changefeedccl: fix job progress regressing #37009
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
When the ChangeFrontier processor for a changefeed is started up, it creates a spanFrontier to track the overall progression for the feed. Each watched span is set to the empty timestamp so we can tell the difference between a rangefeed being behind and one still initializing. However, rangefeed will happily return resolved timestamps less than the one it was started with. If we feed this into the spanFrontier, it can report a timestamp behind the one everything started up at. In the case of a new changefeed, this is fine (though perhaps odd). In the case of a changefeed being restarted from the job's high-water, this means we can overwrite the high-water progress with the lower value, regressing it. At the least, this is very surprising UX, but it's easy to argue that it's a correctness issue. Touches cockroachdb#36879 Release note (bug fix): Fixed a bug where `CHANGEFEED` job progress would regress when the job was restarted.
|
This change is |
nvanbenschoten
approved these changes
Apr 23, 2019
There was a problem hiding this comment.
Reviewed 2 of 2 files at r1, 5 of 5 files at r2.
Reviewable status:complete! 0 of 0 LGTMs obtained (waiting on @danhhz)
pkg/roachpb/api.proto, line 1766 at r2 (raw file):
// // Note that these resolved timestamps may be lower than the timestamp used in // the RangeFeedRequest used to start the RnageFeed.
s/RnageFeed/RangeFeed/
I was surprised that rangefeeds would return resolved timestamps from before the request timestamp and this expectation mismatch was a component in a changefeed bug. It's not clear yet if the contract is wrong or simply underdocumented, so document it louder for now and leave a TODO to reconsider the contract when rangefeeds have more users. Also a very small cleanup on the `(*DistSender).RangeFeed` api while I'm in here. Release note: None
danhhz
commented
Apr 23, 2019
There was a problem hiding this comment.
TFTR!
bors r=nvanbenschoten
Reviewable status:
pkg/roachpb/api.proto, line 1766 at r2 (raw file):
Previously, nvanbenschoten (Nathan VanBenschoten) wrote…
s/RnageFeed/RangeFeed/
Done
craig bot
pushed a commit
that referenced
this pull request
Apr 23, 2019
37009: changefeedccl: fix job progress regressing r=nvanbenschoten a=danhhz When the ChangeFrontier processor for a changefeed is started up, it creates a spanFrontier to track the overall progression for the feed. Each watched span is set to the empty timestamp so we can tell the difference between a rangefeed being behind and one still initializing. However, rangefeed will happily return resolved timestamps less than the one it was started with. If we feed this into the spanFrontier, it can report a timestamp behind the one everything started up at. In the case of a new changefeed, this is fine (though perhaps odd). In the case of a changefeed being restarted from the job's high-water, this means we can overwrite the high-water progress with the lower value, regressing it. At the least, this is very surprising UX, but it's easy to argue that it's a correctness issue. Touches #36879 Release note (bug fix): Fixed a bug where `CHANGEFEED` job progress would regress when the job was restarted. 37030: changefeedccl: log more details when returning with non-retryable errors r=tbg a=danhhz The `%v` and `%+v` were switched from what I intended in the line that logs retryable errors and the one that handles non-retryable errors. Release note: None Co-authored-by: Daniel Harrison <daniel.harrison@gmail.com>
Build succeeded |
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.
When the ChangeFrontier processor for a changefeed is started up, it
creates a spanFrontier to track the overall progression for the feed.
Each watched span is set to the empty timestamp so we can tell the
difference between a rangefeed being behind and one still initializing.
However, rangefeed will happily return resolved timestamps less than the
one it was started with. If we feed this into the spanFrontier, it can
report a timestamp behind the one everything started up at. In the case
of a new changefeed, this is fine (though perhaps odd). In the case of a
changefeed being restarted from the job's high-water, this means we can
overwrite the high-water progress with the lower value, regressing it.
At the least, this is very surprising UX, but it's easy to argue that
it's a correctness issue.
Touches #36879
Release note (bug fix): Fixed a bug where
CHANGEFEEDjob progresswould regress when the job was restarted.