Details
-
Improvement
-
Resolution: Won't Do
-
Major - P3
-
None
-
None
Description
At present chunk splits are initiated by a mongoS when the amount of data written into the chunk (tracked by the mongoS) exceeds the threshold of 20% of the maximum chunk size.
When mongoS is restarted, that re-sets the data written counter, which in return can delay splits causing chunks to grow beyond the splitting threshold and even beyond the maximum chunk size threshold. Eventually such chunks can be split in a cascading fashion, adding (potentially significant) balancing overhead.
The documentation should suggest that:
- frequent restarts of mongoS should be avoided, when possible
- the flushRouterConfig command should only be executed in exceptional circumstances, when there is a good reason to do so. It should not be run for prevention purposes on regular basis