[FLINK-930] Netty Initialization is sometimes very slow #19
Conversation
|
Can someone please review and merge? Travis is running consistently out of time atm and this PR should "fix it". We can then further monitor the Netty ThreadLocalRandom warning. |
| try { | ||
| this.networkConnectionManager.shutdown(); | ||
| } catch (IOException e) { | ||
| LOG.warn("NetworkConnectionManager did not shutdown properly."); |
There was a problem hiding this comment.
I think we should log the exception as well here (I guess the error is unlikely and needs review by the user)
|
I agree with both your points and have addressed them. The thrown exceptions are rethrown as IOExceptions up to the TaskManager's shutdown method, where it is logged. I decided against changing the signature of the TaskManager's shutdown method to rethrow the Exception as I think this should go hand in hand with changes to the other component shutdowns as well. |
|
Nice to see it green again: https://travis-ci.org/uce/incubator-flink/builds/27754618 |
|
Looks good to me. Will merge. |
This closes apache#19
Rebase on release-1.9
…or so that checkpoints can correctly fail This closes apache#19.
This is FLINK-930.
The issue with the Netty warning was reported in netty/netty#2412 and has been fixed in
4.0.19.Final. I couldn't reproduce the reported problem (with and without the changes in this PR).This PR contains the following changes:
4.0.19.Finalto4.0.20.FinalandExecutionModefor TaskManagers, where inLOCALmode theNetworkConnectionManageris not started (e.g. no ServerSocket binding etc.)The 2. change speeds up the startup and shutdown time of the TaskManagers noticeably and solves/circumvents the reported problem with each test taking ~ 3 seconds for startup/shutdown.