[
https://issues.apache.org/jira/browse/FLINK-17918?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17121532#comment-17121532
]
Yu Li commented on FLINK-17918:
-------------------------------
Thanks [~jark], I was about to add the same comments, that the observed
behavior of `CopyOnWriteStateMap` is by design.
Actually this condition (don't hold the reference and further modifying the
object beyond `processElement`) applies to all kinds of `StateTable` for
`HeapKeyedStateBackend` and is the source of performance gain (kept in object
format w/o serialization/deserialization, or any similar op like deep copy).
And a big +1 to explicitly write this down in our document.
> Blink Jobs are loosing data on recovery
> ---------------------------------------
>
> Key: FLINK-17918
> URL: https://issues.apache.org/jira/browse/FLINK-17918
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Checkpointing, Table SQL / Runtime
> Affects Versions: 1.11.0
> Reporter: Piotr Nowojski
> Priority: Blocker
> Fix For: 1.11.0
>
>
> After trying to enable unaligned checkpoints by default, a lot of Blink
> streaming SQL/Table API tests containing joins or set operations are throwing
> errors that are indicating we are loosing some data (full records, without
> deserialisation errors). Example errors:
> {noformat}
> [ERROR] Failures:
> [ERROR] JoinITCase.testFullJoinWithEqualPk:775 expected:<List(1,1, 2,2,
> 3,3, null,4, null,5)> but was:<List(2,2, 3,3, null,1, null,4, null,5)>
> [ERROR] JoinITCase.testStreamJoinWithSameRecord:391 expected:<List(1,1,1,1,
> 1,1,1,1, 2,2,2,2, 2,2,2,2, 3,3,3,3, 3,3,3,3, 4,4,4,4, 4,4,4,4, 5,5,5,5,
> 5,5,5,5)> but was:<List()>
> [ERROR] SemiAntiJoinStreamITCase.testAntiJoin:352 expected:<0> but was:<1>
> [ERROR] SetOperatorsITCase.testIntersect:55 expected:<MutableList(1,1,Hi,
> 2,2,Hello, 3,2,Hello world)> but was:<List()>
> [ERROR] JoinITCase.testJoinPushThroughJoin:1272 expected:<List(1,0,Hi,
> 2,1,Hello, 2,1,Hello world)> but was:<List(2,1,Hello, 2,1,Hello world)>
> {noformat}
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
