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| size-all-mem-tables | The approximate size of active, unflushed immutable, and pinned immutable memtables in bytes. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| num-entries-active-mem-table | The number of entries in the active memtable. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| num-entries-imm-mem-tables | The number of entries in the unflushed immutable memtables. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| num-deletes-active-mem-table | The number of delete entries in the active memtable. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| num-deletes-imm-mem-tables | The number of delete entries in the unflushed immutable memtables. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| mem-table-flush-pending | This metric reports 1 if a memtable flush is pending, otherwise it reports 0. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| num-running-flushes | The number of currently running flushes. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| compaction-pending | This metric reports 1 if at least one compaction is pending, otherwise it reports 0. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| num-running-compactions | The number of currently running compactions. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| estimate-pending-compaction-bytes | The estimated total number of bytes a compaction needs to rewrite on disk to get all levels down to under target size (only valid for level compaction). | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| total-sst-files-size | The total size in bytes of all SST files. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| live-sst-files-size | The total size in bytes of all SST files that belong to the latest LSM tree. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| num-live-versions | Number of live versions of the LSM tree. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| block-cache-capacity | The capacity of the block cache in bytes. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| block-cache-usage | The memory size of the entries residing in block cache in bytes. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| block-cache-pinned-usage | The memory size for the entries being pinned in the block cache in bytes. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| estimate-num-keys | The estimated number of keys in the active and unflushed immutable memtables and storage. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| estimate-table-readers-mem | The estimated memory in bytes used for reading SST tables, excluding memory used in block cache. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
| background-errors | The total number of background errors. | kafka.streams:type=stream-state-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),[store-scope]-id=([-.\w]+) |
Record Cache Metrics
All of the following metrics have a recording level ofdebug:
| Metric/Attribute name | Description | Mbean name |
|---|---|---|
| hit-ratio-avg | The average cache hit ratio defined as the ratio of cache read hits over the total cache read requests. | kafka.streams:type=stream-record-cache-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),record-cache-id=([-.\w]+) |
| hit-ratio-min | The minimum cache hit ratio. | kafka.streams:type=stream-record-cache-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),record-cache-id=([-.\w]+) |
| hit-ratio-max | The maximum cache hit ratio. | kafka.streams:type=stream-record-cache-metrics,thread-id=([-.\w]+),task-id=([-.\w]+),record-cache-id=([-.\w]+) |
Others
We recommend monitoring GC time and other stats and various server stats such as CPU utilization, I/O service time, etc. On the client side, we recommend monitoring the message/byte rate (global and per topic), request rate/size/time, and on the consumer side, max lag in messages among all partitions and min fetch request rate. For a consumer to keep up, max lag needs to be less than a threshold and min fetch rate needs to be larger than 0.6.9 ZooKeeper
Stable version
The current stable branch is 3.5. Kafka is regularly updated to include the latest release in the 3.5 series.ZooKeeper Deprecation
With the release of Apache Kafka 3.5, Zookeeper is now marked deprecated. Removal of ZooKeeper is planned in the next major release of Apache Kafka (version 4.0), which is scheduled to happen no sooner than April 2024. During the deprecation phase, ZooKeeper is still supported for metadata management of Kafka clusters, but it is not recommended for new deployments. There is a small subset of features that remain to be implemented in KRaft see current missing features for more information.
Migration
Migration of an existing ZooKeeper based Kafka cluster to KRaft is currently Preview and we expect it to be ready for production usage in version 3.6. Users are recommended to begin planning for migration to KRaft and also begin testing to provide any feedback. Refer to ZooKeeper to KRaft Migration for details on how to perform a live migration from ZooKeeper to KRaft and current limitations.
3.x and ZooKeeper Support
The final 3.x minor release, that supports ZooKeeper mode, will receive critical bug fixes and security fixes for 12 months after its release.
ZooKeeper and KRaft timeline
For details and updates on tentative timelines for ZooKeeper removal and planned KRaft feature releases, refer to KIP-833.
Operationalizing ZooKeeper
Operationally, we do the following for a healthy ZooKeeper installation:- Redundancy in the physical/hardware/network layout: try not to put them all in the same rack, decent (but don't go nuts) hardware, try to keep redundant power and network paths, etc. A typical ZooKeeper ensemble has 5 or 7 servers, which tolerates 2 and 3 servers down, respectively. If you have a small deployment, then using 3 servers is acceptable, but keep in mind that you'll only be able to tolerate 1 server down in this case.
