Details
-
Bug
-
Status: Closed (View Workflow)
-
Blocker
-
Resolution: Fixed
-
1.4
-
None
-
Ubuntu 18, not-root dev environment
-
2019-06
Description
Looking into why our S3 stuff wasn't deleting everything it should, I found it was a CS-side problem in develop as of commit 355e2f91. I can't say when it started happening. It is not currently happening in develop-1.2.
Running test000 once takes up about 7.2GB of disk space. Running it again, it should use the same 7.2GB, but instead uses about 12GB. It adds about 5GB of disk usage per run.
Some add'l info...
Ubuntu 18.04
running as non-root
SKIP_OAM_INIT & starting with startcs/stopcs scripts
Attachments
Issue Links
- is caused by
-
MCOL-498 provide option to not preallocate disk on extents
-
- Closed
-
Activity
In a test build I took develop and reverted all the MCOL-498 commits. The bug went away. So it is caused by MCOL-498.
Cause is commit https://github.com/mariadb-corporation/mariadb-columnstore-engine/pull/738/commits/7cf0d55dd0d43c24a0a405c050b7fbbbe77863fe
The issue worths the explanation.
The proactive fill-block-with-empty-magic-values happens to overwrite a legitimate block with data if the first value of the block has been erased by DELETE operation. This logic is fired by INSERT SQL operation. I moved the the logic into a higher level to allow me to call the logic only when it is needed.
Build verified: 1.4.0-1
engine commit:
975463c
The reported issue no longer occurring.
Confirmed using non-root and skip OAM. When a table is dropped the extent files are not being deleted and are remaining in the extent map. Essentially orphaned.