confluent-kafka-dotnet icon indicating copy to clipboard operation
confluent-kafka-dotnet copied to clipboard

Published 20 hours ago • verified publisher icon confluentinc

Violating partition assignment

Open ksdvishnukumar opened this issue 10 months ago • 1 comments

Description

I came across the below scenario in one of environment. I have a Topic (EventHub) with 3 partitions. I spanned up 3 consumers ideally each consumer should be assigned with one partition each. But from the PartitionAssignmenedHandler I see that, one consumer(Consumer1) took all the 3 partitions. But I observed that, Consumer2 received the partition 1 data even though there is no assignment to Consumer2 (Via PartitionAssignmenedHandler) and No revoke/loss assignment to Consumer1 (SetPartitionsRevokedHandler/SetPartitionsLostHandler). The core of the partition distribution itself violating which leads to duplicate message receiving. I understand that there is no rebalancing happened in this case and also no error/exception occurred.

Consumers are deployed to AKS and consumer are connecting to Azure Eventhub using Confluent Kafka Package (2.1.1)

How to reproduce

Could not be able to reproduce in the lower environment as this issue happened in production so there is no way to get the Debug logs.

Checklist

Please provide the following information:

  • [x] A complete (i.e. we can run it), minimal program demonstrating the problem. No need to supply a project file. There is no specific code could reproduce it

  • [x] Confluent.Kafka nuget version. (2.1.1)

  • [x] Apache Kafka version. - Using Azure Eventhub

  • [x] Client configuration. AutoOffsetReset=Earliest EnableAutoOffsetStore = true EnableAutoCommit = false AutoCommitIntervalMs = 10Seconds EnablePartitionEof = true ConnectionsMaxIdleMs = 3 Min PartitionAssignmentStrategy = PartitionAssignmentStrategy.CooperativeSticky IsolationLevel = IsolationLevel.ReadUncommitted SocketNagleDisable = true SocketKeepaliveEnable = true MetadataMaxAgeMs = 180000 SessionTimeoutMs = 30000 MaxPollIntervalMs = 300000 CancellationDelayMaxMs = 200,

  • [x] Operating system. Ubuntu

  • [x] Provide logs (with "debug" : "..." as necessary in configuration). As its happened in production no way to enable debug as it generates plenty of logs and also not able to reproduce it in lower environment

  • [x] Provide broker log log-excerpts.csv.

  • [x] Critical issue. Yes

ksdvishnukumar avatar Apr 01 '24 13:04 ksdvishnukumar

Hi @mhowlett , @edenhill

Is there any workaround on this can be implemented based on the Consumer configuration?

ksdvishnukumar avatar Apr 11 '24 12:04 ksdvishnukumar