Operations in sharding that hold distributed locks for a long time without contacting the config server, such as chunk migration, need to periodically confirm that the config servers have not been converted from SCCC to CSRS or vice versa. Such upgrades and downgrades ensure that these operations will eventually fail, but if the node does not check during the operation, it may prevent other new operations from making progress, since it blocks the local node from switching out its catalog manager implementation.
- is related to
-
SERVER-22524 Only interrupt mapReduce on catalog manager swap if it is outputting to a sharded collection
- Closed