[
https://issues.apache.org/jira/browse/KAFKA-20295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18094901#comment-18094901
]
Roland Sommer commented on KAFKA-20295:
---------------------------------------
If I open the (only) log-file, I run into
{code:java}
[2026-07-09 06:09:34,218] ERROR Encountered shell fault: Unhandled fault in
MetadataLoader#handleCommit. Last image offset was 217548874
(org.apache.kafka.server.fault.LoggingFaultHandler)
java.lang.IllegalStateException: Unknown metadata version for FeaturesImage:
metadataVersion(Optional.empty)
at
org.apache.kafka.image.FeaturesImage.lambda$metadataVersionOrThrow$0(FeaturesImage.java:73)
at java.base/java.util.Optional.orElseThrow(Optional.java:403)
at
org.apache.kafka.image.FeaturesImage.metadataVersionOrThrow(FeaturesImage.java:72)
at
org.apache.kafka.image.loader.MetadataLoader.maybePublishMetadata(MetadataLoader.java:372)
at
org.apache.kafka.image.loader.MetadataBatchLoader.applyDeltaAndUpdate(MetadataBatchLoader.java:285)
at
org.apache.kafka.image.loader.MetadataBatchLoader.maybeFlushBatches(MetadataBatchLoader.java:221)
at
org.apache.kafka.image.loader.MetadataLoader.lambda$handleCommit$0(MetadataLoader.java:406)
at
org.apache.kafka.queue.KafkaEventQueue$EventContext.run(KafkaEventQueue.java:134)
at
org.apache.kafka.queue.KafkaEventQueue$EventHandler.handleEvents(KafkaEventQueue.java:217)
at
org.apache.kafka.queue.KafkaEventQueue$EventHandler.run(KafkaEventQueue.java:188)
at java.base/java.lang.Thread.run(Thread.java:1583)
[2026-07-09 06:09:34,305] ERROR Encountered shell fault: Unhandled error
initializing new publishers (org.apache.kafka.server.fault.LoggingFaultHandler)
java.lang.IllegalStateException: Unknown metadata version for FeaturesImage:
metadataVersion(Optional.empty)
at
org.apache.kafka.image.FeaturesImage.lambda$metadataVersionOrThrow$0(FeaturesImage.java:73)
at java.base/java.util.Optional.orElseThrow(Optional.java:403)
at
org.apache.kafka.image.FeaturesImage.metadataVersionOrThrow(FeaturesImage.java:72)
at
org.apache.kafka.image.loader.MetadataLoader.initializeNewPublishers(MetadataLoader.java:318)
at
org.apache.kafka.image.loader.MetadataLoader.lambda$scheduleInitializeNewPublishers$0(MetadataLoader.java:288)
at
org.apache.kafka.queue.KafkaEventQueue$EventContext.run(KafkaEventQueue.java:134)
at
org.apache.kafka.queue.KafkaEventQueue$EventHandler.handleEvents(KafkaEventQueue.java:217)
at
org.apache.kafka.queue.KafkaEventQueue$EventHandler.run(KafkaEventQueue.java:188)
at java.base/java.lang.Thread.run(Thread.java:1583)
{code}
and if I try to list the existing controllers afterwards, I just get an empty
result:
{code:java}
>> ls image/cluster/controllers/
>>
{code}
> Removed controllers still in metadata, blocking finalizing upgrade to 4.2.0
> ---------------------------------------------------------------------------
>
> Key: KAFKA-20295
> URL: https://issues.apache.org/jira/browse/KAFKA-20295
> Project: Kafka
> Issue Type: Bug
> Components: controller
> Environment: Kafka 4.2.0 (Scala 2.13) running on Debian Trixie 13.3
> Reporter: Roland Sommer
> Priority: Major
>
> While upgrading our kafka clusters to new operating systems I switched to
> dynamic voter configuration and removed controller instances with
> {{/opt/kafka/bin/kafka-metadata-quorum.sh}} and the {{remove-controller}}
> subcommand. Inspecting the cluster with {{describe}} only shows the actual
> running nodes.
> Now during the update to 4.2.0, the final metadata upgrade step complains
> about
> {code:java}
> Could not upgrade eligible.leader.replicas.version to 1. The update failed
> for all features since the following feature had an error: Invalid update
> version 29 for feature metadata.version. Controller 351 only supports
> versions 7-27{code}
> with 351 being an ID of an already removed controller. Inspecting a snapshot
> with {{/opt/kafka/bin/kafka-metadata-shell.sh}} indeed shows all controller
> ids of already removed controllers:
> {code:java}
> >> ls image/cluster/controllers/
> 158 206 351 584 611 686 {code}
> while other tools only show the expected nodes:
> {code:java}
> ~$ /opt/kafka/bin/kafka-metadata-quorum.sh --bootstrap-controller
> localhost:9093 describe --replication --human-readable
> NodeId DirectoryId LogEndOffset Lag LastFetchTimestamp LastCaughtUpTimestamp
> Status
> 158 2gsvOvnT7urpZcA_-LUy5w 196823524 0 7 ms ago 8 ms ago Leader
> 611 27Ii-xdAZ7ReQBLsvvJb0A 196823524 0 348 ms ago 348 ms ago Follower
> 206 Q7X9o3XbKxk_3tz4T8torg 196823524 0 348 ms ago 348 ms ago Follower
> 226 7n6aedUEuytkqhBnbe7ESw 196823524 0 348 ms ago 348 ms ago Observer
> 181 tZ17VQ8cYpf7R-LyAQWf2w 196823524 0 349 ms ago 349 ms ago Observer
> 299 P4qXt3K0G5Qg_7w_UdvaNA 196823524 0 348 ms ago 348 ms ago Observer
> 290 bA0pqZFsUa45lRTB6bS4bg 196823524 0 348 ms ago 348 ms ago Observer
> 293 Av_12222lURKVYVt-aNKOQ 196823524 0 348 ms ago 348 ms ago Observer
> 485 glENIgkIng1MYDF8HxxoDQ 196823524 0 349 ms ago 350 ms ago Observer {code}
> Grepping through {{bin/kafka-dump-log.sh --cluster-metadata-decoder}} only
> shows the expected three {{REGISTER_CONTROLLER_RECORD}} entries.
> Is there any clear path for removing those stale nodes?
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