- I/O segregation: if you do a lot of write type traffic you'll almost definitely want the transaction logs on a dedicated disk group. Writes to the transaction log are synchronous (but batched for performance), and consequently, concurrent writes can significantly affect performance. ZooKeeper snapshots can be one such a source of concurrent writes, and ideally should be written on a disk group separate from the transaction log. Snapshots are written to disk asynchronously, so it is typically ok to share with the operating system and message log files. You can configure a server to use a separate disk group with the dataLogDir parameter.
- Application segregation: Unless you really understand the application patterns of other apps that you want to install on the same box, it can be a good idea to run ZooKeeper in isolation (though this can be a balancing act with the capabilities of the hardware).
- Use care with virtualization: It can work, depending on your cluster layout and read/write patterns and SLAs, but the tiny overheads introduced by the virtualization layer can add up and throw off ZooKeeper, as it can be very time sensitive
- ZooKeeper configuration: It's java, make sure you give it 'enough' heap space (We usually run them with 3-5G, but that's mostly due to the data set size we have here). Unfortunately we don't have a good formula for it, but keep in mind that allowing for more ZooKeeper state means that snapshots can become large, and large snapshots affect recovery time. In fact, if the snapshot becomes too large (a few gigabytes), then you may need to increase the initLimit parameter to give enough time for servers to recover and join the ensemble.
- Monitoring: Both JMX and the 4 letter words (4lw) commands are very useful, they do overlap in some cases (and in those cases we prefer the 4 letter commands, they seem more predictable, or at the very least, they work better with the LI monitoring infrastructure)
- Don't overbuild the cluster: large clusters, especially in a write heavy usage pattern, means a lot of intracluster communication (quorums on the writes and subsequent cluster member updates), but don't underbuild it (and risk swamping the cluster). Having more servers adds to your read capacity.
6.10 KRaft
Configuration
Process Roles
In KRaft mode each Kafka server can be configured as a controller, a broker, or both using the process.roles property. This property can have the following values:
- If
process.rolesis set tobroker, the server acts as a broker. - If
process.rolesis set tocontroller, the server acts as a controller. - If
process.rolesis set tobroker,controller, the server acts as both a broker and a controller. - If
process.rolesis not set at all, it is assumed to be in ZooKeeper mode.
Kafka servers that act as both brokers and controllers are referred to as "combined" servers. Combined servers are simpler to operate for small use cases like a development environment. The key disadvantage is that the controller will be less isolated from the rest of the system. For example, it is not possible to roll or scale the controllers separately from the brokers in combined mode. Combined mode is not recommended in critical deployment environments.
Controllers
In KRaft mode, specific Kafka servers are selected to be controllers (unlike the ZooKeeper-based mode, where any server can become the Controller). The servers selected to be controllers will participate in the metadata quorum. Each controller is either an active or a hot standby for the current active controller.
A Kafka admin will typically select 3 or 5 servers for this role, depending on factors like cost and the number of concurrent failures your system should withstand without availability impact. A majority of the controllers must be alive in order to maintain availability. With 3 controllers, the cluster can tolerate 1 controller failure; with 5 controllers, the cluster can tolerate 2 controller failures.
All of the servers in a Kafka cluster discover the quorum voters using the controller.quorum.voters property. This identifies the quorum controller servers that should be used. All the controllers must be enumerated. Each controller is identified with their id, host and port information. For example:
controller.quorum.voters=id1@host1:port1,id2@host2:port2,id3@host3:port3If a Kafka cluster has 3 controllers named controller1, controller2 and controller3, then controller1 may have the following configuration:
process.roles=controller
node.id=1
listeners=CONTROLLER://controller1.example.com:9093
controller.quorum.voters=1@controller1.example.com:9093,2@controller2.example.com:9093,3@controller3.example.com:9093Every broker and controller must set the controller.quorum.voters property. The node ID supplied in the controller.quorum.voters property must match the corresponding id on the controller servers. For example, on controller1, node.id must be set to 1, and so forth. Each node ID must be unique across all the servers in a particular cluster. No two servers can have the same node ID regardless of their process.roles values.
