[
https://issues.apache.org/jira/browse/FLINK-22937?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17360050#comment-17360050
]
Piers commented on FLINK-22937:
-------------------------------
[~yunta]
[^dump.txt]
This is dump file.
And, I found something. Below is code, this is minimal code to reproduce it.
[https://github.com/mofadeyunduo/flink-problem
|https://github.com/mofadeyunduo/flink-problem]
This code can reproduce crash problem, but not same error when i execute on
remote k8s. You local machined must have kafka on 9092 and redis on 6379 and
excute Main method in com.pi.proxy.job.alert.Job. This is simplified for
debuging problem.
Condition (i guess and test) may be trigger problem:
* sink function collect objects that inherit abstract based class, and this
class is implemented as inner class, such as PositiveAlertResult in this below
project
* has many instances to sink, and when sink, i use jedis client, which should
be blocked due to io(i guess)
And on my k8s, i can reproduce it. You can change code to getExecuteEnviroment,
and set backend to rocksdb, and submit to k8s. But you should build data to
kafka yourself, and make as large as you can, will help trigger this problem.
Thanks and sorry for my pool english.
> rocksdb cause jvm to crash
> --------------------------
>
> Key: FLINK-22937
> URL: https://issues.apache.org/jira/browse/FLINK-22937
> Project: Flink
> Issue Type: Bug
> Components: Runtime / State Backends
> Affects Versions: 1.13.1
> Environment: deployment: native kubernates
> Reporter: Piers
> Priority: Major
> Attachments: dump.txt, hs_err_pid1.log
>
>
> JVM crash when running job. Possibly RocksDB caused this.
> This link containers JVM crash log.
> Thanks.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
