The new group coordinator prints the following line at fixed interval
even if no groups were deleted:
```
Generated 0 tombstone records while cleaning up group metadata in 0 milliseconds. (org.apache.kafka.coordinator.group.GroupCoordinatorShard)
```
The time component has some value in its own but it may be better to not
print if when there are not records in order to reduce the spam in the
logs.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>
Given that now we support Java 17 on our brokers, this PR replace the
use of :
- `Collections.singletonList()` and `Collections.emptyList()` with
`List.of()`
- `Collections.singletonMap()` and `Collections.emptyMap()` with
`Map.of()`
- `Collections.singleton()` and `Collections.emptySet()` with `Set.of()`
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>
This patch filters out the topic describe unauthorized topics from the
ConsumerGroupHeartbeat and ConsumerGroupDescribe response.
In ConsumerGroupHeartbeat,
- if the request has `subscribedTopicNames` set, we directly check the
authz in `KafkaApis` and return a topic auth failure in the response if
any of the topics is denied.
- Otherwise, we check the authz only if a regex refresh is triggered and
we do it based on the acl of the consumer that triggered the refresh. If
any of the topic is denied, we filter it out from the resolved
subscription.
In ConsumerGroupDescribe, we check the authz of the coordinator
response. If any of the topic in the group is denied, we remove the
described info and add a topic auth failure to the described group.
(similar to the group auth failure)
Reviewers: David Jacot <djacot@confluent.io>, Lianet Magrans
<lmagrans@confluent.io>, Rajini Sivaram <rajinisivaram@googlemail.com>,
Chia-Ping Tsai <chia7712@gmail.com>, TaiJuWu <tjwu1217@gmail.com>,
TengYao Chi <kitingiao@gmail.com>
This change implements the basic RPC handling StreamsGroupHeartbeat and
StreamsGroupDescribe. This includes:
- Adding an option to enable streams groups on the broker
- Passing describe and heartbeats to the right shard of the group
coordinator
- The handler inside the GroupMetadatManager for StreamsGroupDescribe is
fairly trivial, and is included directly in this PR.
- The handler for StreamsGroupHeartbeat is complex and not included in
this PR yet. Instead, a UnsupportedOperationException is thrown.
However, the interface is already defined: The result of a
streamsGroupHeartbeat is a response, together with a list of internal
topics to be created.
The heartbeat implementation inside the `GroupMetadataManager`, which
actually implements the assignment / reconciliation logic, will come in
a follow-up PR. Also, automatic creation of internal topics will be
created in a follow-up PR.
Reviewers: Bill Bejeck <bill@confluent.io>
### About
The current `SimpleAssignor` in AK assigned all subscribed topic
partitions to all the share group members. This does not match the
description given in
[KIP-932](https://cwiki.apache.org/confluence/pages/viewpage.action?pageId=255070434#KIP932:QueuesforKafka-TheSimpleAssignor).
Here are the rules as mentioned in the KIP by which the assignment
should happen. We have changed the step 3 implementation here due to the
reasons
[described](https://github.com/apache/kafka/pull/18864#issuecomment-2659266502)
-
1. The assignor hashes the member IDs of the members and maps the
partitions assigned to the members based on the hash. This gives
approximately even balance.
2. If any partitions were not assigned any members by (1) and do not
have members already assigned in the current assignment, members are
assigned round-robin until each partition has at least one member
assigned to it.
3. We combine the current and new assignment. (Original rule - If any
partitions were assigned members by (1) and also have members in the
current assignment assigned by (2), the members assigned by (2) are
removed.)
### Tests
The added code has been verified with unit tests and the already present
integration tests.
Reviewers: Andrew Schofield <aschofield@confluent.io>, Apoorv Mittal <apoorvmittal10@gmail.com>, TaiJuWu <tjwu1217@gmail.com>
The last commit in this class mistakenly described the functions to be
for Streams Groups. Just a minor update.
