- 8月 05, 2022
-
-
由 Eric Wu 创作于
The Kafka controller and other parts of Kafka call into the ZKTopicMetadataCollector. We should protect that rest of Kafka against error by making sure that none of the public methods of ZKTopicMetadataCollector throw any Exception.
-
由 Grace Chen 创作于
* Update README with PR docker image push Update README with instructions for how to push docker images after #6526 changes * Update README.md
-
由 Vikas Singh 创作于
KAFKALESS-1247: Fix CPU metric for non-Linux system and ConfluentTelemetryReporterSamplerIntegrationTest timeout issue (#6865) This PR updates CPU metrics to fall back to one reported by OperatingSystemMXBean if SBC is running on non-Linux systems. This fixes the regression on Mac introduced by PR #6714. In addition to this TestUtils.retryOnExceptionWithTimeout had a timeout of 5 seconds whereas the internal sampler.getSamples had a timeout of 300 seconds. As sampler samples once every minute, we need both to be over that limit. This change bumps the first timeout to be 30 seconds more than the sampling timeout. The test was flaky if the samples were not generated within 5 seconds of bringing server up. This change fixes that. Test passes locally, it was failing consistently.
-
由 David Arthur 创作于
Enable some of the dynamic broker reconfiguration tests in KRaft mode Conflicts: KafkaBroker.scala: fix conflicts created by KMETA-249. DynamicBrokerReconfigurationTest.scala: fix conflicts created by extra tests in ce-kafka, and UnifiedLog vs. AbstractLog issue. ClusterLinkConfigEncoder: handle change in PasswordEncoder from concrete class to interface. MetricReporterClusterTestHarness.java: use KafkaBroker, not KafkaServer.
-
由 Divij Vaidya 创作于
Reviewers: Mickael Maison <mickael.maison@gmail.com>
-
由 dengziming 创作于
The code used BrokerRegistrationFencingChange.FENCE when unfencing a broker and used BrokerRegistrationFencingChange.UNFENCE when fencing a broker, this is confusing. This commit flips the values of the two enums and changes their usage at all of the call sites. Reviewers: José Armando García Sancio <jsancio@users.noreply.github.com>
-
由 Colin Patrick McCabe 创作于
KRaft should validate that manual assignments given to createTopics are contiguous. In other words, they must start with partition 0, and progress through 1, 2, 3, etc. ZK mode does this, but KRaft mode previously did not. Also fix a null pointer exception when the placement for partition 0 was not specified. Convert over AddPartitionsTest to use KRaft. This PR converts all of the test except for some of the placement logic tests, which will need to be redone for KRaft mode in a future change. Fix null pointer exception in KRaftMetadataCache#getPartitionInfo. Specifically, we should not assume that the partition will be found in the hash map. This is another case where we had "Some(x)" but it should be "Option(x)." Fix a potential null pointer exception in BrokerServer#state. Reviewers: dengziming <dengziming1993@gmail.com>, Jason Gustafson <jason@confluent.io> Conflicts: BrokerServer.scala: fix some minor conflicts caused by additional config logic in CE AddPartitionsTest.scala: fix some minor conflicts, mostly caused by additional parameters to ZK-specific functions.
-
由 Bob Barrett 创作于
* KC-2217: Remove restriction on secured, non-TLS listeners * Checkstyle fix
-
由 Aishwarya Gune 创作于
* Add system test for triggerEvenClusterLoad command
-
由 Vikas Singh 创作于
This change adds a resource optimization framework to SBC. First version of the framework only has a vanilla cpu detction class that does basic over/under utilization check and generate a metric that can be used to take action on it. Co-authored-by: Shubhankar Ranade <sranade@confluent.io>
- 8月 04, 2022
-
-
由 David Mao 创作于
Building the principal takes ~ 1% of CPU since this is invoked on every request. Another implication is that the Scope used for RBAC is tied to a MultiTenantPrincipal - if we rebuild a principal on every request, then we recompute hash each time. We use a separate authenticator on reauthentication - so this should be safe from a reauth perspective.
