Patch 1 to fix this in the capacity check.
The check added in
HDFS-9549 was to handle a rare case where a block stuck in pending if the capacity is reached.
I think there're 2 options on the fix:
- ignore null blockInfo. We already have logic to remove the blocks that cannot be found on NN
- remove immediately. I feel this may add a source of confusion since we have 1 more place to remove.
To limit the change scope, patch 1 goes with option 1.