Storage Tool
Thekafka-storage.sh random-uuid command can be used to generate a cluster ID for your new cluster. This cluster ID must be used when formatting each server in the cluster with the kafka-storage.sh format command.
This is different from how Kafka has operated in the past. Previously, Kafka would format blank storage directories automatically, and also generate a new cluster ID automatically. One reason for the change is that auto-formatting can sometimes obscure an error condition. This is particularly important for the metadata log maintained by the controller and broker servers. If a majority of the controllers were able to start with an empty log directory, a leader might be able to be elected with missing committed data.
Debugging
Metadata Quorum Tool
The kafka-metadata-quorum tool can be used to describe the runtime state of the cluster metadata partition. For example, the following command displays a summary of the metadata quorum:
> bin/kafka-metadata-quorum.sh --bootstrap-server broker_host:port describe --status
ClusterId: fMCL8kv1SWm87L_Md-I2hg
LeaderId: 3002
LeaderEpoch: 2
HighWatermark: 10
MaxFollowerLag: 0
MaxFollowerLagTimeMs: -1
CurrentVoters: [3000,3001,3002]
CurrentObservers: [0,1,2]Dump Log Tool
The kafka-dump-log tool can be used to debug the log segments and snapshots for the cluster metadata directory. The tool will scan the provided files and decode the metadata records. For example, this command decodes and prints the records in the first log segment:
> bin/kafka-dump-log.sh --cluster-metadata-decoder --files metadata_log_dir/__cluster_metadata-0/00000000000000000000.logThis command decodes and prints the records in the a cluster metadata snapshot:
> bin/kafka-dump-log.sh --cluster-metadata-decoder --files metadata_log_dir/__cluster_metadata-0/00000000000000000100-0000000001.checkpointMetadata Shell
The kafka-metadata-shell tool can be used to interactively inspect the state of the cluster metadata partition:
> bin/kafka-metadata-shell.sh --snapshot metadata_log_dir/__cluster_metadata-0/00000000000000000000.log
>> ls /
brokers local metadataQuorum topicIds topics
>> ls /topics
foo
>> cat /topics/foo/0/data
{
"partitionId" : 0,
"topicId" : "5zoAlv-xEh9xRANKXt1Lbg",
"replicas" : [ 1 ],
"isr" : [ 1 ],
"removingReplicas" : null,
"addingReplicas" : null,
"leader" : 1,
"leaderEpoch" : 0,
"partitionEpoch" : 0
}
>> exit
Deploying Considerations
- Kafka server's
process.roleshould be set to eitherbrokerorcontrollerbut not both. Combined mode can be used in development environments, but it should be avoided in critical deployment environments. - For redundancy, a Kafka cluster should use 3 controllers. More than 3 controllers is not recommended in critical environments. In the rare case of a partial network failure it is possible for the cluster metadata quorum to become unavailable. This limitation will be addressed in a future release of Kafka.
- The Kafka controllers store all the metadata for the cluster in memory and on disk. We believe that for a typical Kafka cluster 5GB of main memory and 5GB of disk space on the metadata log director is sufficient.
Missing Features
The following features are not fully implemented in KRaft mode:
- Supporting JBOD configurations with multiple storage directories
- Modifying certain dynamic configurations on the standalone KRaft controller
- Delegation tokens
ZooKeeper to KRaft Migration
ZooKeeper to KRaft migration is considered an Early Access feature and is not recommended for production clusters.
The following features are not yet supported for ZK to KRaft migrations:
- Downgrading to ZooKeeper mode during or after the migration
- Other features not yet supported in KRaft
Please report issues with ZooKeeper to KRaft migration using the project JIRA and the "kraft" component.
Terminology
We use the term "migration" here to refer to the process of changing a Kafka cluster's metadata system from ZooKeeper to KRaft and migrating existing metadata. An "upgrade" refers to installing a newer version of Kafka. It is not recommended to upgrade the software at the same time as performing a metadata migration.
We also use the term "ZK mode" to refer to Kafka brokers which are using ZooKeeper as their metadata system. "KRaft mode" refers Kafka brokers which are using a KRaft controller quorum as their metadata system.
