[GitHub] [kafka] wenzelcheng closed pull request #11360: Floating point number comparison problem
wenzelcheng closed pull request #11360: URL: https://github.com/apache/kafka/pull/11360 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: jira-unsubscr...@kafka.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [kafka] lbradstreet opened a new pull request #11376: KAFKA-13342: LeaderAndIsrRequest should not be sent for topic queued for deletion
lbradstreet opened a new pull request #11376: URL: https://github.com/apache/kafka/pull/11376 In some cases a broker may be lost during a topic deletion and before its replica has moved to OfflineReplica state. When the broker comes back up the controller will send it a LeaderAndIsrRequest containing the partition even though it is already in the deleting state in the controller. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: jira-unsubscr...@kafka.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[jira] [Commented] (KAFKA-13331) Slow reassignments in 2.8 because of large number of UpdateMetadataResponseReceived(UpdateMetadataResponseData(errorCode=0),) Events
[ https://issues.apache.org/jira/browse/KAFKA-13331?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17423761#comment-17423761 ] GEORGE LI commented on KAFKA-13331: --- besides slowness of the reassignments, when the EventQueue is long, with a lot of UpdateMetadataRequestResponse events. The cluster is also not behaving correctly (slow / timeout with some RPC calls). e.g. using {{kafka-topics.sh --bootstrap-server}} will time out like below. because it puts {{listPartitionReassignments}} in the EventQueue. The old way {{kafka-topics.sh --zookeeper}} returned fast. {code} $ /usr/lib/kafka/bin/kafka-topics.sh --bootstrap-server :9092 --topic --describe Error while executing topic command : Call(callName=listPartitionReassignments, deadlineMs=1632339660352, tries=1, nextAllowedTryMs=1632339660520) timed out at 1632339660420 after 1 attempt(s) [2021-09-22 19:41:00,425] ERROR org.apache.kafka.common.errors.TimeoutException: Call(callName=listPartitionReassignments, deadlineMs=1632339660352, tries=1, nextAllowedTryMs=1632339660520) timed out at 1632339660420 after 1 attempt(s) Caused by: org.apache.kafka.common.errors.DisconnectException: Cancelled listPartitionReassignments request with correlation id 9 due to node 10043 being disconnected (kafka.admin.TopicCommand$) {code} Any RPC calls putting into the EventQueue will be affected during the reassignments (batch size 50+) in our environment. > Slow reassignments in 2.8 because of large number of > UpdateMetadataResponseReceived(UpdateMetadataResponseData(errorCode=0),) > Events > > > Key: KAFKA-13331 > URL: https://issues.apache.org/jira/browse/KAFKA-13331 > Project: Kafka > Issue Type: Bug > Components: admin, controller, core >Affects Versions: 2.8.1, 3.0.0 >Reporter: GEORGE LI >Priority: Critical > Fix For: 3.1.0 > > Attachments: Screen Shot 2021-09-28 at 12.57.34 PM.png > > > Slowness is observed when doing reassignments on clusters with more brokers > (e.g. 80 brokers). > After investigation, it looks like the slowness is because for > reassignments, it sends the UpdateMetadataRequest to all the broker for every > topic partition affected by the reassignment (maybe some optimization can be > done). e.g. > for a reassignment with batch size of 50 partitions. it will generate about > 10k - 20k ControllerEventManager.EventQueueSize and the p99 EventQueueTimeMs > will be 1M. if the batch size is 100 partitions, about 40K > EventQeuueSize and 3M p99 EventQueueTimeMs. See below screen shot on the > metrics. > !Screen Shot 2021-09-28 at 12.57.34 PM.png! > it takes about 10-30minutes to process 100 reassignments per batch. and > 20-30 seconds for 1 reassignment per batch even the topic partition is almost > empty. Version 1.1, the reassignment is almost instant. > Looking at what is in the ControllerEventManager.EventQueue, the majority > (depends on the how many brokers in the cluster, it can be 90%+) is > {{UpdateMetadataResponseReceived(UpdateMetadataResponseData(errorCode=0),)}} > events. which is introduced in this commit: > {code} > commit 4e431246c31170a7f632da8edfdb9cf4f882f6ef > Author: Jason Gustafson > Date: Thu Nov 21 07:41:29 2019 -0800 > MINOR: Controller should log UpdateMetadata response errors (#7717) > > Create a controller event for handling UpdateMetadata responses and log a > message when a response contains an error. > > Reviewers: Stanislav Kozlovski , Ismael > Juma > {code} > Checking how the events in the ControllerEventManager are processed for > {{UpdateMetadata response}}, it seems like it's only checking whether there > is an error, and simply log the error. but it takes about 3ms - 60ms to > dequeue each event. Because it's a FIFO queue, other events were waiting in > the queue. > {code} > private def processUpdateMetadataResponseReceived(updateMetadataResponse: > UpdateMetadataResponse, brokerId: Int): Unit = { > if (!isActive) return > if (updateMetadataResponse.error != Errors.NONE) { > stateChangeLogger.error(s"Received error > ${updateMetadataResponse.error} in UpdateMetadata " + > s"response $updateMetadataResponse from broker $brokerId") > } > } > {code} > There might be more sophisticated logic for handling the UpdateMetadata > response error in the future. For current version, would it be better to > check whether the response error code is Errors.NONE before putting into > the Event Queue? e.g. I put this additional check and see the Reassignment > Performance increase dramatically on the 80 brokers cluster. > {code} > val up
[jira] [Commented] (KAFKA-13342) LISR sent for topic queued for deletion in controller
