Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
Description
As of now, when deleting a table, we do the following regarding the filesystem layout:
1. Move the table data to the temp directory (hbase/.tmp)
2. Archive the region directories of the table in the temp directory one by one:
https://github.com/apache/hbase/blob/b322d0a3e552dc228893408161fd3fb20f6b8bf1/hbase-server/src/main/java/org/apache/hadoop/hbase/master/procedure/DeleteTableProcedure.java#L319-L323
However, step 2 will take a long time when the table has a huge number of regions. So I propose doing step 2 in multi-threads in this Jira.
Also, during master startup, we do the same process as step 2:
https://github.com/apache/hbase/blob/b322d0a3e552dc228893408161fd3fb20f6b8bf1/hbase-server/src/main/java/org/apache/hadoop/hbase/master/MasterFileSystem.java#L313-L319
We should make it multi-threaded, similarly.
Attachments
Attachments
Issue Links
- links to