[
https://issues.apache.org/jira/browse/KAFKA-12818?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17348590#comment-17348590
]
Tom Bentley commented on KAFKA-12818:
-------------------------------------
Since your limit > request you're going to be at risk of the OOMKiller anyway,
whether or not there's a memory leak. You were exceeding your request even with
the FileConfigProvider.
I find it hard to believe the DirectoryConfigProvider itself is responsible for
somehow using _that much_ (~200m) memory above what we see in the first curve.
It's normally only used once when the config file is parsed and the values
substituted. It doesn't watch the files (even if they do change). So I think we
need better memory usage diagnostics here than just a graph of RSS.
> Memory leakage when kafka connect 2.7 uses directory config provider
> --------------------------------------------------------------------
>
> Key: KAFKA-12818
> URL: https://issues.apache.org/jira/browse/KAFKA-12818
> Project: Kafka
> Issue Type: Bug
> Components: KafkaConnect
> Affects Versions: 2.7.0
> Environment: Azure AKS / Kubernetes v1.20
> Reporter: Viktor Utkin
> Priority: Critical
> Attachments: Screenshot 2021-05-20 at 14.53.05.png
>
>
> Hi, we noticed a Memory leakage problem when kafka connect 2.7 uses directory
> config provider. We've got an OOM in kubernetes environment. K8s kills
> container when limit reached. At same time we've not get any OOM in Java.
> Heap dump did't show us anything interesting.
> JVM config:
> {code:java}
> -XX:+HeapDumpOnOutOfMemoryError
> -XX:HeapDumpPath=/tmp/
> -XX:+UseContainerSupport
> -XX:+OptimizeStringConcat
> -XX:MaxRAMPercentage=75.0
> -XX:InitialRAMPercentage=50.0
> -XX:MaxMetaspaceSize=256M
> -XX:MaxDirectMemorySize=256M
> -XX:+UseStringDeduplication
> -XX:+AlwaysActAsServerClassMachine{code}
>
> Kafka Connect config:
> {code:java}
> "config.providers": "directory"
> "config.providers.directory.class":
> "org.apache.kafka.common.config.provider.DirectoryConfigProvider"{code}
>
> Kubernetes pod resources limits:
> {code:java}
> resources:
> requests:
> cpu: 1500m
> memory: 2Gi
> limits:
> cpu: 3000m
> memory: 3Gi
> {code}
>
> doker image used: confluentinc/cp-kafka-connect:6.1.1
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
