Ilya Kasnacheev created IGNITE-14711:
----------------------------------------
Summary: Client discovery thread interrupt/stop causes endless
communication reconnect attempt
Key: IGNITE-14711
URL: https://issues.apache.org/jira/browse/IGNITE-14711
Project: Ignite
Issue Type: Bug
Components: networking
Affects Versions: 2.10
Reporter: Ilya Kasnacheev
Original issue: if tcp-client-disco-sock-reader thread dies on client node, it
will never disconnect from the cluster despite NODE_FAILED, and will endlessly
try to open communication connections to server while getting "Remote node does
not observe current node in topology" exceptions on client and "Close incoming
connection, unknown node" on server.
Generalized issue: stop()ing or interrupt()ing discovery threads cause cluster
to hang in many cases, where it is expected that any such node will:
* Restart the thread and continue normally
* Disconnect from the cluster to re-establish discovery connection
* Stop and close all remaining threads.
See the attached reproducer.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
