[
https://issues.apache.org/jira/browse/FLINK-7467?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17200824#comment-17200824
]
Nico Kruber commented on FLINK-7467:
------------------------------------
I'm not aware of any changes in this regard but was also not following any
updates. I will unassign myself since I'm not actually working on this.
In general, I don't think has too high priority though and may just be
theoretical / very rare.
> a slow client on the BlobServer may block writing operations
> ------------------------------------------------------------
>
> Key: FLINK-7467
> URL: https://issues.apache.org/jira/browse/FLINK-7467
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Coordination, Runtime / Network
> Affects Versions: 1.3.0, 1.3.1, 1.3.2, 1.4.0
> Reporter: Nico Kruber
> Assignee: Nico Kruber
> Priority: Major
>
> Since FLINK-6020, a locking mechanism was introduced to isolate reading from
> writing operations. This, however, will cause a slow client to slow down all
> write operations and may not be desirable at this level.
> The current usage patterns, however, first write all BLOBs (during job
> submission) and then read the needed ones when the TaskManagers need them.
> The slowdown may thus materialize only when multiple jobs are submitted
> concurrently and these phases overlap.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
