This PR reduces `maxInterval` for `resendExponentialBackoff` in
`BrokerLifecycleManager` class from `broker.session.timeout.ms` to half
of its value. Setting `maxInterval` to `broker.session.timeout.ms`
caused brokers to be fenced if a resend attempt occurred near the
timeout threshold, leading to unnecessary broker fencing.
Reviewers: Colin P. McCabe <cmccabe@apache.org>
Implements auto-topic creation when handling the streams group
heartbeat.
Inside KafkaApis, the handler for streamsGroupHeartbeat uses the result
of the streams group heartbeat inside the group coordinator to attempt
to create all missing internal topics using AutoTopicCreationManager.
CREATE TOPIC ACLs are checked. The unit tests class
AutoTopicCreationManagerTest is brought back (it was recently deleted
during a ZK removal PR), but testing only the kraft-related
functionality.
Reviewers: Bruno Cadonna <bruno@confluent.io>
### Committer Checklist (excluded from commit message)
- [ ] Verify design and implementation
- [ ] Verify test coverage and CI build status
- [ ] Verify documentation (including upgrade notes)
Remove kafka.cluster.Broker and BrokerEndPointNotAvailableException as they were used by zk path.
Reviewers: TengYao Chi <kitingiao@gmail.com>, Ken Huang <s7133700@gmail.com>, Chia-Ping Tsai <chia7712@gmail.com>
The PR implements the SharePartitionMetrics as defined in KIP-1103, with
one change. The metric `FetchLockRatio` is defined as `Meter` in KIP but
is implemented as `HIstogram`. There was a discussion about same on
KIP-1103 discussion where we thought that `FetchLockRatio` is
pre-aggregated but while implemeting the rate from `Meter` can go above
100 as `Meter` defines rate per time period. Hence it makes more sense
to implement metric `FetchLockRatio` as `Histogram`.
Reviewers: Andrew Schofield <aschofield@confluent.io>
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>
For the KRaft implementation there is a race between the network thread,
which read bytes in the log segments, and the KRaft driver thread, which
truncates the log and appends records to the log. This race can cause
the network thread to send corrupted records or inconsistent records.
The corrupted records case is handle by catching and logging the
CorruptRecordException. The inconsistent records case is handle by only
appending record batches who's partition leader epoch is less than or
equal to the fetching replica's epoch and the epoch didn't change
between the request and response.
For the ISR implementation there is also a race between the network
thread and the replica fetcher thread, which truncates the log and
appends records to the log. This race can cause the network thread send
corrupted records or inconsistent records. The replica fetcher thread
already handles the corrupted record case. The inconsistent records case
is handle by only appending record batches who's partition leader epoch
is less than or equal to the leader epoch in the FETCH request.
Reviewers: Jun Rao <junrao@apache.org>, Alyssa Huang <ahuang@confluent.io>, Chia-Ping Tsai <chia7712@apache.org>
The PR handles fetch for `compacted` topics. The fix was required only
when complete batch disappears from the topic log, and same batch is
marked re-available in Share Partition state cache. Subsequent log reads
will not result the disappeared batch in read response hence respective
batch will be left as available in the state cache.
The PR checks for the first fetched/read batch base offset and if it's
greater than the position from where the read occurred (fetch offset)
then if there exists any `available` batches in the state cache then
they will be archived.
Reviewers: Andrew Schofield <aschofield@confluent.io>, Abhinav Dixit <adixit@confluent.io>
When a cluster is configured with a dynamic controller quorum, KRaft replica's endpoint are computed using the advertised.listeners property and not the quorum.controller.voters property. This change in the configuration makes it difficult to keeping all previous node configurations compatible with the new endpoint discovery functionality.
The least intrusive solution is to rely on Kafka's reverse hostname lookup when the hostname is not specified. The effective advertised controller listener now remove '0.0.0.0' hostname if the endpoint came from the listener configuration and not the advertised.listener configuration.
Reviewers: José Armando García Sancio <jsancio@apache.org>, Alyssa Huang <ahuang@confluent.io>
The PR handles slicing of fetched records based on acquire response for
share fetch. There could be additional bytes fetched from log but
acquired offsets can be a subset, typically with `max fetch records`
configuration. Rather sending additional bytes of fetched data to client
we should slice the file and wire only needed batches.
Note: If the acquired offsets are within a batch then we need to send
the entire batch within the file record. Hence rather checking for
individual batches, PR finds the first and last acquired offset, and
trims the file for all batches between (inclusive) these two offsets.
Reviewers: Christo Lolov <lolovc@amazon.com>, Andrew Schofield <aschofield@confluent.io>, Jun Rao <junrao@gmail.com>
The PR does following:
1. Adds `fetchOffset` to `acquire` API in SharePartition.
2. Adds a ShareFetchPartitionData class efficiently handle the
propagation of fetchOffset information.
3. Updates SharePartitionTests to make common code so such improvements
does not require all tests changes for future PRs.
Reviewers: Andrew Schofield <aschofield@confluent.io>
PR implements the final set of ShareGroupMetrics,
RequestTopicPartitionsFetchRatio and TopicPartitionsAcquireTimeMs, as
defined in KIP-1103:
https://cwiki.apache.org/confluence/display/KAFKA/KIP-1103%3A+Additional+metrics+for+cooperative+consumption
Note: Metric `RequestTopicPartitionsFetchRatio` is calculated as
percentage as Histogram API doesn't record double.