Reviewers: Andrew Schofield <aschofield@confluent.io>, Apoorv Mittal <apoorvmittal10@gmail.com>, Sushant Mahajan <smahajan@confluent.io>
I was looking into GroupCoordinatorConfigs to review configurations that
we will ship with Apache Kafka 4.0. I found out that it was pretty
disorganised. This patch cleans up the format and re-groups the
configurations which are related.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>
* In this PR, we have added GC side impl to call the delete state share
coord RPC using the persister.
* We will be using the existing `GroupCoordinatorService.deleteGroups`.
The logic will be modified as follows:
* After sanitization, we will call a new
`runtime.scheduleWriteOperation` (not read for consistency) with
callback `GroupCoordinatorShard.sharePartitions`. This will return a Map
of share partitions of the groups which are of SHARE type. We need to
pass all groups as WE CANNOT DETERMINE the type of the group in the
service class.
* Then using the map we will create requests which could be passed to
the persister and make the appropriate calls.
* Once this future completes, we will continue with the existing flow of
group deletion.
* If the group under inspection is not share group - the read callback
should return an empty map.
* Tests have been added wherever applicable.
Reviewers: David Jacot <djacot@confluent.io>, Andrew Schofield <aschofield@confluent.io>
* Due to recent changes in the way group count metrics are initialized
and updated, the current share group count code has become obsolete as
well as non-functional.
* The update method for the share group count which should be called
from `ShareGroup` cannot be called either. This is because the
constructor has been changed to NOT accept the
`GroupCoordinatorShardMetrics` ref.
* In this PR, we remedy the situation by bringing share group count code
at par with consumer and streams groups.
* Additionally the metric name for share groups with group state
attributes was not aligned with streams and consumer groups as mentioned
in https://github.com/apache/kafka/pull/17011#discussion_r1960309578.
This PR aligns them too.
Reviewers: Andrew Schofield <aschofield@confluent.io>
Cleanup code to avoid rawtype, and add suppressions where necessary.
Change the build to fail on rawtype warning.
Reviewers: Apoorv Mittal <apoorvmittal10@gmail.com>, Andrew Schofield <aschofield@confluent.io>
This patch changes the default value of `group.coordinator.threads` to `4` and sets it priority to `HIGH`. This change makes it consistent with how we handle `num.network.threads` and `num.io.threads`. The patch also tweaks the upgrade notes.
Reviewers: Ismael Juma <ismael@juma.me.uk>
Adds streams group to the GroupMetadataManager, and implements loading
the records from the offset topic into state. The state also contains
two timers (rebalance timeout and session timeout) that are started
after the group coordinator has been loaded.
Reviewers: Bruno Cadonna <bruno@confluent.io>, Bill Bejeck <bill@confluent.io>
Implements a memory model for representing streams groups in the group coordinator, as well as group count and rebalance metrics.
Reviewers: Bill Bejeck <bill@confluent.io>, Bruno Cadonna <bruno@confluent.io>
At the moment, we require specifying builtin server side assignors by their full class name. This is not convenient and also exposed their full class name as part of our public API. This patch changes it to accept specifying builtin server side assignor by their short name (uniform or range) while continuing to accept full class name for customer assignors.
Reviewers: Jeff Kim <jeff.kim@confluent.io>
A class to build a new target assignment based on the provided parameters. As a result, it yields the records that must be persisted to the log and the new member assignments as a map.
Compared to the feature branch, I extended the unit tests (testing also standby and warm-up task logic) and adopted simplifications due to the TasksTuple class.
Reviewers: Bruno Cadonna <cadonna@apache.org>, Bill Bejeck <bbejeck@apache.org>
During testing we discovered that the empty group count is not updated in group conversion, but when the new group is transition to other state, the empty group count is decremented. This could result in negative empty group count.
We can have a new consumer group count implementation that follows the pattern we did for the classic group count. The timeout task periodically refreshes the metrics based on the current groups soft state.
Reviewers: Jeff Kim <jeff.kim@confluent.io>
CoordinatorRecordSerde does not validate the version of the value to check whether the version is supported by the current version of the software. This is problematic if a future and unsupported version of the record is read by an older version of the software because it would misinterpret the bytes. Hence CoordinatorRecordSerde must throw an error if the version is unknown. This is also consistent with the handling in the old coordinator.
