[MCOL-3745] Investigate slow-down in DDL after a restart Created: 2020-01-27  Updated: 2020-03-13  Resolved: 2020-02-28

Status: Closed
Project: MariaDB ColumnStore
Component/s: Storage Manager
Affects Version/s: 1.4.2
Fix Version/s: 1.4.3

Type: Task Priority: Major
Reporter: Todd Stoffel Assignee: Patrick LeBlanc (Inactive)
Resolution: Fixed Votes: 0
Labels: None
Environment:

Sky docker image


Issue Links:
Relates
relates to MCOL-3751 Reduce duration or type of locks held... Closed
Sprint: 2020-1, 2020-2, 2020-3

 Description   

Todd noticed during testing that if he killed either controllernode or DMLProc, and let it restart, drop table would be occasionally be slow.

Ben reproduced the problem and saw correlation with upstream bandwidth usage. When upstream traffic stops, that's when the drop table stmt would return.
Presumably, SM is sending a lot of data to the cloud, which is blocking the ops for dropping a table.

Ben also couldn't reproduce it when using the 'fake cloud' module.

Need to investigate.



 Comments   
Comment by Patrick LeBlanc (Inactive) [ 2020-01-31 ]

Yes it's contention on a few hot files that get changed on many ops like in this sequence. Access to the BRM files, and the syscat files is blocked while they are being sync'd to cloud storage. On home networks with limited upstream that could be quite a while. For now I'll add documentation to relevant parameters in the config file to reduce the observed latency. I'll add a ticket to try to avoid locking files during network xfers.

Comment by Patrick LeBlanc (Inactive) [ 2020-02-03 ]

The intermediate fix for this is documentation in the config file.

Comment by Patrick LeBlanc (Inactive) [ 2020-02-28 ]

It was merged some time ago.

Generated at Thu Feb 08 02:45:07 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.