Reviewers: Andrew Schofield <aschofield@confluent.io>, Abhinav Dixit <adixit@confluent.io>
- ELR is enabled (ELRV_1) by default if the cluster is created with its bootstrap metadata version >= IBP_4_1_IV0.
- ELRV_1 can be manually enabled iff the metadata version is >= IBP_4_0_IV1.
Reviewers: Ismael Juma <ismael@juma.me.uk>, Colin P. McCabe <cmccabe@apache.org>, David Jacot <djacot@confluent.io>
3 tests which were marked flaky in ShareConsumerTest do not have any
failure on trunk since the test was converted to use `ClusterTestExtensions`.
Reviewers: Sushant Mahajan <smahajan@confluent.io>, Apoorv Mittal <apoorvmittal10@gmail.com>, Andrew Schofield <aschofield@confluent.io>
The current Docker Hub documentation for Kafka is based on the use of static voters. Since Kafka 4.0 utilizes dynamic voters, users following the doc of docker hub may encounter unexpected behavior. Due to the limited time available for the 4.0.0 release, a simple and quick solution is to revert to using static voters within the Docker image. This can be achieved by adding a configuration file with static voter definitions to the kafka/docker folder, keeping it separate from the main kafka/config directory. This approach allows us to encourage the use of dynamic voters in typical deployments while maintaining compatibility within the Docker image.
Reviewers: Vedarth Sharma <142404391+VedarthConfluent@users.noreply.github.com>, Chia-Ping Tsai <chia7712@gmail.com>
fix the following behavior changes.
1) in log4j 1, users can't change the logger by parent if the logger is declared by properties explicitly. For example, `org.apache.kafka.controller` has level explicitly in the properties. Hence, we can't use "org.apache.kafka=INFO" to change the level of `org.apache.kafka.controller` to INFO. By contrast, log4j2 allows us to change all child loggers by the parent logger.
2) in log4j2, we can change the level of root to impact all loggers' level. By contrast, log4j 1 can't.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>
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>
3.3.0 was the first KRaft release that was deemed production-ready and also
when KIP-778 (KRaft to KRaft upgrades) landed. Given that, it's reasonable
for 4.x to only support upgrades from 3.3.0 or newer (the metadata version also
needs to be set to "3.3" or newer before upgrading).
Noteworthy changes:
1. `AlterPartition` no longer includes topic names, which makes it possible to
simplify `AlterParitionManager` logic.
2. Metadata versions older than `IBP_3_3_IV3` have been removed and
`IBP_3_3_IV3` is now the minimum version.
3. `MINIMUM_BOOTSTRAP_VERSION` has been removed.
4. Removed `isLeaderRecoverySupported`, `isNoOpsRecordSupported`,
`isKRaftSupported`, `isBrokerRegistrationChangeRecordSupported` and
`isInControlledShutdownStateSupported` - these are always `true` now.
Also removed related conditional code.
5. Removed default metadata version or metadata version fallbacks in
multiple places - we now fail-fast instead of potentially using an incorrect
metadata version.
6. Update `MetadataBatchLoader.resetToImage` to set `hasSeenRecord`
based on whether image is empty - this was a previously existing issue that
became more apparent after the changes in this PR.
7. Remove `ibp` parameter from `BootstrapDirectory`
8. A number of tests were not useful anymore and have been removed.
I will update the upgrade notes via a separate PR as there are a few things that
need changing and it would be easier to do so that way.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>, Jun Rao <junrao@gmail.com>, David Arthur <mumrah@gmail.com>, Colin P. McCabe <cmccabe@apache.org>, Justine Olshan <jolshan@confluen.io>, Ken Huang <s7133700@gmail.com>
- update reflection-config.json and resource-config.json to include log4j2 and jackson
- remove unused jackson scala library
- fix the incorrect path of log4j2.yaml
- adopt workaround (--standalone) to make this PR work and it will be fixed by KAFKA-18737)
Reviewers: TengYao Chi <kitingiao@gmail.com>, Chia-Ping Tsai <chia7712@gmail.com>
- Fixed the RemoteIndexCacheTest that fails with caffeine > 3.1.1
Reviewers: Luke Chen <showuon@gmail.com>, Kamal Chandraprakash <kamal.chandraprakash@gmail.com>
The stale/invalid files that ends-with ".deleted" and ".tmp" should be cleaned when the broker gets restarted.
- fix the remote-index-cache test to use the logDir instead of topicDir
- fix the flaky test
Reviewers: Luke Chen <showuon@gmail.com>
In PR #18267, we removed old message format for cases in ConsumerWithLegacyMessageFormatIntegrationTest. Although test cases can pass, they don't fulfill original purpose. We can't send old message format since 4.0, so I change cases to append old records by ReplicaManager directly.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>
Frequently updating the trust store can cause unexpected termination of the AsyncConsumer background thread.
1. To resolve this issue, reuse the same AdminClient instead of recreating it.
2. Add error logging when fail to initialize resources for the consumer network thread.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>
KAFKA-16720 aims to add the support for the AlterShareGroupOffsets AdminClient. Key Changes in the PR:
1. Added handing of alterShareGroupOffsets() in KafkaAdminClient and introduce AlterShareGroupOffsetRequest/AlterShareGroupOffsetResponse/AlterShareGroupOffsetsOptions classes.