-
由 Daniel Gospodinow 创作于
Adjusting the log level to something less stressful.
-
由 Grace Chen 创作于
* Test PR builds tagging images * docker login * gradle properties * Try PR builds on semaphore instead * Empty push to trigger semaphore CI * Try with make build-docker * Re-add copying gradle properties * Use old make docker * Try removing extraneous secrets * nit * Return secrets * Push only on PR titles containing pr-image string * nit * Test commit to trigger image build * Point to PR title * Test no build when not in PR title
-
由 Aman Singh 创作于
* KAFKA-6945: KIP-373, allow users to create delegation token for others (#10738) Reviewers: Manikumar Reddy <manikumar.reddy@gmail.com> Co-authored-by: Viktor Somogyi-Vass <viktorsomogyi@gmail.com>
-
由 kpatelatwork 创作于
* KGLOBAL-1775 Authenticate NetworkRegionId is allowed to talk to this cluster deleted as we will include this in the main PR * minor nits * minor compile issues * added unit tests for PlainSaslServer authenticate * initialize SaslServer only once as authenticate can be called many times during mechanisms like SCRAM * removed extra {} in logs * added Reconfigurable authenticate callback handle as that seems to be one decent way to to test the networkRegionId injection * used newConfigs instead of configs * removed ReconfigurableAuthenticateCallbackHandler based on review comments before embarking on a new approach to testing this * checkstyle issues * removed unused NetworkRegionId
-
由 Vikas Singh 创作于
* KAFKALESS-1261: Use newly added follower fetch rate metric This change updates the code to use follower fetch reate metric that was added to Kafka to capture the replication load that partition have from followers perspective. This change adds code that make the metric part of Metric Sample. The metric has not been used to populate cluster model or use to attribute the cpu to a replica. That will be done in future changes. Unit tests pass. The tests are generic so they test all metrics to be in sane state in all data-structures that we need.
-
由 Colin Patrick McCabe 创作于
* MINOR: convert some more junit tests to support KRaft Introduce TestUtils#waitUntilLeaderIsElectedOrChangedWithAdmin, a ZK-free alternative to TestUtils#waitUntilLeaderIsElectedOrChanged. Convert PlaintextProducerSendTest, SslProducerSendTest, TransactionsWithMaxInFlightOneTest, AddPartitionsToTxnRequestServerTest and KafkaMetricsReporterTest to support KRaft Reviewers: dengziming <dengziming1993@gmail.com>, David Arthur <mumrah@gmail.com> Conflicts: BaseProducerSendTest.scala: fix a minor conflict
-
Read and write access to the TreeMap in snapshots needs to be synchronized. Reviewers: David Arthur <mumrah@gmail.com>
-
由 Jason Gustafson 创作于
Updates relevant tests in `ReassignPartitionsIntegrationTest` for KRaft. We skip JBOD tests since it is not supported and we skip `AlterPartition` upgrade tests since they are not relevant. Reviewers: Kvicii <Karonazaba@gmail.com>, David Arthur <mumrah@gmail.com> Conflicts: in ReassignPartitionsIntegrationTest, the default values for some configurations were different in ce-kafka. Also, there were some additional ce-kafka-only tests that I had to annotate as ZK-specific.
-
由 Colin Patrick McCabe 创作于
Convert ProducerCompressionTest, MirrorMakerIntegrationTest, EdgeCaseRequestTest to kraft. Make it explicit that ServerShutdownTest#testControllerShutdownDuringSend is ZK-only. Reviewers: David Arthur <mumrah@gmail.com> Conflicts: ServerShutdownTest.scala: disable testNoCleanShutdownAfterFailedStartupDueToCorruptLogs in ZK mode due to CPKAFKA-8947, which appears to be a pre-existing problem.
-
由 Prabhash Kumar 创作于
-
由 Aman Singh 创作于
Added metrics for monitoring Kafka Management events Testing: Tested metrics by generating the respective events.
- 8月 03, 2022