Details
-
Bug
-
Resolution: Done
-
Major - P3
-
None
-
2.3.1
-
None
-
ALL
Description
Whenever the configuration of a replica set changes, not all instances (whether they are kept by a thread servicing a client connection, idle in the pool or active) of the connections would know about this event. This can cause some unintuitive behaviors especially when mongos tries to use a very old replica set connection and gets a "not master" error.
Proposed fix would be to make the replica set connection retry (meaning, re-evaluate the target host) when it gets the "not master" error (and perhaps consider for some class of errors as well) for the first time.