2. Corresponding test in KafkaAdminClientTest.
3. Added ALTER_SHARE_GROUP_OFFSETS API (will finish it in next PR and the share coordinator pieces)
Reviewers: poorv Mittal <apoorvmittal10@gmail.com>, Andrew Schofield <aschofield@confluent.io>, Chia-Ping Tsai <chia7712@gmail.com>
During the transaction commit phase, it is normal to hit CONCURRENT_TRANSACTION error before the transaction markers are fully propagated. Instead of letting the client to retry the produce request, it is better to retry on the server side.
Reviewers: Artem Livshits <alivshits@confluent.io>, Justine Olshan <jolshan@confluent.io>
Rewrite BrokerMetricNamesTest using ReplicaManager.MetricNames, ensuring that all metrics are always included. This helps prevent issues like PartitionsWithLateTransactionsCount not being correctly included in the test before.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>
The case testDescribeQuorumRequestToControllers shutdowns raft client but not the controller. This makes client has chance to send a request to the controller and get NOT_LEADER_OR_FOLLOWER error. However, if the raft client finishes shutdown before handling the request, the request will not be handled. Shutdown the controller before doing KafkaFuture#get for the client request, so we can make sure the request is handled by another controller eventually.
Signed-off-by: PoAn Yang <payang@apache.org>
Reviewers: Luke Chen <showuon@gmail.com>
Sometimes we didn't get into abortable state before aborting, so the epoch didn't get bumped. Now we force abortable state with an attempt to send before aborting so the epoch bump occurs as expected.
Reviewers: Jeff Kim <jeff.kim@confluent.io>
* KAFKA-18758: NullPointerException in shutdown following InvalidConfigurationException
Add checks for null in shutdown as BrokerLifecycleManager is not instantiaited if LogManager constructor throws an Exception
After log4j migration, we need to update the logging configuration in KafkaDockerWrapper from log4j1 to log4j2.
Reviewers: Manikumar Reddy <manikumar.reddy@gmail.com>
it's crucial to utilize a try-finally block to ensure proper closure of the ReplicaManager. Failing to do so can result in an unreleased thread from the purgatory, potentially leading to errors in subsequent integration tests that incorporate thread leak detection.
Reviewers: poorv Mittal <apoorvmittal10@gmail.com>, Chia-Ping Tsai <chia7712@gmail.com>
This commit ensures that the ClientQuotaCallback#updateClusterMetadata method is executed in KRaft mode. This method is triggered whenever a topic or cluster metadata change occurs. However, in KRaft mode, the current implementation of the updateClusterMetadata API is inefficient due to the requirement of creating a full Cluster object. To address this, a follow-up issue (KAFKA-18239) has been created to explore more efficient mechanisms for providing cluster information to the ClientQuotaCallback without incurring the overhead of a full Cluster object creation.
Reviewers: Mickael Maison <mickael.maison@gmail.com>, TaiJuWu <tjwu1217@gmail.com>, Chia-Ping Tsai <chia7712@gmail.com>
If Log4j Core is absent, most calls to Log4jController and Loggers will end up with a NoClassDefFoundError.
This changeset:
- Profits from the major version bump to rename k.util.Log4jController to LoggingController.
- Removes o.a.l.l.Level from the signature of public methods of o.a.k.connect.runtime.Loggers and replaces it with String.
- Provides an additional no-op implementation of k.util.LoggingController and o.a.k.connect.runtime.Loggers: if Log4j Core is not present on the runtime classpath the no-op implementation will be used.
Reviewers: Mickael Maison <mickael.maison@gmail.com>, Chia-Ping Tsai <chia7712@gmail.com>
Remove broker.id.generation.enable and reserved.broker.max.id, which are not used in KRaft mode.
Remove inter.broker.protocol.version, which is not used in KRaft mode.
Reviewers: PoAn Yang <payang@apache.org>, Ismael Juma <ismael@juma.me.uk>, Chia-Ping Tsai <chia7712@gmail.com>
Since zk has been removed in 4.0, config handlers no longer need to handle the "<default>" value. This PR streamlines the config update process by eliminating the unnecessary string checks for "<default>"
Reviewers: Christo Lolov <lolovc@amazon.com>, Ismael Juma <ismael@juma.me.uk>, Chia-Ping Tsai <chia7712@gmail.com>
related to KAFKA-18206, set features in EmbeddedKafkaCluster in both streams and connect module, note that this PR also fix potential transaction with empty records in sendPrivileged method as transaction version 2 doesn't allow this kind of scenario.
Reviewers: Justine Olshan <jolshan@confluent.io>
It appears this test was failing because the transaction was never aborting and the concurrent transactions errors would not go away.
ccab9eb introduced the test failure because it requires the transaction to complete, but I suspect the lack of completion was happening before the change.
The timeout for the write is based on the transactional timeout, and 100ms seemed too small -- thus the requests to update the state would often repeatedly time out.
Also removed the loop since it was not necessary.
Reviewers: Jeff Kim <jeff.kim@confluent.io>, Calvin Liu <caliu@confluent.io>
We need to re-enable the unclean shutdown detection when in ELR mode, which was inadvertently removed during the development process.
