[HDFS-8718] Block replicating cannot work after upgrading to 2.7

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 2.7.0
• Assignee: Unassigned
• Reporter: Bing Jiang
• Created: 07/Jul/15 09:04
• Updated: 09/Aug/18 16:49
• Resolved: 09/Aug/18 16:42

Description

Decommission a datanode from hadoop, and hdfs can calculate the correct number of blocks to be replicated from web-ui.

Decomissioning
Node	Last contact	Under replicated blocks	Blocks with no live replicas	Under Replicated Blocks 
In files under construction
TS-BHTEST-03:50010 (172.22.49.3:50010)		25641	0	0

From NN's log, the work of block replicating cannot be enforced due to inconsistent expected storage type.

Node /default/rack_02/172.22.49.5:50010 [
  Storage [DISK]DS-3915533b-4ae4-4806-bf83caf1446f1e2f:NORMAL:172.22.49.5:50010 is not chosen since storage types do not match, where the required storage type is ARCHIVE.
  Storage [DISK]DS-3e54c331-3eaf-4447-b5e4-9bf91bc71b17:NORMAL:172.22.49.5:50010 is not chosen since storage types do not match, where the required storage type is ARCHIVE.
  Storage [DISK]DS-d44fa611-aa73-4415-a2de-7e73c9c5ea68:NORMAL:172.22.49.5:50010 is not chosen since storage types do not match, where the required storage type is ARCHIVE.
  Storage [DISK]DS-cebbf410-06a0-4171-a9bd-d0db55dad6d3:NORMAL:172.22.49.5:50010 is not chosen since storage types do not match, where the required storage type is ARCHIVE.
  Storage [DISK]DS-4c50b1c7-eaad-4858-b476-99dec17d68b5:NORMAL:172.22.49.5:50010 is not chosen since storage types do not match, where the required storage type is ARCHIVE.
  Storage [DISK]DS-f6cf9123-4125-4234-8e21-34b12170e576:NORMAL:172.22.49.5:50010 is not chosen since storage types do not match, where the required storage type is ARCHIVE.
  Storage [DISK]DS-7601b634-1761-45cc-9ffd-73ee8687c2a7:NORMAL:172.22.49.5:50010 is not chosen since storage types do not match, where the required storage type is ARCHIVE.
  Storage [DISK]DS-1d4b91ab-fe2f-4d5f-bd0a-57e9a0714654:NORMAL:172.22.49.5:50010 is not chosen since storage types do not match, where the required storage type is ARCHIVE.
  Storage [DISK]DS-cd2279cf-9c5a-4380-8c41-7681fa688eaf:NORMAL:172.22.49.5:50010 is not chosen since storage types do not match, where the required storage type is ARCHIVE.
  Storage [DISK]DS-630c734f-334a-466d-9649-4818d6e91181:NORMAL:172.22.49.5:50010 is not chosen since storage types do not match, where the required storage type is ARCHIVE.
  Storage [DISK]DS-31cd0d68-5f7c-4a0a-91e6-afa53c4df820:NORMAL:172.22.49.5:50010 is not chosen since storage types do not match, where the required storage type is ARCHIVE.
]
2015-07-07 16:00:22,032 WARN org.apache.hadoop.hdfs.protocol.BlockStoragePolicy: Failed to place enough replicas: expected size is 1 but onl
y 0 storage types can be selected (replication=3, selected=[], unavailable=[DISK, ARCHIVE], removed=[DISK], policy=BlockStoragePolicy{HOT:7,
 storageTypes=[DISK], creationFallbacks=[], replicationFallbacks=[ARCHIVE]})
2015-07-07 16:00:22,032 WARN org.apache.hadoop.hdfs.server.blockmanagement.BlockPlacementPolicy: Failed to place enough replicas, still in n
eed of 1 to reach 3 (unavailableStorages=[DISK, ARCHIVE], storagePolicy=BlockStoragePolicy{HOT:7, storageTypes=[DISK], creationFallbacks=[],
 replicationFallbacks=[ARCHIVE]}, newBlock=false) All required storage types are unavailable:  unavailableStorages=[DISK, ARCHIVE], storageP
olicy=BlockStoragePolicy{HOT:7, storageTypes=[DISK], creationFallbacks=[], replicationFallbacks=[ARCHIVE]}

We have upgraded the hadoop cluster from 2.5 to 2.7.0 previously. I believe the feature of ARCHIVE STORAGE has been enforced, but how about the block's storage type after upgrading?

The default BlockStoragePolicy is hot, and I guess those blocks do not have the correct information bit of BlockStoragePolicy, so it cannot be handled well.
After I shutdown the datanode, the under-replicated blocks can be asked to copy. So the workaround is to shutdown the datanode.

Could anyone take a look at the issue?

Issue Links

• duplicates: HDFS-10453 ReplicationMonitor thread could stuck for long time due to the race between replication and delete of same file in a large cluster.