Bill Burcham created GEODE-10391:
------------------------------------
Summary: Region Operation During Primary Change in P2P-only
Configuration Results in Spurious Entry{NotFound|Exists}Exception
Key: GEODE-10391
URL: https://issues.apache.org/jira/browse/GEODE-10391
Project: Geode
Issue Type: Bug
Components: regions
Affects Versions: 1.16.0
Reporter: Bill Burcham
When a primary moves while a region operation, e.g. create, is in-flight, i.e.
started but not yet acknowledged, the operation will be retried automatically,
until the operation succeeds or fails.
When a member notices another member has crashed, the surviving member requests
(from the remaining members) data for which the crashed member had been primary
(delta-GII/sync). This sync is necessary to regain consistency in case the
(retrying) requester fails before it can re-issue the request to the new
primary.
In GEODE-5055 we learned that we needed to delay that sync request long enough
for the new primary to be chosen and for the original requester to make a new
request against the new primary. If we didn't delay the sync, the primary could
end up with the entry in the new state (as if the operation had completed) but
without the corresponding event tracker data needed to conflate the retried
event.
The fix for GEODE-5055 introduced a delay, but only for configurations where
clients were present. If only peers were present there would be no delay. This
ticket pertains to the P2P-only case.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