Preparing for migration
Before beginning the migration, the Kafka brokers must be upgraded to software version 3.5.0 and have the "inter.broker.protocol.version" configuration set to "3.5". See Upgrading to 3.5.0 for upgrade instructions.
It is recommended to enable TRACE level logging for the migration components while the migration is active. This can be done by adding the following log4j configuration to each KRaft controller's "log4j.properties" file.
log4j.logger.org.apache.kafka.metadata.migration=TRACE
It is generally useful to enable DEBUG logging on the KRaft controllers and the ZK brokers during the migration.
Provisioning the KRaft controller quorum
Two things are needed before the migration can begin. First, the brokers must be configured to support the migration and second, a KRaft controller quorum must be deployed. The KRaft controllers should be provisioned with the same cluster ID as the existing Kafka cluster. This can be found by examining one of the "meta.properties" files in the data directories of the brokers, or by running the following command.
./bin/zookeeper-shell.sh localhost:2181 get /cluster/id
The KRaft controller quorum should also be provisioned with the latest metadata.version.
This is done automatically when you format the node with the kafka-storage.sh tool.
For further instructions on KRaft deployment, please refer to the above documentation.
In addition to the standard KRaft configuration, the KRaft controllers will need to enable support for the migration as well as provide ZooKeeper connection configuration.
Here is a sample config for a KRaft controller that is ready for migration:
# Sample KRaft cluster controller.properties listening on 9093 process.roles=controller node.id=3000 controller.quorum.voters=3000@localhost:9093 controller.listener.names=CONTROLLER listeners=CONTROLLER://:9093 # Enable the migration zookeeper.metadata.migration.enable=true # ZooKeeper client configuration zookeeper.connect=localhost:2181 # Other configs ...
Note: The KRaft cluster node.id values must be different from any existing ZK broker broker.id.
In KRaft-mode, the brokers and controllers share the same Node ID namespace.
Enabling the migration on the brokers
Once the KRaft controller quorum has been started, the brokers will need to be reconfigured and restarted. Brokers may be restarted in a rolling fashion to avoid impacting cluster availability. Each broker requires the following configuration to communicate with the KRaft controllers and to enable the migration.
- controller.quorum.voters
- controller.listener.names
- The controller.listener.name should also be added to listener.security.property.map
- zookeeper.metadata.migration.enable
Here is a sample config for a broker that is ready for migration:
# Sample ZK broker server.properties listening on 9092 broker.id=0 listeners=PLAINTEXT://:9092 advertised.listeners=PLAINTEXT://localhost:9092 listener.security.protocol.map=PLAINTEXT:PLAINTEXT,CONTROLLER:PLAINTEXT # Set the IBP inter.broker.protocol.version=3.5 # Enable the migration zookeeper.metadata.migration.enable=true # ZooKeeper client configuration zookeeper.connect=localhost:2181 # KRaft controller quorum configuration controller.quorum.voters=3000@localhost:9093 controller.listener.names=CONTROLLER
Note: Once the final ZK broker has been restarted with the necessary configuration, the migration will automatically begin. When the migration is complete, an INFO level log can be observed on the active controller:
Completed migration of metadata from Zookeeper to KRaft
Migrating brokers to KRaft
Once the KRaft controller completes the metadata migration, the brokers will still be running in ZK mode. While the KRaft controller is in migration mode, it will continue sending controller RPCs to the ZK mode brokers. This includes RPCs like UpdateMetadata and LeaderAndIsr.
To migrate the brokers to KRaft, they simply need to be reconfigured as KRaft brokers and restarted. Using the above
broker configuration as an example, we would replace the broker.id with node.id and add
process.roles=broker. It is important that the broker maintain the same Broker/Node ID when it is restarted.
The zookeeper configurations should be removed at this point.
# Sample KRaft broker server.properties listening on 9092 process.roles=broker node.id=0 listeners=PLAINTEXT://:9092 advertised.listeners=PLAINTEXT://localhost:9092 listener.security.protocol.map=PLAINTEXT:PLAINTEXT,CONTROLLER:PLAINTEXT # Don't set the IBP, KRaft uses "metadata.version" feature flag # inter.broker.protocol.version=3.5 # Remove the migration enabled flag # zookeeper.metadata.migration.enable=true # Remove ZooKeeper client configuration # zookeeper.connect=localhost:2181 # Keep the KRaft controller quorum configuration controller.quorum.voters=3000@localhost:9093 controller.listener.names=CONTROLLER
Each broker is restarted with a KRaft configuration until the entire cluster is running in KRaft mode.
