[OPENDJ-6682] Divergences in replication after upgrade from 2.6.4: entry missing in the changelog Created: 27/Sep/19 Updated: 11/Dec/19 Resolved: 11/Dec/19
|Reporter:||carole forel||Assignee:||carole forel|
|Epic Link:||Bugs 7.0|
Found with rev (d7c1991a9f8).
To check the upgrade task that does the ecl configuration migration, we have added a test that does the following:
Then both servers are stopped and upgraded one after the other, successfully.
and add an entry on this suffix and check the changelog:
Even if we wait for a bit, nothing for dc=othersuffix appears in the changelog:
To reproduce the issue:
|Comment by Fabio Pistolesi [ 09/Dec/19 ]|
I think it is another MCP and ReplicaOfflineMsg related problem: 2.6.4 does not know about it.
Given 3 instances on 2.6.4 and doing a rolling upgrade, means the first upgraded server goes from N replicaids to only one. The changelog is deleted on the upgraded server, but when it starts, it connects to remaining 2.6.4 servers, who start replicating all data back to it, including from the replicaIDs deleted from the configuration.
|Comment by Fabio Pistolesi [ 10/Dec/19 ]|
After discussion, the decision is not to support direct upgrade from 2.6.4 to 7.0 directly. Instead, mention in the release notes upgrade from 2.6.4 is a two step process (using an intermediate version first) and stop upgrade to run if run on a 2.6.4 instance.
|Comment by carole forel [ 11/Dec/19 ]|
We have removed from our jenkins ci jobs any tests related to 2.6.4 upgrade.
|Comment by Mark Craig [ 11/Dec/19 ]|
Removing the release-notes label because I've added hand-written information in the release notes instead. (All the label will help me do is run a query that includes the bug in the fixed issues list, which would be misleading in this case.)