Reviewers: Jeff Kim <jeff.kim@confluent.io>
The full class name of the assignors if part of our public api. Hence, we should ensure that they are not changed by mistake. This patch adds a unit test verifying them.
Reviewers: Sean Quah <squah@confluent.io>, Jeff Kim <jeff.kim@confluent.io>
A class with helper methods to create records stored in the __consumer_offsets topic.
Compared to the feature branch, I added unit tests (most functions were not tested) and adopted the new interface for constructing coordinator records introduced by David.
Reviewers: Bruno Cadonna <cadonna@apache.org>
Implements the current assignment builder, analogous to the current assignment builder of consumer groups. The main difference is the underlying assigned resource, and slightly different logic around process IDs: We make sure to move a task only to a new client, once the task is not owned anymore by any client with the same process ID (sharing the same state directory) - in any role (active, standby or warm-up).
Compared to the feature branch, the main difference is that I refactored the separate treatment of active, standby and warm-up tasks into a compound datatype called TaskTuple (which is used in place of the more specific Assignment class). This also has effects on StreamsGroupMember.
Reviewers: Bruno Cadonna <cadonna@apache.org>, Bill Bejeck <bbejeck@apache.org>
This patch does a few things:
1) Replace ApiMessageAndVersion by ApiMessage in CoordinatorRecord for the key
2) Leverage the fact that ApiMessage exposes the apiKey. Hence we don't need to specify the key anymore.
Reviewers: Andrew Schofield <aschofield@confluent.io>
The internal topic manager takes a requested topology and returns a configured topology, as well as any internal topics that need to be created.
Shares with the client-side "InternalTopicManager" the name only.
Reviewers: Bruno Cadonna <cadonna@apache.org>
This patch updates the transaction coordinator record to use the new coordinator record definition.
Reviewers: Andrew Schofield <aschofield@confluent.io>
Apache Kafka 4.0 will only support KRaft and 3.0-IV1 is the minimum version supported by KRaft. So, we can assume that Apache Kafka 4.0 will only communicate with brokers that are 3.0-IV1 or newer.
Note that KRaft was only marked as production-ready in 3.3, so we could go further and set the baseline to 3.3. I think we should have that discussion, but it made sense to start with the non controversial parts.
Reviewers: Jun Rao <junrao@gmail.com>, Chia-Ping Tsai <chia7712@gmail.com>, David Jacot <david.jacot@gmail.com>
A simplified port of "CopartitionedTopicsEnforcer" from the client-side to the group coordinator.
This class is responsible for enforcing the number of partitions in copartitioned topics. For each copartition group, it checks whether the number of partitions for all topics in the group is the same, and enforces the right number of partitions for repartition topics whose number of partitions is not enforced by the topology.
Compared to the client-side version, the implementation uses immutable data structures, and returns the computed number of partitions instead of modifying mutable data structures and calling the admin client.
Reviewers: Bruno Cadonna <cadonna@apache.org>
This patch updates the GroupCoordinatorSerde and the ShareGroupCoordinatorSerde to leverage the CoordinatorRecordType to deserialize records. With this, newly added record are automatically picked up. In other words, the serdes work with all defined records without doing anything.
Reviewers: Andrew Schofield <aschofield@confluent.io>
A simplified port of "RepartitionTopics" from the client-side to the group coordinator.
Compared to the client-side version, the implementation uses immutable data structures, and returns the computed number of partitions instead of modifying mutable data structures and calling the admin client.
Reviewers: Bruno Cadonna <cadonna@apache.org>
Adds a class that represent the topology of a Streams group sent by a Streams client in the Streams group heartbeat during initialization to the group coordinator.
This topology representation is used together with the partition metadata on the broker to create a configured topology.
Reviewer: Lucas Brutschy <lbrutschy@confluent.io>
* KAFKA-18321: Add StreamsGroupMember, MemberState and Assignment classes
This commit adds the classes to represent a Streams group member in the
consumer coordinator.
