Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
-
None
-
Reviewed
Description
In our cluster, we has many SSD and a little HDD. (Most table configured storage policy ONE_SSD, and all wals is configured ALL_SSD)
when all cluster down, It's difficult to recovery cluster. Because HDD Disk IO bottleneck (Almost all disk io util is 100%).
I think the most hdfs operation when recovery is split wal to recovered edits dir, And read it.
And it goes better when i stop hbase and set all recovered.edits to ALL_SSD.
So we can get benifit of recovery time if we set recovered.edits dir to better storage like WAL.
Now i reuse config item hbase.wal.storage.policy to set recovered.edits storage type. Because I did not find a scenario where they use different storage Policy
Attachments
Issue Links
- is a parent of
-
HBASE-26678 Backport HBASE-26579 to branch-1
- Resolved
- links to