[
https://issues.apache.org/jira/browse/STORM-1952?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15370716#comment-15370716
]
Robert Joseph Evans commented on STORM-1952:
--------------------------------------------
There are two situations here.
1) The topology is still up and running, and for some reason it got rescheduled
some place else.
2) The topology is killed and will not ever come back.
This JIRA is trying to address the first one because when it happens it can
take a long time for the topology to come back up on a node that it was running
on previously (think about a rolling upgrade). This is why I suggested using
the dist cache code with LRU, because it will keep the jar/etc around longer,
but not forever, if the disk space is needed for something else.
[~kabhwan],
I think your comment is about the second situation and if so yes you are
correct about that too. If we do modify the topology specific download to use
the cache code it would be good to add an optimization to remove them from the
cache when we know it will never be used again. But that I see as an
optimization, not necessarily as a requirement.
> Keeping topology code for supervisor until topology got killed
> --------------------------------------------------------------
>
> Key: STORM-1952
> URL: https://issues.apache.org/jira/browse/STORM-1952
> Project: Apache Storm
> Issue Type: Improvement
> Components: storm-core
> Affects Versions: 1.0.0, 2.0.0, 1.0.1
> Reporter: Jungtaek Lim
>
> It's based on review comment from [~sriharsha].
> https://github.com/apache/storm/pull/1528/files#r69152524
> Please feel free to change reporter if you would like to.
> In supervisor we're removing topology code when assignments for that
> supervisor has gone.
> But there's valid scenario to need to keep the topology code though
> assignments for that supervisor is none, for example, rebalancing.
> So it would be better for supervisor to keep topology code until topology has
> been killed (and all topology workers assigned to that supervisor are also
> killed).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
