The documentation recommends doing an fsync+lock of a config server when taking a backup of a sharded cluster, but all of my tests using this method seem to result in blocked inserts after a few seconds ( I think when it needs to split the chunk ).
3 replica set members per shard
3 config servers
chunksize=1 to force splits earlier
Steps to reproduce
1. Stop balancer
2. Verify balancer stopped
3. Run test script to generate inserts
4. fsync+lock config server
5. Wait a few seconds for block
I think I can work around the issue by shutting down the config database instead of doing an fsync and lock, but wanted to check to see whether or not this was a bug or just something missing in the documentation.