Roman Puchkovskiy created IGNITE-17944:
------------------------------------------
Summary: Logging storm on server node failure/disconnect during
loading through a DataStreamers
Key: IGNITE-17944
URL: https://issues.apache.org/jira/browse/IGNITE-17944
Project: Ignite
Issue Type: Bug
Reporter: Roman Puchkovskiy
This can be reproduced in RebalanceIteratorLargeEntriesOOMTest when changing
#additionalRemoteJvmArgs() to return this:
return Arrays.asList("-Xmx128m", "-Xms128m", "-XX:+HeapDumpOnOutOfMemoryError",
"-XX:+CrashOnOutOfMemoryError");
On my machine, when the remote node crashes, the machine becomes barely
responsive because all the cores peak to 100% of load.
There happens a LOT of logging at that phase, probably a lot of Buffers of
DataStreamers get cancelled, so a lot of futures get cancelled, which is made
via completing with an exception, which causes a lot of logging.
It seems that same thing might happen in production if a node suddently crashes
amidst data loading by a client.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
