[fix](tablet clone) fix replica last failed version after clone done #25898
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Contributor
Author
|
run buildall |
Contributor
|
(From new machine)TeamCity pipeline, clickbench performance test result: |
github-actions
bot
added
the
approved
Contributor
|
PR approved by at least one committer and no changes requested. |
Contributor
|
PR approved by anyone and no changes requested. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Proposed changes
partially pick: #25236
it may cause case replica's version = partition's visible version = partition's committed version, and replica's last failed version = partition's + 1. case as follow:
a. suppose partition visible version = 10, committed version = 11, the committed txn version T is not published yet;
b. clone a replica A to B, after cloning, B's version = 10.
c. publish txn T, A version = 11. Also due to txn publish bug, B's version = 11 (PR #23706 has fix this). (Another case is: even if this PR is fix and be publish txns ok, but if it powerf off and restart, the be may miss the version 11 because it's not ready to sync disk, and the writes will lost)
d. publish another txn, A and B's version become 12;
e. now B's backend version is: [1, 10], [12, 12], it will report it missing version = 11; then fe will mark this replica's last failed version = version + 1 = 12 + 1 = 13;
f. fe will let B clone version 11 from A, after cloning, B will contains all the version. But Replica.updateVersion has bug,
if it found its version not change (12 -> 12)，it will not set replica's last failed version to -1.
Then later, B will always be: version = 12, last failed version = 13.
To Fix this problem, after cloning, if replica's version >= partition's commit version(more precise, replica's version = partition's commit version), and replica's last failed version > partition's commit (more precise, replica's last failed version = partition's commit version + 1), then we should set replica's last failed version to -1.
Further comments
If this is a relatively large or complex change, kick off the discussion at dev@doris.apache.org by explaining why you chose the solution you did and what alternatives you considered, etc...