[
https://issues.apache.org/jira/browse/FLINK-14400?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Andrey Zagrebin resolved FLINK-14400.
-------------------------------------
Fix Version/s: 1.10.0
Release Note: merge into master by 9fed0ddc5bc015f98246a2d8d9adbe5fb2b91ba4
Resolution: Done
> Shrink the scope of MemoryManager from TaskExecutor to slot
> ------------------------------------------------------------
>
> Key: FLINK-14400
> URL: https://issues.apache.org/jira/browse/FLINK-14400
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / Coordination, Runtime / Task
> Reporter: Andrey Zagrebin
> Assignee: Andrey Zagrebin
> Priority: Major
> Labels: pull-request-available
> Fix For: 1.10.0
>
> Time Spent: 20m
> Remaining Estimate: 0h
>
> MemoryManager currently manages the memory bookkeeping for all slots/tasks
> inside one TaskExecutor. For better abstraction and isolation of slots, we
> can shrink its scope and make it per slot. The memory limits are fixed now
> per slot at the moment of slot creation. All operators, sharing the slot,
> will get their fixed fractional limits.
> In future, we might make it possible for operators to over-allocate beyond
> their fraction limit if there is some available free memory in the slot but
> it should be possible to reclaim the over-allocated memory at any time if
> other operator decides to claim its fair share within its limit.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
