[HDFS-16476] Increase the number of metrics used to record PendingRecoveryBlocks

• Type: Improvement
• Status: In Progress
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 2.9.2, 3.4.0
• Component/s: metrics, namenode
• Labels: pull-request-available
• Assignee: JiangHua Zhu
• Reporter: JiangHua Zhu
• Created: 22/Feb/22 04:26
• Updated: 05/Mar/22 17:43

Description

The complete process of block recovery is as follows:
1. NameNode collects which blocks need to be recovered.
2. The NameNode issues instructions to some DataNodes for execution.
3. DataNode tells NameNode after execution is complete.

Now there is no way to know how many blocks are being recovered. The number of metrics used to record PendingRecoveryBlocks should be increased, which is good for increasing the robustness of the cluster.

Here are some logs of DataNode execution:
2022-02-10 23:51:04,386 [12208592621] - INFO [IPC Server handler 38 on 8025:FsDatasetImpl@2687] - initReplicaRecovery: changing replica state for blk_xxxx from RBW to RUR
2022-02-10 23:51:04,395 [12208592630] - INFO [IPC Server handler 47 on 8025:FsDatasetImpl@2708] - updateReplica: BP-xxxx:blk_xxxx, recoveryId=18386356475, length=129869866, replica=ReplicaUnderRecovery, blk_xxxx, RUR

Here are some logs that NameNdoe receives after completion:
2022-02-22 10:43:58,780 [8193058814] - INFO [IPC Server handler 15 on 8021:FSNamesystem@3647] - commitBlockSynchronization(oldBlock=BP-xxxx, newgenerationstamp=18551926574, newlength=16929, newtargets=[xxxx1:1004, xxxx2:1004, xxxx3:1004]) successful

Issue Links

• links to: GitHub Pull Request #4010