Finalizing the migration
Once all brokers have been restarted in KRaft mode, the last step to finalize the migration is to take the KRaft controllers out of migration mode. This is done by removing the "zookeeper.metadata.migration.enable" property from each of their configs and restarting them one at a time.
# Sample KRaft cluster controller.properties listening on 9093 process.roles=controller node.id=3000 controller.quorum.voters=3000@localhost:9093 controller.listener.names=CONTROLLER listeners=CONTROLLER://:9093 # Disable the migration # zookeeper.metadata.migration.enable=true # Remove ZooKeeper client configuration # zookeeper.connect=localhost:2181 # Other configs ...
6.11 Tiered Storage
Tiered Storage Overview
Kafka data is mostly consumed in a streaming fashion using tail reads. Tail reads leverage OS's page cache to serve the data instead of disk reads. Older data is typically read from the disk for backfill or failure recovery purposes and is infrequent.
In the tiered storage approach, Kafka cluster is configured with two tiers of storage - local and remote. The local tier is the same as the current Kafka that uses the local disks on the Kafka brokers to store the log segments. The new remote tier uses external storage systems, such as HDFS or S3, to store the completed log segments. Please check KIP-405 for more information.
Note: Tiered storage is considered as an early access feature, and is not recommended for use in production environments
Configuration
Broker Configurations
By default, Kafka server will not enable tiered storage feature. remote.log.storage.system.enable
is the property to control whether to enable tiered storage functionality in a broker or not. Setting it to "true" enables this feature.
RemoteStorageManager is an interface to provide the lifecycle of remote log segments and indexes. Kafka server
doesn't provide out-of-the-box implementation of RemoteStorageManager. Configuring remote.log.storage.manager.class.name
and remote.log.storage.manager.class.path to specify the implementation of RemoteStorageManager.
RemoteLogMetadataManager is an interface to provide the lifecycle of metadata about remote log segments with strongly consistent semantics.
By default, Kafka provides an implementation with storage as an internal topic. This implementation can be changed by configuring
remote.log.metadata.manager.class.name and remote.log.metadata.manager.class.path.
When adopting the default kafka internal topic based implementation, remote.log.metadata.manager.listener.name
is a mandatory property to specify which listener the clients created by the default RemoteLogMetadataManager implementation.
Topic Configurations
After correctly configuring broker side configurations for tiered storage feature, there are still configurations in topic level needed to be set.
remote.storage.enable is the switch to determine if a topic wants to use tiered storage or not. By default it is set to false.
After enabling remote.storage.enable property, the next thing to consider is the log retention.
When tiered storage is enabled for a topic, there are 2 additional log retention configurations to set:
local.retention.msretention.mslocal.retention.bytesretention.bytes
local are to specify the time/size the "local" log file can accept before moving to remote storage, and then get deleted.
If unset, The value in retention.ms and retention.bytes will be used.
Quick Start Example
Apache Kafka doesn't provide an out-of-the-box RemoteStorageManager implementation. To have a preview of the tiered storage feature, the LocalTieredStorage implemented for integration test can be used, which will create a temporary directory in local storage to simulate the remote storage.
To adopt the `LocalTieredStorage`, the test library needs to be built locally
# please checkout to the specific version tag you're using before building it # ex: `git checkout 3.6.2` ./gradlew clean :storage:testJar
After build successfully, there should be a `kafka-storage-x.x.x-test.jar` file under `storage/build/libs`. Next, setting configurations in the broker side to enable tiered storage feature.