Reviewers: Bill Bejeck <bill@confluent.io>, Lucas Brutschy <lbrutschy@confluent.io>
A simplified port of "ChangelogTopics" from the client-side to the group coordinator
Compared to the client-side version, the implementation uses immutable data structures, and returns the computed number of partitions instead of modifying mutable data structures and calling the admin client.
Reviewers: Bruno Cadonna <cadonna@apache.org>
Clients in the Streams Rebalance Protocol send an "unconfigured" representation of the topology to the broker. That is, the number of input topics and (some) internal topics is not fixed, regular expressions are not resolved. The broker takes this description of the topology and, together with the current state of the topics on the broker, derives a ConfiguredTopology. The configured topology is what is being returned from StreamsGroupDescribe, and has all number of partitions defined, and regular expressions resolved. The configured topology also contains missing internal topics that need to be created, and potentially configuration errors, such as missing source topics.
In this change, we add the internal data structures for representing the configured topology. They differ in some details from the data structures used in the RPCs. Most importantly, they can be evolved independently of the public interface.
Reviewers: Bruno Cadonna <cadonna@apache.org>
Introduces interfaces for defining task assignors. Task assignors are pure functions, mapping the state of the group and a topology to a target assignment. We include a mock assignor, which we will be able to use when testing / benchmarking without the complexities of the sticky task assignor and the high-availability task assignor. We may remove the mock assignor in before the streams rebalance protocol goes GA.
The consumer groups introduce these interfaces to establish a clear separation between the group coordinator code and the pluggable assignors, which may live outside the group coordinator code. We have removed pluggable assignors in KIP-1071, but I think it still makes sense to keep these interfaces for having a clean interface for people to code against. This will pay off, if we plan on making the task assignors pluggable later.
Reviewers: Bill Bejeck <bbejeck@gmail.com>, David Jacot <djacot@confluent.io>
Included in this change:
1. Remove deprecated protocol api versions from json files.
3. Remove fields that are no longer used from json files (affects ListOffsets, OffsetCommit, DescribeConfigs).
4. Remove record down-conversion support from KafkaApis.
5. No longer return `Errors.UNSUPPORTED_COMPRESSION_TYPE` on the fetch path[1].
6. Deprecate `TopicConfig. MESSAGE_DOWNCONVERSION_ENABLE_CONFIG` and made the relevant
configs (`message.downconversion.enable` and `log.message.downcoversion.enable`) no-ops since
down-conversion is no longer supported. It was an oversight not to deprecate this via KIP-724.
7. Fix `shouldRetainsBufferReference` to handle null request schemas for a given version.
8. Simplify producer logic since it only supports the v2 record format now.
9. Fix tests so they don't exercise protocol api versions that have been removed.
10. Add upgrade note.
Testing:
1. System tests have a lot of failures, but those tests fail for trunk too and I didn't see any issues specific to this change - it's hard to be sure given the number of failing tests, but let's not block on that given the other testing that has been done (see below).
3. Java producers and consumers with version 0.9-0.10.1 don't have api versions support and hence they fail in an ungraceful manner: the broker disconnects and the clients reconnect until the relevant timeout is triggered.
4. Same thing seems to happen for the console producer 0.10.2 although it's unclear why since api versions should be supported. I will look into this separately, it's unlikely to be related to this PR.
5. Console consumer 0.10.2 fails with the expected error and a reasonable message[2].
6. Console producer and consumer 0.11.0 works fine, newer versions should naturally also work fine.
7. kcat 1.5.0 (based on librdkafka 1.1.0) produce and consume fail with a reasonable message[3][4].
8. kcat 1.6.0-1.7.0 (based on librdkafka 1.5.0 and 1.7.0 respectively) consume fails with a reasonable message[5].
9. kcat 1.6.0-1.7.0 produce works fine.
10. kcat 1.7.1 (based on librdkafka 1.8.2) works fine for consumer and produce.
11. confluent-go-client (librdkafka based) 1.8.2 works fine for consumer and produce.
