[MCOL-3736] mcsadmin suspenddatabasewrites can return before S3 sync is complete Created: 2020-01-21  Updated: 2020-03-13  Resolved: 2020-02-13

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

Type: Bug Priority: Blocker
Reporter: Ben Thompson (Inactive) Assignee: Daniel Lee (Inactive)
Resolution: Fixed Votes: 0
Labels: None

Sprint: 2020-2

 Description   

The change for MCOL-3577 to allow suspenddatabasewrites to sync the S3 journals and metadata has a logical error that makes it possible to complete before journal files have been sync'd and allow backup scripts to capture metadata mismatched from s3 bucket contents.

 Comments   
Comment by Ben Thompson (Inactive) [ 2020-02-12 ]

Test this by running mcsadmin suspenddatabasewrites with S3 / storagemanager. Verify the metadata directory files all have the correct objects matching objects in the S3 bucket. The way this was captured was through a script that does the following quickly

mcsadmin suspenddatabasewrites
copy entire /root/storagemanager/metadata to another location

at this point if the suspend really waited for all storagemanager data to sync you should find all objects match metadata. Before the fix files referenced in metadata would not exist in S3 bucket.

Comment by Daniel Lee (Inactive) [ 2020-02-13 ]

Build verified: 1.4.3-1 BB nightly

engine commit:
8588678

cpimported a 10g lineitem table
Immediately run suspendDatabaseWrites. The command waited for a while before completing.
Data directory copied matched

Generated at Thu Feb 08 02:45:03 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.