Fix data loss consistency violation #6019
Merged
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
bharathv
requested changes
Aug 15, 2022
dotnwat
reviewed
Aug 16, 2022
|
PartitionBalancerTest.test_full_nodes - #5884 |
dotnwat
previously approved these changes
Aug 19, 2022
| help='Path to the log desired to be analyzed') | ||
| parser.add_argument( | ||
| '--path', | ||
| type=str, |
There was a problem hiding this comment.
if you'd like to integrate more tightly with argument parsing, you can change type = str to type = validate_path and validate_path to:
def validate_path(path):
if not os.path.exists(path):
raise ArgumentTypeError(f"Path doesn't exist {path}")
controller = join(path, "redpanda", "controller")
if not os.path.exists(controller):
raise ArgumentTypeError(f"Each redpanda data dir should have controller piece but {controller} isn't found")
return path
Kafka API doesn't have explicit begin txn API, when a transaction
coordinator recieves first add_partition or add_group it starts a
transaction. Also Redpanda defers disk flushes to the commit moment.
Combinations of those thing caues a problem:
1. a client issues add_partition to txn coordinator
2. txn coordinator starts a transaction (this is in memory state)
3. the client writes a message to data partition
4. redpanda treats the write as acks=1 and acks the request before
the replication is finished (it's safe to do it because on commit
redpanda checks that all pending replication is done)
5. txn coordinator & data partition experience re-election and the
in memory state is lost
6. the client issues add_group to txn coordinator
7. txn coordinator is unaware about the ongoing txn and starts new
transaction
8. the client commits the txn and only the consumer group change is
written
Since the data record was written with acks=1 just before re-election
it gets lost and the client hasn't figured out that there was some-
thing wrong with transaction.
The fix is to write to the txn coordinator's log when a transaction
starts; in this case when a crush-induced re-election happens new txn
coordinator has an opportunity to detect an ongoing txn and fail it.
the checks didn't handle well empty dirs
|
SIPartitionMovementTest.test_shadow_indexing - #4702 |
This pull request was closed.
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.
Cover letter
Kafka API doesn't have explicit begin txn API, when a transaction coordinator recieves first add_partition or add_group it starts a transaction. Also Redpanda defers disk flushes to the commit moment. Combinations of those thing caues a problem:
Since the data record was written with acks=1 just before re-election it gets lost and the client hasn't figured out that there was something wrong with transaction.
Fixes #6018
Backport Required
UX changes
Release notes
Bug Fixes