[
https://issues.apache.org/jira/browse/FLINK-21504?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Roman Khachatryan updated FLINK-21504:
--------------------------------------
Description:
Goal: enable TMs to discard state after subsumption (FLINK-23139).
Proposed solution: include earliest non-subsumed checkpoint ID into checkpoint
confirmation notification.
was:
As discussed in FLIP-151
[thread|https://lists.apache.org/thread.html/r0b90330094ef356997399e4b657e42e6b09fdd731d3069d8bb9841d9%40%3Cdev.flink.apache.org%3E]
both FLIP-151 and FLIP-158 would benefit from adding latest confirmed
checkpoint ID to checkpoint barriers and RPC calls. More specifically, knowing
the latest confirmed checkpoint at the time of performing a new one would allow
to:
* avoid otherwise unnecessary re-uploading of state changes
* cleanup state changes uploaded to DFS and sent to JM but not used by the JM
Besides that, adding earliest non-subsumed checkpoint ID would be another way
to prevent FLINK-21351.
cc: [~sewen]
> State ownership: notify TMs about checkpoint subsumption
> --------------------------------------------------------
>
> Key: FLINK-21504
> URL: https://issues.apache.org/jira/browse/FLINK-21504
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / Checkpointing
> Reporter: Roman Khachatryan
> Priority: Major
> Labels: auto-unassigned
>
> Goal: enable TMs to discard state after subsumption (FLINK-23139).
> Proposed solution: include earliest non-subsumed checkpoint ID into
> checkpoint confirmation notification.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
