-
Type:
Improvement
-
Resolution: Done
-
Priority:
Minor - P4
-
None
-
Affects Version/s: 3.0.12, 3.2.6
-
Component/s: Sharding
-
Labels:
-
Sharding EMEA
-
Sharding 16 (06/24/16)
If the recipient shard fails to start migration early in the cycle, before it has installed the pending chunk into the metadata, its cleanup code will still try to clean it up.
This results in a misleading message in the logs:
2016-05-20T13:10:50.088+0100 I SHARDING [migrateThread] starting receiving-end of migration of chunk { files_id: MinKey, n: MinKey } -> { files_id: ObjectId('000085341e8011e6a61c8380'), n: 0 } for collection db.coll from rs/host1:38101,host2:38102 at epoch 573c7245e4f3add3194fa270
2016-05-20T13:10:50.089+0100 W SHARDING [migrateThread] failed to create index before migrating data. error: IndexOptionsConflict: Index with name: files_id_1_n_1 already exists with different options
...
2016-05-20T13:10:50.091+0100 W SHARDING [migrateThread] cannot remove pending chunk [{ files_id: MinKey, n: MinKey }, { files_id: ObjectId('000085341e8011e6a61c8380'), n: 0 }), this shard does not contain the chunk
We should make sure to not do this cleanup if the pending chunk range has not been installed in the metadata.