[backport][v23.3.x] tx/compaction: don't deduplicate control batches #17100
Conversation
This file contains 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
Control batches (tx_commit, tx_abort) are special data batches that
indicate the completion of a transaction. Because these are technically
data batches, they participate in deduplication.
It was previously thought that as long as we kept the last control batch
in a segment, that deduplicating such records was safe. However,
consider the following sequence:
node 1:
1. begin tx 1, write data batch in tx 1
2. write many non-tx data batches, up to offset 100
3. abort tx 1
4. begin tx 2, write data batch in tx 2, abort tx 2 which is offset 110
5. begin recovery, replicating data to node 2, sending records [0, 50]
6. roll segment
node 2:
7. receives data [0, 50], including [begin tx 1, data tx 1, data]
node 1:
8. self compact segment: since all abort records have the same key, only
abort tx 2 is kept
9. continue recovery, replicating data to node 2, sending records in the
range [50, 110], but is missing abort tx 1
node 2:
10. receives data [50, 110], and now has [begin tx 1, data tx 1, data,
begin tx 2, data tx 2, abort tx 2]
11. roll segment
12. self compact segment: rm_stm on this node doesn't know that tx 1 is
considered aborted
13. data tx 1 is left in the segment as committed
If instead, at step 8, we had kept [abort tx 1] as well, this wouldn't
have been an issue: node 2 would have replayed the abort and correctly
compacted away data tx 1.
This commit does this disallowing the deduplication of control batches
entirely. These batches are only used in transactions, and are expected
to be uncommon. This happens in the `is_compactible()` method, which is
used at compaction time to determine whether the entire batch should be
kept without consulting further record-based maps.
I'm explicilty doing it here rather than e.g. filtering control batches
when writing to the compaction index because we'll ultimately need to
filter here anyway: older versions of Redpanda will have these control
records in their compaction index.
(cherry picked from commit 4ce6552)
bharathv
changed the title
andrwng
approved these changes
Mar 15, 2024
1 task
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
This is a select backport of commit from #16295
Fixes: #16679
Backports Required
Release Notes
Bug Fixes