Details
-
Improvement
-
Resolution: Unresolved
-
Major - P3
-
None
-
2.6.11, 3.0.5, 3.1.4
-
None
-
Replication
Description
Now that SERVER-18280 is fixed, it is possible to see mongos messages such as:
2015-08-03T06:06:55.726+0000 I NETWORK [conn24] node HOSTNAME:PORT believes it is primary, but its election id of 55bf04a0c0f569a0268e3ab5 is older than the most recent election id for this set, 55bf04aacc82c473f5fa767f
|
However, when attempting to diagnose the source of this message, these election ids are not present anywhere in the shard logs (at default level).
When a member is elected, it should log the election id it that it will be reporting. eg:
-2015-08-03T06:05:20.523+0000 I REPL [ReplicationExecutor] replSet election succeeded, assuming primary role
|
+2015-08-03T06:05:20.523+0000 I REPL [ReplicationExecutor] replSet election succeeded, assuming primary role, election id 55bf04a0c0f569a0268e3ab5
|
2015-08-03T06:05:20.523+0000 I REPL [ReplicationExecutor] transition to PRIMARY
|
2015-08-03T06:05:21.478+0000 I REPL [rsSync] transition to primary complete; database writes are now permitted |
The workaround is to use the timestamp of the election id ObjectId as a heuristic, ie. noticing that
> ObjectId("55bf04a0c0f569a0268e3ab5").getTimestamp()
|
ISODate("2015-08-03T06:05:20Z")
|
and then going looking for an election around 06:05:20.
Attachments
Issue Links
- related to
-
SERVER-18280 ReplicaSetMonitor should use electionId to avoid talking to old primaries
-
- Closed
-