[jira] [Updated] (FLINK-26992) Passing directly between threads PojoSerializer may cause ConcurrentModificationException

["Piotr Nowojski (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/FLINK-26992?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Piotr Nowojski updated FLINK-26992:
-----------------------------------
    Summary: Passing directly between threads PojoSerializer may cause ConcurrentModificationException  (was: PojoSerializer may cause concurrent exception passing directly between threads)

> Passing directly between threads PojoSerializer may cause ConcurrentModificationException
> -----------------------------------------------------------------------------------------
>
>                 Key: FLINK-26992
>                 URL: https://issues.apache.org/jira/browse/FLINK-26992
>             Project: Flink
>          Issue Type: Bug
>          Components: Runtime / State Backends
>    Affects Versions: 1.16.0
>            Reporter: Yuan Mei
>            Priority: Major
>
> Extract from FLINK-26835
> While investigating this issue, we have found that probably state backends are also using non-thread safe serialisers from different threads.
> For example: {{RocksFullSnapshotStrategy#syncPrepareResources}} is passing {{keySerializer}} from the task thread, to the async thread in order to serialize the serializer itself. {{RocksIncrementalSnapshotStrategy.RocksDBIncrementalSnapshotOperation#materializeMetaData}} seems to be doing the same thing. If {{PojoSerializer}} is used as {{keySerializer}} I think this will lead to the same problems as above. Iterating through the {{PojoSerializer#subclassSerializerCache}} from the the async checkpoint thread, while the map can be changed from the task thread. It looks like in all of those places the serializer should have been duplicated ({{{}#duplicate{}}}) before being passed to another thread. Maybe this should happen in {{{}RocksDBSnapshotStrategyBase{}}}. I don't know about other state backends.
>  
> =======
> TODO:
> I second that each thread obtains ownership of the Serializer passed in itself.
>  * Figure out whether the current way of passing Serializer really causing problems in the state backend (concurrent modification possible).
>  * What other places have Serializer directly passed between threads.



--
This message was sent by Atlassian Jira
(v8.20.1#820001)