maksaska commented on code in PR #13627:
URL: https://github.com/apache/ignite/pull/13627#discussion_r4539763129
##########
modules/core/src/main/java/org/apache/ignite/internal/processors/cache/distributed/dht/preloader/GridDhtPartitionsExchangeFuture.java:
##########
@@ -2391,8 +2394,21 @@ private String exchangeTimingsLogMessage(String header,
List<String> timings) {
cleanIdxRebuildFutures = false;
- for (CacheGroupContext grp : cctx.cache().cacheGroups())
- grp.topology().onExchangeDone(this,
grp.affinity().readyAffinity(res), false);
+ for (CacheGroupContext grp : cctx.cache().cacheGroups()) {
+ if (grp.topology().onExchangeDone(this,
grp.affinity().readyAffinity(res), false))
+ locChangedGrps.add(grp);
+ }
+
+ if (!locChangedGrps.isEmpty()
+ &&
cctx.kernalContext().rollingUpgrade().features().isActive(PART_STATE_FROM_OWNER_FEATURE))
{
Review Comment:
I'd keep the check. Until the cluster version is finalized, a node of the
new version should behave as the current cluster version, as
RollingUpgradeProcessor#enableVersionUpgrade puts it: "nodes running a higher
version operate in a compatibility mode that emulates the behavior of the
current cluster version". So the cluster has two states: the old behavior
before finalization and the new one after it. Without the check there would be
a third state in the middle, where new nodes send their maps while every node
still writes the owners' state itself.
Yes, the bug stays until finalization, as it does without this PR. On
activation every server resends its maps once
(GridCachePartitionExchangeManager#onLocalJoin), so owners whose last exchange
ended before it are seen too. If we drop the feature (see the main thread),
this check goes with it.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]