- 11月 17, 2022
-
-
由 Patrick Sullivan 创作于
-
由 Sanjana Kaundinya 创作于
- 11月 16, 2022
-
-
由 kpatelatwork 创作于
* KGLOBAL-2389 NettySslEngineFactory to support trustmanager algorithm * parameterized the test with both factories * removed unused method and change visibility
-
由 Vincent Jiang 创作于
This reverts commit 2615c2ef. https://github.com/confluentinc/ce-kafka/pull/7650 introduced a change that breaks kafkatest.tests.core.downgrade_tiering_test We found another issue with this PR change in devel: when tier compaction is enabled for the first time, the replicas with includeCompactionFields==false cannot handle TierSegmentUploadInitiate events with uploadType=Compaction properly because TierSegmentUploadInitiate events are not materialized into proper TierObjectMetadata objects. To ensure it works correctly, we need to two rolls to enable tier compaction: in roll #1, set includeCompactionFields=true. From this point, back compatibility is broken. in roll #2, enable tier compaction.
-
由 Anastasia Vela 创作于
This PR creates a new ClientQuotaType called ProducerId, as well as creates its quota manager called the ProducerIdQuotaManager for managing the quota for the producer id count. The ProducerIdQuotaManager extends the ClientQuotaManager class to utilize the QuotaCallback to manage the quotas and metricTag. Each client is registered with the 2 original sensors (rate metric and token bucket metric) as well as an additional count metric to keep track of the number of producer ids owned by that client. Added a series of tests to ensure throttling behavior is as expected.
-
由 Akhilesh C 创作于
[skip secret scan] KGLOBAL-2023: Controller support to remove cluster link references in ACLs on removing cluster link. (#7459) KGLOBAL-2023: Controller support to remove cluster link references in ACLs on removing cluster links. When a customer deletes a cluster link, the current behavior is to remove the link id referenced in the synced ACLs. Upon removing the link id, if there's a local version available, we delete the cluster-linked synced ACL. Otherwise, we create a new local ACL and delete the cluster-link synced ACL. In KRaft mode, we depend on the link coordinator to clear the link ID for such synced ACLs. If the link coordinator Broker restarts during this process, the synced ACLs will forever have the dangling link id and cause unnecessary confusion to the customer for ACL management. In KRaft mode, we already clear the link references in mirror topic and partition states in KRaft Controller based on the `RemoveClusterLinkRecord` record. With this change, we use the same record to remove the referenced link id in synced ACLS. For this change to work, I had to change the notion of the immutability of ACL. For instance, if a local version of the synced ACL doesn't exist, we re-use the same ACL ID to represent the converted local ACL. This way, we can correctly replay the `RemoveClusterLinkRecord` in the Broker. Otherwise, if we use a new ID in the Controller, there is no way for the Broker to know the id unless we generate a new record for the ACL. We want to avoid causing multiple records for new ACLs after cluster link deletion (as they could quickly run into thousands of records). So changed `AclControlManager` and `StandardAuthorizer` to work with the new behavior of updating ACLs for the same id.
-
由 Lingnan Liu 创作于
* KAFKALESS-1376: GoalOptimizationHistory * Add more javadoc * Fine-tune test * Change fields to private * Remove unnecessary complexity and add more docs * Consider bad listeners, more logs and more docs * Add documentation on listeners' responsibility of ensuring monotonic updates of topic partition history * Add docs on epoch update
-
由 Colin Patrick McCabe 创作于
We should avoid using Message.highestSupportedVersion to generate metadata records. Instead, we need to pick the correct record version based on the current metadata version which is in effect. In cases where there is only one record version that we know how to generate, we can hard code that version, but it should just be a simple constant zero. Reviewers: Jason Gustafson <jason@confluent.io> Conflicts: Also remove ce-kafka-specific uses of highestSupportedVersion.
-
由 Luke Young 创作于
[security] upgrade cc-base docker image to v20.6.0-jdk-17 and cc-base-alpine docker image to v2.16.0 (#8011)
-
由 Yang Yu 创作于
This PR adds misclassification detection logic for stray partition deletion. A new PartitionMarkedStrayException is added. Will return this exception for: LeaderAndIsr request that makes a stray partition leader LeaderAndIsr request that makes a stray partition follower, but is not part of the incoming reassignment Early stray partition deletion for some benign cases: LeaderAndIsr request that makes a stray partition follower and is part of an incoming reassignment: will initiate an early deletion for the stray partition as the newly reassigned replica should override the stray partition. StopReplica request with deletePartitions=true: will initiate an early deletion for the stray partition as it is scheduled to be deleted anyways.
-
由 Stanislav Kozlovski 创作于
This patch adds a metric that tracks whether SBC v2 is enabled (incremental goals enabled) kafka.databalancer:type=GoalOptimizer,name=incremental-balancing-enabled
- 11月 15, 2022
-
-
由 k-raina 创作于
Co-authored-by: Matthew de Detrich <matthew.dedetrich@aiven.io>
-
由 Stanislav Kozlovski 创作于
When the event queue was first introduced, we logged all events' handling at INFO level. This proved to be very noisy, particularly around the MetadataEvent. Any log search in Kibana will show that this log forms a large percentage of all SBC KRaft logs. This generates unnecessary load in our ccloud production log clusters, log pipeline and makes investigation into logs much more burdensome This PR adds a configurable log level for this sort of log so that each event can override it. This PR also overrides the log level for the following read events: - BalancerStatus - EvenClusterLoadStatus - ListBrokerAdditions - ListBrokerRemovals and the noisy UpdateMetadataEvent to all be at DEBUG. As you can see in our soak cluster pkcc-7wx51, 13% of all logs are Handling SBCMetadataUpdateEvent
-
由 Sushant Mahajan 创作于
* Added support for identity in v2 schema audit logs.
- 11月 11, 2022
-
-
由 Stanislav Kozlovski 创作于
This patch deletes the unused Strategy#DEMOTED. It was deprecated in February 2022 but there is no real reason to keep it in the code at all - it is entirely unused and only brings confusion to the product.
-
由 andymg3 创作于
KGLOBAL-1501: Fix error message propagation and improve error message when mirror topic name is not expected (#8053)
-
由 Confluent Jenkins Bot 创作于
-
由 Confluent Jenkins Bot 创作于
-
由 Corey Christous 创作于
-
由 Mathew Hogan 创作于
Reviewer: @alok123t
-
由 Lingnan Liu 创作于
* KAFKALESS-1376: Add SuspendedTopicPartitionPool * Use MockTime in unit test * Consider epoch * Address comments * Added boolean return value to add() method * Removed timeout in await method in tests * Make tests more stable * Convert test methods to be public * Add a method to add history without indexing * Rename take method to takeExpired