Fix potential data inconsistency under heavy ddl operation (#5044)#5051
Fix potential data inconsistency under heavy ddl operation (#5044)#5051ti-chi-bot wants to merge 1 commit intopingcap:release-6.0from
Conversation
Signed-off-by: ti-chi-bot <ti-community-prow-bot@tidb.io>
|
[REVIEW NOTIFICATION] This pull request has not been approved. To complete the pull request process, please ask the reviewers in the list to review by filling The full list of commands accepted by this bot can be found here. DetailsReviewer can indicate their review by submitting an approval review. |
|
@ti-chi-bot: This cherry pick PR is for a release branch and has not yet been approved by release team. To merge this cherry pick, it must first be approved by the collaborators. AFTER it has been approved by collaborators, please ping the release team in a comment to request a cherry pick review. DetailsInstructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. |
|
I will fix the conflict when 6.0.1 is needed. |
|
close as no further patch version for the 6.0 branch |
This is an automated cherry-pick of #5044
What problem does this PR solve?
Issue Number: close #5032
Problem Summary: Currently we cache decoding schema for decoding raft data if a table schema doesn't change. And we judge whether a table schema has changed based on the table schema version.
But the schema version is not strictly consistent with the actual table schema which can be seen at https://github.com/pingcap/tiflash/blob/master/dbms/src/TiDB/Schema/SchemaBuilder.cpp#L362. That is when applying different schema changes in a diff, the table schema version will be set to the latest schema version after the first schema change is applied.
More concretely, when a lossy ddl change occurs, it will trigger
drop columnandadd columnschema changes and also rewrite some data at the same time. The schema changes will be applied one by one, and because tiflash updates the table schema version ahead of time when applying schema diff, after applying thedrop columnschema change, it will update the schema version to the latest schema version.And if the decode thread tries to obtain the current schema for decoding data before the
add columnis applied, the current schema and the latest schema version will be cached. Then after the subsequentadd columnoperation is applied, the table schema version will not be updated, so the cache of the decode thread will not be invalidated.Therefore, the decode thread will decode the new data with an older schema, considering that the new added column is a dropped column and discarding it.
In addition, there is a lower chance of triggering this problem in the case of frequent add column and insert data.
What is changed and how it works?
Add an internal schema version for
DecodingStorageSchemaSnapshot.Check List
Tests
Side effects
Documentation
Release note