# Sample Zookeeper/Kraft broker server.properties listening on PLAINTEXT://:9092 remote.log.storage.system.enable=true # Setting the listener for the clients in RemoteLogMetadataManager to talk to the brokers. remote.log.metadata.manager.listener.name=PLAINTEXT # Please provide the implementation info for remoteStorageManager. # This is the mandatory configuration for tiered storage. # Here, we use the `LocalTieredStorage` built above. remote.log.storage.manager.class.name=org.apache.kafka.server.log.remote.storage.LocalTieredStorage remote.log.storage.manager.class.path=/PATH/TO/kafka-storage-x.x.x-test.jar # These 2 prefix are default values, but customizable remote.log.storage.manager.impl.prefix=rsm.config. remote.log.metadata.manager.impl.prefix=rlmm.config. # Configure the directory used for `LocalTieredStorage` # Note, please make sure the brokers need to have access to this directory rsm.config.dir=/tmp/kafka-remote-storage # This needs to be changed if number of brokers in the cluster is more than 1 rlmm.config.remote.log.metadata.topic.replication.factor=1 # Try to speed up the log retention check interval for testing log.retention.check.interval.ms=1000
Following quick start guide to start up the kafka environment. Then, create a topic with tiered storage enabled with configs:
# remote.storage.enable=true -> enables tiered storage on the topic # local.retention.ms=1000 -> The number of milliseconds to keep the local log segment before it gets deleted. Note that a local log segment is eligible for deletion only after it gets uploaded to remote. # retention.ms=3600000 -> when segments exceed this time, the segments in remote storage will be deleted # segment.bytes=1048576 -> for test only, to speed up the log segment rolling interval # file.delete.delay.ms=10000 -> for test only, to speed up the local-log segment file delete delay bin/kafka-topics.sh --create --topic tieredTopic --bootstrap-server localhost:9092 \ --config remote.storage.enable=true --config local.retention.ms=1000 --config retention.ms=3600000 \ --config segment.bytes=1048576 --config file.delete.delay.ms=1000
Try to send messages to the `tieredTopic` topic to roll the log segment:
bin/kafka-producer-perf-test.sh --topic tieredTopic --num-records 1200 --record-size 1024 --throughput -1 --producer-props bootstrap.servers=localhost:9092
Then, after the active segment is rolled, the old segment should be moved to the remote storage and get deleted. This can be verified by checking the remote log directory configured above. For example:
> ls /tmp/kafka-remote-storage/kafka-tiered-storage/tieredTopic-0-jF8s79t9SrG_PNqlwv7bAA 00000000000000000000-knnxbs3FSRyKdPcSAOQC-w.index 00000000000000000000-knnxbs3FSRyKdPcSAOQC-w.snapshot 00000000000000000000-knnxbs3FSRyKdPcSAOQC-w.leader_epoch_checkpoint 00000000000000000000-knnxbs3FSRyKdPcSAOQC-w.timeindex 00000000000000000000-knnxbs3FSRyKdPcSAOQC-w.log
Lastly, we can try to consume some data from the beginning and print offset number, to make sure it will successfully fetch offset 0 from the remote storage.
bin/kafka-console-consumer.sh --topic tieredTopic --from-beginning --max-messages 1 --bootstrap-server localhost:9092 --property print.offset=true
Please note, if you want to disable tiered storage at the cluster level, you should delete the tiered storage enabled topics explicitly. Attempting to disable tiered storage at the cluster level without deleting the topics using tiered storage will result in an exception during startup.
bin/kafka-topics.sh --delete --topic tieredTopic --bootstrap-server localhost:9092
After topics are deleted, you're safe to set remote.log.storage.system.enable=false in the broker configuration.
Limitations
While the early access release of Tiered Storage offers the opportunity to try out this new feature, it is important to be aware of the following limitations:
- No support for clusters with multiple log directories (i.e. JBOD feature)
- No support for compacted topics
- Cannot disable tiered storage at the topic level
- Deleting tiered storage enabled topics is required before disabling tiered storage at the broker level
- Admin actions related to tiered storage feature are only supported on clients from version 3.0 onwards
For more information, please check Tiered Storage Early Access Release Note.
7. Security
8. Kafka Connect
9. Kafka Streams
Kafka Streams is a client library for processing and analyzing data stored in Kafka. It builds upon important stream processing concepts such as properly distinguishing between event time and processing time, windowing support, exactly-once processing semantics and simple yet efficient management of application state.
Kafka Streams has a low barrier to entry: You can quickly write and run a small-scale proof-of-concept on a single machine; and you only need to run additional instances of your application on multiple machines to scale up to high-volume production workloads. Kafka Streams transparently handles the load balancing of multiple instances of the same application by leveraging Kafka's parallelism model.
Learn More about Kafka Streams read this Section.