Reviewers: David Mao <dmao@confluent.io>, Jun Rao <junrao@gmail.com>
https://issues.apache.org/jira/browse/KAFKA-18575 solved a critical race condition by returning with CONCURRENT_TRANSACTIONS early when the transaction was still completing.
In testing, it was discovered that this early return could cause performance regressions.
Prior to KIP-890 the addpartitions call was a separate call from the producer. There was a previous change https://issues.apache.org/jira/browse/KAFKA-5477 that decreased the retry backoff to 20ms. With KIP-890 and making the call through the produce path, we go back to the default retry backoff which takes longer. Prior to 18575 we introduce a slight delay when sending to the coordinator, so prior to 18575, we are less likely to return quickly and get stuck in this backoff. However, based on results from produce benchmarks, we can still run into the default backoff in some scenarios.
This PR reverts KAFKA-18575, and doesn't return early and wait until the coordinator for checking if a transaction is ongoing. Instead, it will fix the handling with the verification guard so we don't hit the edge condition.
Also cleans up some of the verification text that was unclear.
Reviewers: Jeff Kim <jeff.kim@confluent.io>, Artem Livshits <alivshits@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>
Return produce v0-v2 as supported versions in `ApiVersionsResponse`, but disable support
for it everywhere else.
Since clients pick the highest supported version by both client and broker during version
negotiation, this solves the problem with minimal tech debt (even though it's not ideal that
`ApiVersionsResponse` becomes inconsistent with the actual protocol support).
Add one test for the socket server handling (in `ProcessorTest`) and one test for the
client behavior (in `ProduceRequestTest`). Adjust a couple of api versions tests to verify
the new behavior.
Finally, include a few clean-ups in `ApiKeys`, `Protocol`, `ProduceRequest`,
`ProduceRequestTest` and `BrokerApiVersionsCommandTest`.
Reference to related librdkafka issue:
https://github.com/confluentinc/librdkafka/issues/4956
Reviewers: Jun Rao <junrao@gmail.com>, Chia-Ping Tsai <chia7712@gmail.com>, Stanislav Kozlovski <stanislav_kozlovski@outlook.com>
This change implement some of the metrics enumerated in KIP-853.
The KafkaRaftMetrics object now exposes number-of-voters, number-of-observers and uncommitted-voter-change. The number-of-observers and uncommitted-voter-change metrics are only present on the active controller or leader, since it does not make sense for other replicas to report these metrics.
In order to make these two metrics thread-safe, KafkaRaftMetrics needs to be passed into LeaderState, and therefore QuorumState. This introduces a circularity since the KafkaRaftMetrics constructor takes in QuorumState. To break the circularity for now, the logic using QuorumState will be moved to the KafkaRaftMetrics#initialize method.
The BrokerServerMetrics object now exposes ignored-static-voters. The ControllerServerMetrics object now exposes IgnoredStaticVoters. To implement both metrics for "ignored static voters", this PR introduces the ExternalKRaftMetrics interface, which allows for higher layer metrics objects to be accessible within the raft module.
Reviewers: José Armando García Sancio <jsancio@apache.org>
Fixed the typo that used the wrong producer ID and epoch when returning so that we handle epoch overflow correctly.
We also had to rearrange the concurrent transaction handling so that we don't self-fence when we start the new transaction with the new producer ID.
I also tested this with a modified version of the code where epoch overflow happens on the first epoch bump (every request has a new producer id)
Reviewers: Artem Livshits <alivshits@confluent.io>, Jeff Kim <jeff.kim@confluent.io>
While testing, it was found that the not_enough_replicas error was super common and could be easily confused. Since we are already bumping the request, we can signify that the produce request may return this error and new clients can handle it
(Note, the java client should be able to handle this already as a retriable error, but other client libraries may need to implement this change)
Reviewers: Justine Olshan <jolshan@confluent.io>
Ensure we always return empty records (including cases where an error is returned).
We also remove `nullable` from `records` since it is effectively expected to be
non-null by a large percentage of clients in the wild.
This behavior regressed in fe56fc9 (KAFKA-18269). Empty records were
previously set via `FetchResponse.recordsOrFail(partitionData)` in the
now-removed `maybeConvertFetchedData` method.
Added an integration test that fails without this fix and also update many
tests to set `records` to `empty` instead of leaving them as `null`.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>, David Arthur <mumrah@gmail.com>
During testing, we identified that kafka-python (and aiokafka) relies on metadata request v0 and
hence we need to add these back to comply with the premise of KIP-896 - i.e. it should not
break the clients listed within it.
I reverted the changes from #18218 related to the removal of metadata versions 0-3.
I will submit a separate PR to undeprecate these API versions on the relevant 3.x branches.
kafka-python (and aiokafka) work correctly (produce & consume) with this change on
top of the 4.0 branch.
Reviewers: David Arthur <mumrah@gmail.com>
The following tests were previously reported as flaky but were only annotated with a comment in pull request #18558 due to module dependency limitations:
testAdminClientApisAuthenticationFailure
testOutdatedCoordinatorAssignment
testThrottledProducerConsumer
With the introduction of the new test infrastructure #18602 , which allows all modules to use the @Flaky annotation, these tests should now be updated to include the @Flaky annotation.
Reviewers: TengYao Chi <kitingiao@gmail.com>, Chia-Ping Tsai <chia7712@gmail.com>
The PR implements the BrokerTopicMetrics defined in KIP-1103.
