[
https://issues.apache.org/jira/browse/CAMEL-16914?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17408731#comment-17408731
]
Otavio Rodolfo Piske commented on CAMEL-16914:
----------------------------------------------
This ALWAYS happen in a block of 128 records: it works OK for 128 records, then
skip 128, then works OK for the subsequent 128 and so on.
After a a preliminary observation, I think this is being caused because the
code is assuming the key is always String:
https://github.com/apache/camel/blob/camel-3.11.x/components/camel-kafka/src/main/java/org/apache/camel/processor/idempotent/kafka/KafkaIdempotentRepository.java#L456
This seems to be messing up with the add/confirms/remove logic:
{code:java}
[1] 2021-09-02 10:42:51,230 DEBUG
[org.apa.cam.pro.ide.kaf.KafkaIdempotentRepository$TopicPoller] (Camel
(camel-1) thread #0 - KafkaIdempotentRepository) Adding to cache messageId:q
[1] 2021-09-02 10:42:52,229 DEBUG
[org.apa.cam.pro.ide.kaf.KafkaIdempotentRepository$TopicPoller] (Camel
(camel-1) thread #0 - KafkaIdempotentRepository) Adding to cache
messageId:r{code}
> camel-kafka: possible corruption of idempotency messages when using
> KafkaIdempotentRepository
> ----------------------------------------------------------------------------------------------
>
> Key: CAMEL-16914
> URL: https://issues.apache.org/jira/browse/CAMEL-16914
> Project: Camel
> Issue Type: Task
> Components: camel-kafka
> Affects Versions: 3.11.1
> Reporter: Otavio Rodolfo Piske
> Priority: Major
> Attachments: FakeDataProducer.java, MessageLossReproducer.java
>
>
> [^FakeDataProducer.java][^MessageLossReproducer.java]I noticed that when
> using the KafkaIdempotentRepository, sometimes it seems that it is losing or
> corrupting the ID information, causing loss of messages.
>
> When reading the topic used for publishing the idempotency data with kafkacat
> I get:
>
> {code:java}
> ERROR: Failed to format message in fake-data-inserts [1] at offset 39:
> expected key end-of-input, but 2 bytes remaining : terminating{code}
>
> As a result, this causes the loss of messages because the consumers cannot go
> further than the corrupted record.
>
> I found this while using Camel K and Strimzi (on a setup with 3 Kafkas using
> 1 topic with 3 partitions).
>
> I couldn't test w/ 3.12 yet, but I suspect it would also have the same
> problem.
>
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