[ https://issues.apache.org/jira/browse/KAFKA-13342?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17423756#comment-17423756 ] Lucas Bradstreet commented on KAFKA-13342: -- Under some scenarios I think this could cause partition dirs to be recreated. > LISR sent for topic queued for deletion in controller > - > > Key: KAFKA-13342 > URL: https://issues.apache.org/jira/browse/KAFKA-13342 > Project: Kafka > Issue Type: Bug >Reporter: Lucas Bradstreet >Priority: Minor > > Under certain conditions in some system tests a broker will be hard killed > during a topic deletion and before its replica has moved to OfflineReplica > state. When the broker comes back up the controller will send it a > LeaderAndIsrRequest containing the partition causing it to recreate the > partition locally even though it is in deleting state in the controller. -- This message was sent by Atlassian Jira (v8.3.4#803005)
[jira] [Updated] (KAFKA-13342) LISR sent for topic queued for deletion in controller
[ https://issues.apache.org/jira/browse/KAFKA-13342?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Lucas Bradstreet updated KAFKA-13342: - Priority: Minor (was: Major) > LISR sent for topic queued for deletion in controller > - > > Key: KAFKA-13342 > URL: https://issues.apache.org/jira/browse/KAFKA-13342 > Project: Kafka > Issue Type: Bug >Reporter: Lucas Bradstreet >Priority: Minor > > Under certain conditions in some system tests a broker will be hard killed > during a topic deletion and before its replica has moved to OfflineReplica > state. When the broker comes back up the controller will send it a > LeaderAndIsrRequest containing the partition causing it to recreate the > partition locally even though it is in deleting state in the controller. -- This message was sent by Atlassian Jira (v8.3.4#803005)
[jira] [Created] (KAFKA-13342) LISR sent for topic queued for deletion in controller
Lucas Bradstreet created KAFKA-13342: Summary: LISR sent for topic queued for deletion in controller Key: KAFKA-13342 URL: https://issues.apache.org/jira/browse/KAFKA-13342 Project: Kafka Issue Type: Bug Reporter: Lucas Bradstreet Under certain conditions in some system tests a broker will be hard killed during a topic deletion and before its replica has moved to OfflineReplica state. When the broker comes back up the controller will send it a LeaderAndIsrRequest containing the partition causing it to recreate the partition locally even though it is in deleting state in the controller. -- This message was sent by Atlassian Jira (v8.3.4#803005)
[jira] [Created] (KAFKA-13341) Quotas are not applied to requests with null clientId
David Mao created KAFKA-13341: - Summary: Quotas are not applied to requests with null clientId Key: KAFKA-13341 URL: https://issues.apache.org/jira/browse/KAFKA-13341 Project: Kafka Issue Type: Bug Reporter: David Mao ClientQuotaManager.DefaultQuotaCallback will not check for the existence of a default quota if a request's clientId is null. This results in null clientIds bypassing quotas. Null clientIds are permitted in the protocol, so this seems like a bug. This looks like it may be a regression introduced by https://github.com/apache/kafka/pull/7372 -- This message was sent by Atlassian Jira (v8.3.4#803005)
[jira] [Updated] (KAFKA-13341) Quotas are not applied to requests with null clientId
[ https://issues.apache.org/jira/browse/KAFKA-13341?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] David Mao updated KAFKA-13341: -- Priority: Major (was: Minor) > Quotas are not applied to requests with null clientId > - > > Key: KAFKA-13341 > URL: https://issues.apache.org/jira/browse/KAFKA-13341 > Project: Kafka > Issue Type: Bug >Reporter: David Mao >Priority: Major > > ClientQuotaManager.DefaultQuotaCallback will not check for the existence of a > default quota if a request's clientId is null. This results in null clientIds > bypassing quotas. > Null clientIds are permitted in the protocol, so this seems like a bug. > This looks like it may be a regression introduced by > https://github.com/apache/kafka/pull/7372 -- This message was sent by Atlassian Jira (v8.3.4#803005)
[GitHub] [kafka] satishd commented on pull request #11058: KAFKA-12802 Added a file based cache for consumed remote log metadata for each partition to avoid consuming again incase of broker restarts.
satishd commented on pull request #11058: URL: https://github.com/apache/kafka/pull/11058#issuecomment-933099724 Thanks @junrao for the review comments. Addressed them with the latest commit and replies. Added a scenario to verify checkpointed offsets in the [test](https://github.com/apache/kafka/pull/11058/commits/554df8c5e58f5dc14b5d1a3476f011184116a088#diff-8c57d1a1451531841bccd4de7f38b838cfa8444e0257c5097480e92e3e0fe72bR145). -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: jira-unsubscr...@kafka.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [kafka] upsidedownsmile opened a new pull request #11375: KAFKA-10865: Log transformed record in WorkerSinkTask
upsidedownsmile opened a new pull request #11375: URL: https://github.com/apache/kafka/pull/11375 This PR adds a log message with the **topic**, **key** and **value** to the transformed record in **WorkerSinkTask** similar to was already being done in **WorkerSinkTask**. Also fixed a small typo in javadocs where there was a double period. ### Committer Checklist (excluded from commit message) - [ ] Verify design and implementation - [ ] Verify test coverage and CI build status - [ ] Verify documentation (including upgrade notes) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: jira-unsubscr...@kafka.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [kafka] wenzelcheng closed pull request #11360: Floating point number comparison problem
wenzelcheng closed pull request #11360: URL: https://github.com/apache/kafka/pull/11360 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: jira-unsubscr...@kafka.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org