The PR also corrected the share-acknowledgement-rate and share-acknowledgement-count metrics defined in KIP-932 as they are moved to BrokerTopicMetrics, necessary changes to KIP-932 broker metrics will be done once we complete KIP-1103.
Reviewers: Andrew Schofield <aschofield@confluent.io>, Chia-Ping Tsai <chia7712@gmail.com>, Jun Rao <junrao@gmail.com>
This patch reorganizes our test infrastructure into three Gradle modules:
":test-common:test-common-internal-api" is now a minimal dependency which exposes interfaces and annotations only. It has one project dependency on server-common to expose commonly used data classes (MetadataVersion, Feature, etc). Since this pulls in server-common, this module is Java 17+. It cannot be used by ":clients" or other Java 11 modules.
":test-common:test-common-util" includes the auto-quarantined JUnit extension. The @Flaky annotation has been moved here. Since this module has no project dependencies, we can add it to the Java 11 list so that ":clients" and others can utilize the @Flaky annotation
":test-common:test-common-runtime" now includes all of the test infrastructure code (TestKitNodes, etc). This module carries heavy dependencies (core, etc) and so it should not normally be included as a compile-time dependency.
In addition to this reorganization, this patch leverages JUnit SPI service discovery so that modules can utilize the integration test framework without depending on ":core". This will allow us to start moving integration tests out of core and into the appropriate sub-module. This is done by adding ":test-common:test-common-runtime" as a testRuntimeOnly dependency rather than as a testImplementation dependency. A trivial example was added to QuorumControllerTest to illustrate this.
Reviewers: Ismael Juma <ismael@juma.me.uk>, Chia-Ping Tsai <chia7712@gmail.com>
All the work that we have done to automate and to simplify the coordinator records allows us to simplify the related MessageParsers in DumpLogSegments. They can all share the same based implementation. This is nice because it ensures that we handle all those records similarly.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>
This patch is a first step towards removing `ReplicaManager#becomeLeaderOrFollower`. It updates the `LocalLeaderEndPointTest` tests.
Reviewers: Christo Lolov <lolovc@amazon.com>, Ismael Juma <ismael@juma.me.uk>
There is a subtle race condition with transactions V2 if a transaction is still completing when checking if we need to add a partition, but it completes when the request reaches the coordinator.
One approach was to remove the verification for TV2 and just check the epoch on write, but a simpler one is to simply return concurrent transactions from the partition leader (before attempting to add the partition). I've done this and added a test for this behavior.
Locally, I reproduced the race but adding a 1 second sleep when handling the WriteTxnMarkersRequest and a 2 second delay before adding the partition to the AddPartitionsToTxnManager. Without this change, the race happened on every second transaction as the first one completed. With this change, the error went away.
As a followup, we may want to clean up some of the code and comments with respect to verification as the code is used by both TV0 + verification and TV2. But that doesn't need to complete for 4.0. This does :)
Reviewers: Jeff Kim <jeff.kim@confluent.io>, Artem Livshits <alivshits@confluent.io>, Calvin Liu <caliu@confluent.io>
While setting Defaults.LogMessageTimestampType to "LogAppendTime", `LogCleanerManagerTest.testLogsUnderCleanupIneligibleForCompaction()` fails with a InvalidTimestampException.
This PR fixes this by regenerating the records instead of previous approach of re-using same records in the test.
Reviewers: Divij Vaidya <diviv@amazon.com>, Kvicii <kvicii.yu@gmail.com>
---------
Co-authored-by: fangxiaobing <fangxiaobing@kuaishou.com>
These methods were previously invoked by ZK components, but we have just removed them.
Reviewers: Ismael Juma <ismael@juma.me.uk>, Chia-Ping Tsai <chia7712@gmail.com>
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>
This patch removes `ReplicaManager#stopReplicas`. I have ensured that removed unit tests are covered by other existing tests or are updated to use kraft.
Reviewers: Ismael Juma <ismael@juma.me.uk>, Chia-Ping Tsai <chia7712@gmail.com>
Although `MetadataCache`'s `getPartitionReplicaEndpoints` takes a single topic-partition, the `KRaftMetadataCache` implementation iterates over all partitions of the matching topic. This is not necessary and can cause significant performance degradation when the topic has a relatively high number of partitions.
Note that this is not a recent regression - it has been a part of `KRaftMetadataCache` since its creation.
Reviewers: Ismael Juma <ismael@juma.me.uk>, David Jacot <djacot@confluent.io>
This patch updates the transaction coordinator record to use the new coordinator record definition.
Reviewers: Andrew Schofield <aschofield@confluent.io>
Kafka 4.0 will remove support for zk mode and will require conversion to kraft
before upgrading to 4.0. The minimum kraft version is 3.0 (aka 3.0-IV1).
This provides an opportunity to remove exclusively server side protocols versions
that only exist to allow direct upgrades from versions older than 3.0 or that are
used only by zk mode.
Since KRaft became production ready in 3.3, we should consider setting the
baseline to 3.3. But that requires more discussion and it can be done via a
separate change (KAFKA-18601).
