Ensure Test Cluster in Azure Discovery Tests Closes #43057
Conversation
* Use `internalCluster().close()` to force all nodes (and not just the datanodes) to shut down even if one fails to shut down in time * More than 10s timeout on node shutdown, if we have a slow run 10s might not suffice in an IT * Force closing httpServer to get cleaner logs if nodes still hang on shut down * Closes elastic#43048
original-brownbear
added
>test
|
Pinging @elastic/es-distributed |
|
Jenkins run elasticsearch-ci/2 |
|
It appears the increased timeout somehow breaks an ML test?? I'm looking into that. Update: yea the ML tests are in fact dependent on the node shutdown running into the interrupt mentioned below within 10s, so I can't just up that here without messing with those tests. |
|
@henningandersen I looked into the code some more and I'm not so convinced that upping the timeout here will help (closing a node waits for 10s then interrupts the node's threadpool and waits another 10s so we shouldn't hang here really). I'd still merge this if it's ok with you just to get a clean stack trace for where exactly things are stuck (currently, we're not stopping all the nodes and neither do we stop the web server on failure so it's hard to tell). |
* Use `internalCluster().close()` to force all nodes (and not just the datanodes) to shut down even if one fails to shut down in time * Force closing httpServer to get cleaner logs if nodes still hang on shut down * Relates elastic#43048
* Use `internalCluster().close()` to force all nodes (and not just the datanodes) to shut down even if one fails to shut down in time * Force closing httpServer to get cleaner logs if nodes still hang on shut down * Relates #43048
internalCluster().close()to force all nodes (and not just the datanodes) to shut down even if one fails to shut down in time