[
https://issues.apache.org/jira/browse/HIVE-15958?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15871116#comment-15871116
]
Siddharth Seth commented on HIVE-15958:
---------------------------------------
Looks good.
There's still a chance of a bunch of extra connections being made if AMNodeInfo
is not cached. It gets removed when fragmentCount goes to 0, i.e. all fragments
for this AM are complete or executing (not in the wait queue).
Would be good to clean the AMNodeInfo cache on query complete instead of count
going to 0. Can be a follow up jira though, since this fixes the main issue.
[~gopalv] - this does not undo that change. This is related to unnecessary
connection being created for kill messages. Tasks still use their own threads
for heartbeats. Wasn't HADOOP-11772 related to not being able to get a
connection, rather than everything serialized on a single connection? I think
we still need to move to everything on a single connection at some point.
> LLAP: IPC connections are not being reused for umbilical protocol
> -----------------------------------------------------------------
>
> Key: HIVE-15958
> URL: https://issues.apache.org/jira/browse/HIVE-15958
> Project: Hive
> Issue Type: Bug
> Components: llap
> Affects Versions: 2.2.0
> Reporter: Rajesh Balamohan
> Assignee: Prasanth Jayachandran
> Attachments: HIVE-15958.1.patch
>
>
> During concurrency testing, observed 1000s of ipc thread creations. Ideally,
> the connections to same hosts should be reused.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