Protocol changes:
* Remove RequestHeader v0 (only used by ControlledShutdown v0)
* Remove WriteTxnMarkers v0
* Remove all versions of ControlledShutdown, LeaderAndIsr, StopReplica, UpdateMetadata
In order to remove all versions safely, extend generator to support setting
"versions" to "none". In this case, we no longer generate the `*Data` classes,
but we still reserve the id for the relevant protocol api (so it doesn't get
accidentally used for something else). The protocol documentation is correct
after these changes.
We kept a simplified version of `LeaderAndIsr{Request|Response}` because
it's used by many tests that are still relevant in kraft mode. Once
KAFKA-18486 is done, it may be possible to remove it (I left a comment on
the ticket). Similarly, KAFKA-18487 may make it possible to remove
the introduced `StopReplicaPartitionState` (left a comment on that ticket too).
There are a number of places that were adjusted to include an
`ApiKeys.hasValidVersion` check.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>
The reason for flakiness is PR #18080 which modifies the linger.ms config from 0 to 5. ClientIdQuotaTest are testing "Low enough quota that a producer sending a small payload in a tight loop should get throttled", thus this config change Influence this test scenario.
This commits uses the older value of 0ms for linger.ms for ClientIdQuotaTest tests.
Reviewers: Ismael Juma <ismael@juma.me.uk>, TaiJuWu <tjwu1217@gmail.com>, Divij Vaidya <diviv@amazon.com>
Since maxOverCleanerThreads does not have a corresponding unit test,
I have added a unit test for it. maintainUncleanablePartitions has been
thoroughly tested in tests, so I simply replaced the old implementation
with the new one.
Reviewers: Ismael Juma <ismael@juma.me.uk>, Chia-Ping Tsai <chia7712@gmail.com>
This was removed during removal of zk code (#18542), but
we should instead convert it to work with kraft.
Reviewers: Ismael Juma <ismael@juma.me.uk>, Chia-Ping Tsai <chia7712@gmail.com>
Currently, each log.append() will add at most 1 index entry, even when the appended data is larger than log.index.interval.bytes. One potential issue is that if a follower restarts after being down for a long time, it may fetch data much bigger than log.index.interval.bytes at a time. This means that fewer index entries are created, which can increase the fetch time from the consumers.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>, Jun Rao <junrao@gmail.com>
This PR implements the second part of KIP-996 and KAFKA-16164 (tasks KAFKA-16607, KAFKA-17642, KAFKA-17643, KAFKA-17675) which encompass the response handling of PreVotes, addition of new ProspectiveState, update to metrics, and addition of Raft simulation tests.
Voters now transition to ProspectiveState first before CandidateState to prevent unnecessary epoch bumps. Voters in ProspectiveState send PreVotes requests which are Vote requests with PreVote set to true.
Follower grants PreVotes if it has not yet fetched successfully from leader. Leader denies all PreVotes. Unattached, Prospective, Candidate, and Resigned will grant PreVotes if the requesting replica's log is at least as long as theirs. Granted PreVotes are not persisted like standard votes. It is possible for a voter to grant several PreVotes in the same epoch.
The only state which is allowed to transition directly to CandidateState is ProspectiveState. This happens on reception of majority of granted PreVotes or if at least one voter doesn't support PreVote requests.
Prospective will transition to Follower after election loss/timeout if it was already aware of last known leader and the leader's endpoint, or at any point if it discovers the leader.
Prospective will transition to Unattached after election loss/timeout if it does not know the leader endpoints.
After electionTimeout, Resigned now always transitions to Unattached and increases the epoch.
Prospective grants standard votes if it has not already granted a standard vote (no votedKey), has no leaderId, and the recipient's log is current enough
Candidate no longer backs off after election timeout. Candidate still backs off after election loss.
Reviewers: José Armando García Sancio <jsancio@apache.org>
Removed Optional for SharePartitionManager and ClientMetricsManager as zookeeper code is being removed. Also removed asScala and asJava conversion in KafkaApis.handleListClientMetricsResources, moved to java stream.
Reviewers: Mickael Maison <mickael.maison@gmail.com>, Ismael Juma <ismael@juma.me.uk>, Chia-Ping Tsai <chia7712@gmail.com>
The PR implements a way to divide acquired batches into batch size as desirable by client.
The BatchSize is the soft limit and should align the batches in response and cached state in broker at the log batch boundaries.
Reviewers: Andrew Schofield <aschofield@confluent.io>, Abhinav Dixit <adixit@confluent.io>, Jun Rao <junrao@gmail.com>
Fix the issue where producer.commitTransaction under transaction version 2 throws error if no partition or offset is added to transaction. The solution is to avoid sending the endTxnRequest unless producer.send or producer.sendOffsetsToTransaction is triggered.
Reviewers: Justine Olshan <jolshan@confluent.io>
The PR removes dependency of server module on share-coordinator, rather it should be other way. Moved the ShareCoordinatorConfig class from server to share-coordinator.
Reviewers: Ismael Juma <ismael@juma.me.uk>, Chia-Ping Tsai <chia7712@gmail.com>
The REMOTE_LOG_MANAGER_TASK_INTERVAL_MS_PROP in RemoteLogManagerTest is 100 which is too small. If assertions verifyNoMoreInteractions can't run in 100ms, the scheduler will run RLMTask again and the case will fail.
Reviewers: Luke Chen <showuon@gmail.com>
This patch updates the ShareGroupStateMessageParser and OffsetsMessageParser used by the DumpLogSegments command line tool to use the recently introduced json converters for those records. It basically means that new records are automatically supported.
