Alexei Scherbakov created IGNITE-12327:
------------------------------------------
Summary: Cross-cache tx is mapped on wrong primary when enlisted
caches have incompatible assignments.
Key: IGNITE-12327
URL: https://issues.apache.org/jira/browse/IGNITE-12327
Project: Ignite
Issue Type: Bug
Affects Versions: 2.7.6
Reporter: Alexei Scherbakov
Assignee: Alexei Scherbakov
Fix For: 2.8
This is happening when supplier node is left while rebalancing is partially
completed on demander.
Suppose we have 2 cache groups, rebalance is in progress and for first group
rebalance is done and for second group rebalance is partially done (some
partitions are still MOVING).
At this moment supplier node dies and corresponding topology version is (N,0).
New assignment is computed using current state of partitions and for first
group will be ideal and the same as for next topology (N,1) which will be
triggered after all rebalancing is completed by CacheAffinityChangeMessage.
For second group affinity will not be ideal.
If transaction is started while PME is in progress (N, 0)->(N,1), first lock
request will pass remap check if it's enslists rebalanced group. All subsequent
lock requests will use invalid topology from previous assignment.
Possible fix: return actual locked topology version from first lock request and
use it for all subsequent requests.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
