Nikita Amelchev created IGNITE-29114:
----------------------------------------
Summary: Fix chained distributed processes hang when the
coordinator leaves between phases
Key: IGNITE-29114
URL: https://issues.apache.org/jira/browse/IGNITE-29114
Project: Ignite
Issue Type: Bug
Reporter: Nikita Amelchev
Assignee: Nikita Amelchev
A distributed process that starts the next phase from its finish closure relies
on {{{}U.isLocalNodeCoordinator{}}}, which is checked on each node when the
full message arrives. If the coordinator leaves after sending the full message
but before starting the next phase, no other node starts it, so the operation
(master key change, cache group key change, snapshot check and restore, rolling
upgrade finalization) hangs forever. The coordinator check is also made against
the discovery ring state, which is not aligned with the discovery events order,
so two nodes may consider themselves the coordinator at the same time.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
