- 10月 15, 2020
-
-
由 Manikumar Reddy 创作于
- 10月 14, 2020
-
-
由 Lucas Bradstreet 创作于
The recursive chmod adds minutes to startup time in clusters with a lot of segments (20K+). I believe it is unnecessary, however I have been unable to find the reason why it was added in https://github.com/confluentinc/ce-kafka/pull/4/files. I've tested an image with this change by creating a new cluster and then rolling it and encountered no issues doing so. If there were prior permission problems in the past then I think they would have been resolved by prior chmod -R runs after restarts. Prior discussion: https://confluent.slack.com/archives/C016HMTHR60/p1594373262091800 https://confluent.slack.com/archives/C013VERBRV1/p1589837983024200 https://confluent.slack.com/archives/CRPN7CQGY/p1576325410015100
-
由 Cheng Tan 创作于
Make fetch.max.bytes dynamic and expose to cloud users
-
由 Ismael Juma 创作于
Both are now downloaded by the base image. Replace hardcoded prometheus jmx exporter version with environment variable reference. Other noteworthy changes in the new base image: * JDK updated to 14.0.2 * Async profiler updated to 1.8.1 * Prometheus jmx-exporter updated to 0.14 which includes measurable performance improvements when the number of JMX metrics is large. See https://github.com/confluentinc/cc-docker-base/pull/57 for more details. * Update CP from 6.0.0-SNAPSHOT to 6.1.0-SNAPSHOT. Full diff: https://github.com/confluentinc/cc-docker-base/compare/v5.1.0...v7.4.0 Reviewers: Lucas Bradstreet <lucasbradstreet@gmail.com>, Xavier Léauté <xavier@confluent.io>
-
由 David Mao 创作于
Fix quota auto-tuning to use a non-volatile value while sorting. Reduce garbage allocated in maybeAutoTuneQuota Adds a jmh benchmark for ClientQuotaManager.maybeAutoTune
-
由 Vikas Singh 创作于
* CNK-374: Return task history for only configured retention period This change adds a configuration to only return task history for a configurable period of time. The history past that time is not returned. The history older to that may or may not get deleted, but that is implementation defined. Added a new test to test correct history is returned.
-
由 Brian Bushree 创作于
-
由 Rajini Sivaram 创作于
Reviewers: Brian Byrne
- 10月 13, 2020
-
-
由 Brian Bushree 创作于
METRICS-2358: only initialize event logger when enabled, properly close it, and add better support for reconfiguration (#2644)
-
由 David Mao 创作于
Broker-level config changes are processed asynchronously. This PR adds a wait with timeout in disableAutoTopicCreation to ensure that config changes are propagated before returning from the function.
-
由 Vikas Singh 创作于
* CNKAF-1134: Close AdminClient, Producers and Consumers immediately This change update SBK code to close all AdminClient, Producers and Consumers it is using immediately on shutdown. Not doing so spams log with follwoing messages at shutdown while we wait for graceful termination of these clients: [2020-08-26 15:31:59,341] WARN [AdminClient clientId=kafka-cruise-control] Connection to node 0 (/192.168.1.14:9092) could not be established. Broker may not be available. (org.apache.kafka.clients.NetworkClient) [2020-08-26 15:31:59,377] WARN [Consumer clientId=kafka-cruise-control, groupId=ConfluentTelemetryReporterSampler--8352956737911420999] Connection to node 0 (/192.168.1.14:9092) could not be established. Broker may not be available. (org.apache.kafka.clients.NetworkClient) [2020-08-26 15:31:59,405] WARN [Producer clientId=ConfluentBalancerSampleStoreProducer] Connection to node 0 (/192.168.1.14:9092) could not be established. Broker may not be available. (org.apache.kafka.clients.NetworkClient) In addition there was a place where the AdminClient was not getting closed, added code to close that too.
- 10月 12, 2020
-
-
由 Rajini Sivaram 创作于
We track first failure from the metadata writer for each partition and reset the failure when the partition is back in a good state with an active writer. At the moment, active writers reset failure status, but followers don't. If a follower with a failure becomes active writer after more than a day, it marks partitions as failed until the new writer is successfully initialized. This may cause unexpected authorization errors in clients. This commit resets failure status when either local or remote writer is successfully initialized to ensure we don't fail authorization unless no writer is successful throughout the retry timeout. Reviewers: Manikumar Reddy
- 10月 10, 2020
-
-
由 Stanislav Kozlovski 创作于
This patch implements the Admin API and DescribeBrokerAdditions RPC for the add broker command as per https://confluentinc.atlassian.net/wiki/spaces/CNKAF/pages/1549730493/SBK+Add+Broker+API+-+Kafka+1-Pager It most notably adds the DescribeBrokerAdditions Admin API and DescribeBrokerAdditions{Request,Response} RPC. It also introduces a new high-level status called BalancerOperationStatus to the broker additions RPC, soon to follow in the remove brokers RPC
-
由 Agam Brahma 创作于
* KC-938: Add config for dedicated-cluster topics for cluster-linking soak setup. Reference: https://confluentinc.atlassian.net/wiki/spaces/K/pages/1402316523/Cluster+Linking+Soak+Cluster+One-Pager * Bumped chart version * Added more soak-cluster topics * Remove D2, MT mirrored load * Bumped version Co-authored-by:
Agam Brahma <abrahma@confluent.io>
- 10月 09, 2020
-
-
由 Brian Bushree 创作于
-
由 Brian Byrne 创作于
-
由 Aishwarya Gune 创作于
* add check in validation layer for rf equal to alive brokers
-
- 10月 08, 2020
-
-
由 Stanislav Kozlovski 创作于
This patch revises the broker removal state machine by making it more extendable, so as to accommodate the soon-to-come broker addition state machine. Most notably, the responsibility for defining the possible state transitions is now done by a new class called StateTransitioner through a declarative builder interface. Previously, we would define the valid state transitions inside the states themselves, which was harder to grasp and was mixing responsibilities. The StateTransitioner is made generic to accept a state and event that operates on said state, such that the soon-to-come broker addition state machine could easily leverage that transitioner. Additionally, we generalize the BrokerRemovalCallback into a BalancerOperationCallback and we move the BrokerRemovalEvent from the cruisecontrol.brokerremoval package to an inner class of balancer.operation.BrokerRemovalStateMachine so as to colocate the state transitions and events.
-
由 Aneel Nazareth 创作于
-
由 Alok Thatikunta 创作于
* Modify offset cache min * Add test * Update tests * Fix build * PR comments * PR comments * PR comments * nit
- 10月 07, 2020
-
-
由 Manikumar Reddy 创作于
-
由 Alok Thatikunta 创作于
* Validate partitions * Add tests * PR comments * PR comments