- 8月 23, 2020
-
-
由 Ismael Juma 创作于
It includes a significant performance improvement: "Just for what it worth... in some benchmarks we saw a 50% throughput increase while decrease handshake latency from 2000ms to 40ms when this PR is applied" https://github.com/netty/netty-tcnative/pull/568 Reviewers: Gwen Shapira <gwen@confluent.io>
- 8月 22, 2020
-
-
由 Manikumar Reddy 创作于
- 8月 21, 2020
-
-
由 Rajini Sivaram 创作于
Reviewers: Manikumar Reddy
-
由 David Arthur 创作于
Remove static method call to NettySslEngineFactory that was causing undesired Netty classes to be loaded (resulting in exceptions in cases where Netty is not in the classpath). Reviewers: Ismael Juma <ismael@juma.me.uk>
-
由 Ismael Juma 创作于
The dependencies we are shading (eg guava) in these modules are present due to other dependencies and/or modules so this just adds complexity and larger jars (due to the same classes being relocated multiple times while the original classes are also present). While at it: 1. Fixed stale comment in ce-sbk build file regarding why we need the ce-metrics dependency. 2. Added gson as an explicit dependency to ce-sbk as it's used by `BrokerCapacityConfigFileResolver`. 3. Removed scalaReflect dependency from ce-sbk as we handle versioning consistency automatically via a resolutionStrategy. 4. Removed duplicate scalaJava8Compat dependency in ce-sbk. 5. Removed commonsCli dependency from ce-sbk since the mentioned issue should not matter since the `core` module does it already. 6. Remove log4j dependency from ce-sbk since it has no direct references to it. 7. Remove failureaccess dependency from ce-broker-plugins since we don't access is directly. 8. Use `compile` for all library dependencies in ce-sbk as there is no benefit in using `compileOnly`. 9. Force a few more library versions in `resolutionStrategy` to fix issues with multiple jars with diverging versions in the distribution. Reviewers: Xavier Léauté <xavier@confluent.io>
- 8月 20, 2020
-
-
由 Yang Yu 创作于
This PR adds new storage faults to Ducktape system tests. new faults in Kibosh service: -- unwritable: write calls will return preset exception on specified directories and files -- write_delay: delay the write calls for a specific time -- read_corrupt: read calls will return corrupted data on specified directories and files -- write_corrupt: write calls will write corrupted data to disk on specified directories and files. new storage failures: -- power failure: Kafka process is killed abruptly and the last write to data log is corrupted. -- high disk latency: delays are added to write and read calls. -- disk failure: write and read calls on data logs return I/O error. -- lost volume: write and read calls on data logs return device not found error. -- permission deny: write and read calls on data logs return permission denied error. Kafka service is modified to include option for starting a Kafka server on Kibosh filesystem.
-
由 Matthew Wong 创作于
We were seeing NPE in our GCP Tier Soak cluster. The TierDeletedPartitionsCoordinator was reading a tier topic partition when the partition was emigrated due to a rebalance. The log was marked for deletion after 1 minute, but the coordinator was still reading the log for longer than that minute. The log was fully deleted, causing the read to throw an NPE. This seems to only occur when the tier topic is very large and would take a long time to fully read. This PR adds a check to stop reading when the partition is detected to be emigrated. There is an associated test that fails w/o the check and passes w/ the check. Also, the PR adds to the kafka api for handling StopReplicaRequests, notifying the coordinator when the request is related to the tier topic.
-
由 Rigel Bezerra de Melo 创作于
-
由 Kowshik Prakasam 创作于
-
- 8月 19, 2020
-
-
由 Stanislav Kozlovski 创作于
This cruise control metric sampler is not needed anymore - we default to and only support using the TelemetryReporterSampler
-
由 Aneel Nazareth 创作于
-
由 Rajini Sivaram 创作于
Reviewers: Ryan Pridgeon
- 8月 18, 2020
-
-
由 Rajini Sivaram 创作于
-
由 Ismael Juma 创作于
`common` set the version at 2.9.9. Took the chance to fix error-prone-annotations duplicate version issue. Reviewers: Srini Dandu <sdandu@confluent.io>
-
由 Rajini Sivaram 创作于
-
由 Rajini Sivaram 创作于
Conflicts: Jenkinsfile bin/kafka-run-class.sh build.gradle ci.py clients/src/main/java/org/apache/kafka/common/protocol/Errors.java core/src/main/scala/kafka/controller/KafkaController.scala core/src/main/scala/kafka/log/Log.scala core/src/main/scala/kafka/server/ReplicaManager.scala core/src/test/scala/unit/kafka/controller/ControllerIntegrationTest.scala core/src/test/scala/unit/kafka/server/KafkaApisTest.scala gradle.properties gradle/dependencies.gradle settings.gradle streams/quickstart/java/pom.xml streams/quickstart/java/src/main/resources/archetype-resources/pom.xml streams/quickstart/pom.xml tests/kafkatest/__init__.py tests/kafkatest/utils/util.py Additional updates to fix build: checkstyle/suppressions.xml clients/src/main/java/org/apache/kafka/common/metrics/Sensor.java ce-broker-plugins/src/main/java/io/confluent/kafka/multitenant/metrics/PartitionSensors.java
-
由 Andrew Egelhofer 创作于
ducktape diff: https://github.com/confluentinc/ducktape/compare/v0.7.8...v0.7.9 - bcrypt (a dependency of ducktape) dropped Python2.7 support. ducktape-0.7.9 now pins bcrypt to a Python2.7-supported version. Author: Andrew Egelhofer <aegelhofer@confluent.io> Reviewers: Dhruvil Shah <dhruvil@confluent.io>, Manikumar Reddy <manikumar.reddy@gmail.com> Closes #9192 from andrewegel/trunk
-
由 Matthew Wong 创作于
Adding the GCS implementation for Tiered Storage picked the guava dependency of 28.0-android. To align with the version of common repo, we are switching to the version 28.1-jre
-
由 Ismael Juma 创作于
The validator that relied on ZK was removed previously, so there is no reason to keep the dependency (which was causing Netty divergence issues). Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>
-
- 8月 17, 2020
-
-
-
由 Rajini Sivaram 创作于
Reviewers: Manikumar Reddy
-
由 Rajini Sivaram 创作于
Adds cluster link destination metrics in the cluster link interceptor consistent with tenant metrics in the broker interceptor. For cluster links, request metrics refer to egress and response metrics are ingress. All metrics are in `cluster-link-dest-tenant-metrics` group and the metric names are consistent with tenant-metrics. Error metrics are currently not generated for cluster links at tenant level. Reviewers: Sanjana Kaundinya
-