Add retention leases replication tests#38857
Conversation
|
Pinging @elastic/es-distributed |
| } | ||
| } | ||
| } | ||
| group.promoteReplicaToPrimary(newPrimary).get(); |
There was a problem hiding this comment.
Discuss: should we align the retention-leases when a new primary is promoted?
There was a problem hiding this comment.
Under what circumstances can the new primary not hold an up-to-date set of leases already? It might perhaps be missing some renewals but I think that's ok.
There was a problem hiding this comment.
We are adding two new leases to the old primary: L1 and L2. L1 was synced to replica r1; L2 was synced to r2, but the old primary crashed before two leases are properly synced to all two replicas. If any replica is promoted, then the retention leases between copies are not aligned.
There was a problem hiding this comment.
We sync by copying all the leases from the primary to its replicas, so I don't follow how r2 could receive L2 without also receiving L1 (assuming they were added in this order).
However, I think I do see a potential problem:
- primary A shares a lease with one replica B, but not to another replica C
- A crashes
- C discards some history that the lease would have retained
- B is promoted to primary and shares its lease with C
- C cannot accept this lease since it has already discarded this history
I think we can prevent this, with peer-recovery retention leases, by insisting that leases do not "go backwards", i.e. they only retain history that is already being retained by another lease. This would mean that C could not discard the history in the situation above because it must already hold a different lease that retains that history.
|
run elasticsearch-ci/1 |
DaveCTurner
left a comment
There was a problem hiding this comment.
Looks good - I needed to make similar adjustments to support peer-recovery retention leases. I left some minor comments. Also there are now merge conflicts.
| } | ||
| } | ||
| } | ||
| group.promoteReplicaToPrimary(newPrimary).get(); |
There was a problem hiding this comment.
Under what circumstances can the new primary not hold an up-to-date set of leases already? It might perhaps be missing some renewals but I think that's ok.
|
|
||
| @Override | ||
| public void backgroundSync(ShardId shardId, RetentionLeases retentionLeases) { | ||
| sync(shardId, retentionLeases, ActionListener.wrap(r -> {}, e -> fail("fail to sync retention leases [" + e + "]"))); |
There was a problem hiding this comment.
I slightly prefer throw AssertionError("failed to sync retention leases", e); rather than putting the inner stack trace in the message like this.
|
|
||
| public static WritePrimaryResult<Request, Response> performOnPrimary(final Request request, | ||
| final IndexShard primary, | ||
| final Logger logger) { |
There was a problem hiding this comment.
Can we assert that we hold an operation permit here?
There was a problem hiding this comment.
The sync action is too thin now, so I inline it directly to the test.
| return performOnReplica(request, replica, logger); | ||
| } | ||
|
|
||
| public static WriteReplicaResult<Request> performOnReplica(final Request request, final IndexShard replica, final Logger logger) { |
There was a problem hiding this comment.
Can we assert that we hold an operation permit here?
|
Thanks for looking @DaveCTurner. I have addressed your comments. Can you have another look? |
|
Thanks @DaveCTurner. |
* master: Mute failing CCR retention lease unfollow test Add support for ccr follow info api to HLRC. (elastic#39115) Do not create the missing index when invoking getRole (elastic#39039) Relax history check in ShardFollowTaskReplicationTests (elastic#39162) Add retention leases replication tests (elastic#38857) Edits to text in Phrase Suggester doc (elastic#38966) Edits to text in API Conventions docs (elastic#39001)
* master: Mute failing CCR retention lease unfollow test Add support for ccr follow info api to HLRC. (elastic#39115) Do not create the missing index when invoking getRole (elastic#39039) Relax history check in ShardFollowTaskReplicationTests (elastic#39162) Add retention leases replication tests (elastic#38857) Edits to text in Phrase Suggester doc (elastic#38966) Edits to text in API Conventions docs (elastic#39001)
This commit introduces the retention leases to ESIndexLevelReplicationTestCase, then adds some tests verifying that the retention leases replication works correctly in spite of the presence of the primary failover or out of order delivery of retention leases sync requests. Relates #37165
This commit introduces the retention leases to ESIndexLevelReplicationTestCase, then adds some tests verifying that the retention leases replication works correctly in spite of the presence of the primary failover or out of order delivery of retention leases sync requests. Relates #37165
This commit introduces the retention leases to ESIndexLevelReplicationTestCase, then adds some tests verifying that the retention leases replication works correctly in spite of the presence of the primary failover or out of order delivery of retention leases sync requests.
This commit introduces the retention leases to ESIndexLevelReplicationTestCase, then adds some tests verifying that the retention leases replication works correctly in spite of the presence of the primary failover or out of order delivery of retention leases sync requests. Relates elastic#37165
This commit introduces the retention leases to ESIndexLevelReplicationTestCase, then adds some tests verifying that the retention leases replication works correctly in spite of the presence of the primary failover or out of order delivery of retention leases sync requests. Relates elastic#37165
This commit introduces the retention leases to ESIndexLevelReplicationTestCase, then adds some tests verifying that the retention leases replication works correctly in spite of the presence of the primary failover or out of order delivery of retention leases sync requests.
Relates #37165