Reviewers: Mickael Maison <mickael.maison@gmail.com>, Chia-Ping Tsai <chia7712@gmail.com>
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>
To avoid self-fencing in the commit/abort + empty abort scenario, return the concurrent transactions error when we have pending state and do the epoch check second. In this scenario, we will complete the previous transaction before proceeding to the empty abort.
Added a test that failed before the change.
Note -- only the pending state is checked earlier. This is because we don’t return from EndTxn (the first commit) until we already written to the log, transitioned to PrepareX, and have the pending CompleteX state. We don't need to worry about the cases of an EndTxn request coming in with PrepareX without the pending state because that would be an older request and/or retry which are already covered.
Reviewers: Artem Livshits <alivshits@confluent.io>, Jeff Kim <jeff.kim@confluent.io>
The test has become flakier recently and it's easy to reproduce by running the single test (vs
running the the class test suite).
The root cause is that following functions call `RemoteLogMetadataManager#listRemoteLogSegments`.
It returns iterator. If one of function goes through iterator first, another can't get expected result.
I changed `thenReturn` to `thenAnswer` to avoid the issue.
The race is between:
* RLMExpirationTask#cleanupExpiredRemoteLogSegments
* RemoteLogManager#deleteRemoteLogPartition
Reviewers: Ismael Juma <ismael@juma.me.uk>
Signed-off-by: PoAn Yang <payang@apache.org>
Convert v0/v1 record batches to v2 during compaction even if said record batches would be
written with no change otherwise. A few important details:
1. V0 compressed record batch with multiple records is converted into single V2 record batch
2. V0 uncompressed records are converted into single record V2 record batches
3. V0 records are converted to V2 records with timestampType set to `CreateTime` and the
timestamp is `-1`.
4. The `KAFKA-4298` workaround is no longer needed since the conversion to V2 fixes
the issue too.
5. Removed a log warning applicable to consumers older than 0.10.1 - they are no longer
supported.
6. Added back the ability to append records with v0/v1 (for testing only).
7. The creation of the leader epoch cache is no longer optional since the record version
config is effectively always V2.
Add integration tests, these tests existed before #18267 - restored, modified and
extended them.
Reviewers: Jun Rao <jun@confluent.io>
Delete the handlers for LEADER_AND_ISR, STOP_REPLICA, and UPDATE_METADATA. Also, remove the corresponding unit tests in KafkaApisTest.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>
Remove Apache ZooKeeper from the Apache Kafka build. Also remove commons IO, commons CLI, and netty, which were dependencies we took only because of ZooKeeper.
In order to keep the size of this PR manageable, I did not remove all classes which formerly interfaced with ZK. I just removed the ZK types. Fortunately, Kafka generally wrapped ZK data structures rather than using them directly.
Some classes were pretty entangled with ZK, so it was easier just to stub them out. For ZkNodeChangeNotificationListener.scala, PartitionStateMachine.scala, ReplicaStateMachine.scala, KafkaZkClient.scala, and ZookeeperClient.scala, I replaced all the functions with "throw new UnsupportedOperationException". Since the tests for these classes have been removed, as well as the ZK-based broker code, this should be OK as an incremental step.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>
Looking at the [history](https://ge.apache.org/scans/tests?search.rootProjectNames=kafka&search.timeZoneId=Europe%2FZurich&tests.container=kafka.api.PlaintextAdminIntegrationTest&tests.test=testConsumerGroups(String%2C%20String)%5B2%5D), I found out that one source of flakiness is due to syncCommit failing with CommitFailedException. We can ignore it and retry on the next iteration.
```
[2025-01-07 10:17:00,783] ERROR [Consumer instanceId=test_instance_id_1, clientId=test_client_id, groupId=test_group_id] OffsetCommit failed for member VfImExrxT3-w_HNJcTkqnw with stale member epoch error. Last epoch sent: 2 (org.apache.kafka.clients.consumer.internals.CommitRequestManager:773)
Exception in thread "Thread-6" org.apache.kafka.clients.consumer.CommitFailedException: OffsetCommit failed with stale member epoch.The member epoch is stale. The member must retry after receiving its updated member epoch via the ConsumerGroupHeartbeat API.
