bugfix: fix when doing global rollback ,may cause an endless loop #6569
Conversation
sixlei
changed the title
funky-eyes
added
type: bug
| LOGGER.info("xid {} branch {}, undo_log added with {}", xid, branchId, | ||
| State.GlobalFinished.name()); | ||
| try { | ||
| insertUndoLogWithGlobalFinished(xid, branchId, UndoLogParserFactory.getInstance(), conn); |
There was a problem hiding this comment.
Why is it modified here?
There was a problem hiding this comment.
The changes here are to narrow the scope of the catch for SQLIntegrityConstraintViolationException at the outermost level, only handling conflicts with the unique index on the undo_log. If the unique index conflict also occurs when executing the rollback undoExecutor.executeOn for the business, it will enter a deadlock directly here.
There was a problem hiding this comment.
The changes here are to narrow the scope of the catch for SQLIntegrityConstraintViolationException at the outermost level, only handling conflicts with the unique index on the undo_log. If the unique index conflict also occurs when executing the rollback undoExecutor.executeOn for the business, it will enter a deadlock directly here.
I am very sorry, I replied to this paragraph one day after you replied, but it has been pending. I didn't know I had to click submitted. disturb you
There was a problem hiding this comment.
The changes here are to narrow the scope of the catch for SQLIntegrityConstraintViolationException at the outermost level, only handling conflicts with the unique index on the undo_log. If the unique index conflict also occurs when executing the rollback undoExecutor.executeOn for the business, it will enter a deadlock directly here.
There was a problem hiding this comment.
The changes here are to narrow the scope of the catch for SQLIntegrityConstraintViolationException at the outermost level, only handling conflicts with the unique index on the undo_log. If the unique index conflict also occurs when executing the rollback undoExecutor.executeOn for the business, it will enter a deadlock directly here.
Sorry, I just saw your message now. I'll review the code as soon as possible.
|
Due to the author's prolonged lack of response, I will close this PR and seek another developer to address this issue. |
|
Could you please resolve the code conflicts first? |
| LOGGER.info("xid {} branch {}, undo_log added with {}", xid, branchId, | ||
| State.GlobalFinished.name()); | ||
| } | ||
| } catch (SQLIntegrityConstraintViolationException e) { |
There was a problem hiding this comment.
你只是使SQLIntegrityConstraintViolationException的范围缩小了，但是并没将因为唯一索引导致的SQLIntegrityConstraintViolationException 转换成SQLUndoDirtyException或者说抛出BranchRollbackFailed_Unretriable状态的BranchTransactionException，目前的改动我认为是无效的
You have merely narrowed the scope of SQLIntegrityConstraintViolationException without converting SQLIntegrityConstraintViolationException caused by unique indexes into SQLUndoDirtyException or throwing BranchTransactionException with BranchRollbackFailed_Unretriable state. I find the current changes ineffective.
Ⅰ. Describe what this PR did
fix issue #6561
Ⅱ. Does this pull request fix one issue?
fix issue #6561
将对undolog的SQLIntegrityConstraintViolationException异常的try catch范围缩小，只处理undo_log的插入并发，如果有业务回滚时的并发插入异常，直接抛出，按照脏写处理即可。
Ⅲ. Why don't you add test cases (unit test/integration test)?
Ⅳ. Describe how to verify it
Ⅴ. Special notes for reviews