[MXS-3379] Make transaction_replay less strict Created: 2021-01-19  Updated: 2021-12-08  Resolved: 2021-12-08

Status: Closed
Project: MariaDB MaxScale
Component/s: readwritesplit
Affects Version/s: None
Fix Version/s: 6.2.1

Type: New Feature Priority: Major
Reporter: markus makela Assignee: Todd Stoffel (Inactive)
Resolution: Fixed Votes: 1
Labels: None

Issue Links:
Duplicate
duplicates MXS-3894 Invent a configuration option to allo... Closed

 Description   

Allowing the user to configure how strict transaction_replay would help improve the likelihood of transactions being successfully replayed.

One common case is that the OK packet in response to an INSERT contains the auto-generate value for an AUTO_INCREMENT field. This can cause transaction replay to fail if the same ID is not generated during the replay. Depending on the client application, the actual ID returned in the OK packet might not be used and the LAST_INSERT_ID() function is used instead.

This option could also modify the checksum calculation algorith to calculate the resultset checksum in a way in which the order of the resultset rows is not critical. This can happen if the SQL statement doesn't have an ORDER BY clause and the implicit ordering of the database is used. This could be done by multiplying the checksums of each packet in the responses instead of doing the checksum on the whole result.



 Comments   
Comment by markus makela [ 2021-12-08 ]

This is fixed by MXS-3894.

Generated at Thu Feb 08 04:21:00 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.