12. I will test more clients, but I don't think we need to block the PR on that.
Note that this also completes part of KIP-724: produce v2 and lower as well as fetch v3 and lower are no longer supported.
Future PRs will remove conditional code that is no longer needed (some of that has been done in KafkaApis,
but only what was required due to the schema changes). We can probably do that in master only as it does
not change behavior.
Note that I did not touch `ignorable` fields even though some of them could have been
changed. The reasoning is that this could result in incompatible changes for clients
that use new protocol versions without setting such fields _if_ we don't manually
validate their presence. I will file a JIRA ticket to look into this carefully for each
case (i.e. if we do validate their presence for the appropriate versions, we can
set them to ignorable=false in the json file).
[1] We would return this error if a fetch < v10 was used and the compression topic config was set
to zstd, but we would not do the same for the case where zstd was compressed at the producer
level (the most common case). Since there is no efficient way to do the check for the common
case, I made it consistent for both by having no checks.
[2] ```org.apache.kafka.common.errors.UnsupportedVersionException: The broker is too new to support JOIN_GROUP version 1```
[3]```METADATA|rdkafka#producer-1| [thrd:main]: localhost:9092/bootstrap: Metadata request failed: connected: Local: Required feature not supported by broker (0ms): Permanent```
[4]```METADATA|rdkafka#consumer-1| [thrd:main]: localhost:9092/bootstrap: Metadata request failed: connected: Local: Required feature not supported by broker (0ms): Permanent```
[5] `ERROR: Topic test-topic [0] error: Failed to query logical offset END: Local: Required feature not supported by broker`
Reviewers: David Arthur <mumrah@gmail.com>
This patch is the first one in a series to improve how coordinator records are managed. It focuses on making coordinator records first class citizen in the generator.
* Introduce `coordinator-key` and `coordinator-value` in the schema;
* Introduce `apiKey` for those. This is done to avoid relying on the version to determine the type.
* It also allows the generator to enforce some rules: the key cannot use flexible versions, the key must have a single version `0`, there must be a key and a value for a given api key, etc.
* It generates an enum with all the coordinator record types. This is pretty handy in the code.
The patch also updates the group coordinators to use those.
Reviewers: Jeff Kim <jeff.kim@confluent.io>, Andrew Schofield <aschofield@confluent.io>
KIP-1071 specifies records that the group coordinator needs to store
into the consumer offset topic to persist the state of a Streams
group. This records are specified in json files from which the actual
classes for the records are generated.
This commit adds the records needed by the group coordinator to store
the state of a Streams group.
Reviewer: Lucas Brutschy <lbrutschy@confluent.io>
KIP-848 will be release as GA in Apache Kafka 4.0. Hence we need to mark all the related public apis as stable.
Reviewers: Jeff Kim <jeff.kim@confluent.io>
Kafka consumer supports auto.offset.reset config option, which is used when there is no initial offset in Kafka (or) if the current offset does not exist any more on the server. This config currently supports earliest/latest/none options. Currently consumer resets might force applications to reprocess large amounts of data from earlier offsets. With infinite storage, its beneficial to have a duration based offset reset strategy. This will allow applications to consume/initialise from a fixed duration when there is no initial offset in Kafka.
As part of KIP-932, we are adding support for share consumer groups. Share consumer groups supports dynamic group configuration property share.auto.offset.reset. This is used to set the initial Share-Partition Start Offset (SPSO) based on the share.auto.offset.reset configuration. Currently share.auto.offset.reset supports earliest and latest options to automatically reset the offset
Similar to the Kafka Consumer, we will add support for by_duration: config value for share.auto.offset.reset.
Reviewers: Andrew Schofield <aschofield@confluent.io>, Apoorv Mittal <apoorvmittal10@gmail.com>
David Jacot
Sanskar Jhajharia
Dongnuo Lyu
Lucas Brutschy
Abhinav Dixit
Sushant Mahajan
Matthias J. Sax
Parker Chang
Andrew Schofield
Alieh Saeedi
Sean Quah
Lianet Magrans
Ismael Juma
Bruno Cadonna
Jeff Kim
David Jacot
Peter Lee