at org.apache.kafka.clients.consumer.internals.CommitRequestManager.commitSyncExceptionForError(CommitRequestManager.java:481)
at org.apache.kafka.clients.consumer.internals.CommitRequestManager.lambda$commitSyncWithRetries$7(CommitRequestManager.java:472)
at java.base/java.util.concurrent.CompletableFuture.uniWhenComplete(CompletableFuture.java:863)
at java.base/java.util.concurrent.CompletableFuture$UniWhenComplete.tryFire(CompletableFuture.java:841)
at java.base/java.util.concurrent.CompletableFuture.postComplete(CompletableFuture.java:510)
at java.base/java.util.concurrent.CompletableFuture.completeExceptionally(CompletableFuture.java:2162)
at org.apache.kafka.clients.consumer.internals.CommitRequestManager$OffsetCommitRequestState.onResponse(CommitRequestManager.java:776)
at org.apache.kafka.clients.consumer.internals.CommitRequestManager$RetriableRequestState.handleClientResponse(CommitRequestManager.java:893)
at org.apache.kafka.clients.consumer.internals.CommitRequestManager$RetriableRequestState.lambda$buildRequestWithResponseHandling$0(CommitRequestManager.java:883)
at java.base/java.util.concurrent.CompletableFuture.uniWhenComplete(CompletableFuture.java:863)
at java.base/java.util.concurrent.CompletableFuture$UniWhenComplete.tryFire(CompletableFuture.java:841)
at java.base/java.util.concurrent.CompletableFuture.postComplete(CompletableFuture.java:510)
at java.base/java.util.concurrent.CompletableFuture.complete(CompletableFuture.java:2147)
at org.apache.kafka.clients.consumer.internals.NetworkClientDelegate$FutureCompletionHandler.onComplete(NetworkClientDelegate.java:433)
at org.apache.kafka.clients.ClientResponse.onComplete(ClientResponse.java:154)
at org.apache.kafka.clients.NetworkClient.completeResponses(NetworkClient.java:669)
at org.apache.kafka.clients.NetworkClient.poll(NetworkClient.java:661)
at org.apache.kafka.clients.consumer.internals.NetworkClientDelegate.poll(NetworkClientDelegate.java:153)
at org.apache.kafka.clients.consumer.internals.ConsumerNetworkThread.runOnce(ConsumerNetworkThread.java:160)
at org.apache.kafka.clients.consumer.internals.ConsumerNetworkThread.run(ConsumerNetworkThread.java:106)
```
Reviewers: Lianet Magrans <lmagrans@confluent.io>
Remove RaftManager.maybeDeleteMetadataLogDir since it was only used during ZK migration, and that code has been removed.
Similarly, remove RaftManagerTest.testKRaftBrokerDoesNotDeleteMetadataLog which tested that function.
Remove AutoTopicCreationManagerTest since it tests the ZK-mode-only AutoTopicReationManager.
Reviewers: Mickael Maison <mickael.maison@gmail.com>, Chia-Ping Tsai <chia7712@gmail.com>
Following https://github.com/apache/kafka/pull/18261, this patch updates the Share Coordinator to use the new record format.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>, Andrew Schofield <aschofield@confluent.io>
There are times when the controller has a high event processing time, such as during startup, or when creating a topic with many partitions. We can see these processing times in the p99 metric (kafka.controller:type=ControllerEventManager,name=EventQueueProcessingTimeMs), however it's difficult to see exactly which event is causing high processing time.
With DEBUG logs, we see every event along with its processing time. Even with this, it's a bit tedious to find the event with a high processing time.
This PR logs all events which take longer than 2 seconds at ERROR level. This will help identify events that are taking far too long, and which could be disruptive to the operation of the controller. The slow event logging looks like this:
```
[2024-12-20 15:03:39,754] ERROR [QuorumController id=1] Exceptionally slow controller event createTopics took 5240 ms. (org.apache.kafka.controller.EventPerformanceMonitor)
```
Also, every 60 seconds, it logs some event time statistics, including average time, maximum time, and the name of the event which took the longest. This periodic message looks like this:
```
[2024-12-20 15:35:04,798] INFO [QuorumController id=1] In the last 60000 ms period, 333 events were completed, which took an average of 12.34 ms each. The slowest event was handleCommit[baseOffset=0], which took 41.90 ms. (org.apache.kafka.controller.EventPerformanceMonitor)
```
An operator can disable these logs by adding the following to their log4j config:
```
org.apache.kafka.controller.EventPerformanceMonitor=OFF
```
Reviewers: Colin P. McCabe <cmccabe@apache.org>
Remove the flaky annotation from the following tests
* RemoteLogManagerTest#testFetchOffsetByTimestampWithTieredStorageDoesNotFetchIndexWhenExistsLocally
* All the children of BaseConsumerTest#testCoordinatorFailover
* TransactionsTest#testFailureToFenceEpoch
* TransactionsTest#testReadCommittedConsumerShouldNotSeeUndecidedData
* MetricsDuringTopicCreationDeletionTest#testMetricsDuringTopicCreateDelete
* ProduceRequestTest#testProduceWithInvalidTimestamp
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>
Nick Guo
Mahsa Seifikar
Apoorv Mittal
Lucas Brutschy
Xuan-Zhang Gong
TengYao Chi
Dongnuo Lyu
Abhinav Dixit
José Armando García Sancio
xijiu
TaiJuWu
PoAn Yang
mingdaoy
Ismael Juma
Sushant Mahajan
Calvin Liu
Shivsundar R
Matthias J. Sax
Parker Chang
Mickael Maison
Andrew Schofield
Chirag Wadhwa
Kamal Chandraprakash
Ming-Yen Chung
Jimmy Wang
陳昱霖(Yu-Lin Chen)
Ken Huang
Jhen-Yung Hsu
Justine Olshan
Edoardo Comar
Piotr P. Karwasz
Colin Patrick McCabe
Kuan-Po Tseng
Sanskar Jhajharia
Sean Quah
Luke Chen
David Jacot
kevin-wu24
David Arthur
Logan Zhu
Lianet Magrans
Xiaobing Fang
Dimitar Dimitrov
Alyssa Huang
Kirk True
Jason Taylor
Dmitry Werner
Ted Yan
Peter Lee
NICOLAS GUYOMAR
Zhihong Yu