-
Notifications
You must be signed in to change notification settings - Fork 17
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
crash instantly on data loss errors #56
Conversation
nvartolomei
commented
Oct 15, 2024
•
edited
Loading
edited
1439144
to
99c7d5b
Compare
See kgo PR for additional details redpanda-data/kgo-verifier#56
See kgo PR for additional details redpanda-data/kgo-verifier#56
i realize this is draft i didn't intend to hit approve, but it also looked ok. |
I suspect false positives on data loss in franz-go https://github.com/twmb/franz-go/blob/master/pkg/kgo/consumer.go#L2283, so I'll check with Travis. |
@bashtanov @nvartolomei what's the status on this pr? |
Most probably the failures are not related to franz-go, so let's merge it |
@nvartolomei ping |
There’s a failure in an upgrade test happening with an old version. Want to
figure that one out to avoid introducing ci noise.
…On Fri, 18 Oct 2024 at 22:09, Noah Watkins ***@***.***> wrote:
@nvartolomei <https://github.com/nvartolomei> ping
—
Reply to this email directly, view it on GitHub
<#56 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAEETWMYCEIEBQQOJSBVLITZ4F2H3AVCNFSM6AAAAABP7AXDWSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMRTGIZTMMJTGE>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
7f0f3a9
to
8ac4000
Compare
See kgo PR for additional details redpanda-data/kgo-verifier#56
See kgo PR for additional details redpanda-data/kgo-verifier#56
We (nvartolomei and bash) found that GroupReadWorker wouldn't report data loss. The reason is that we didn't fail the test instantly when data loss is detected instead we relied on monotonicity validation to fail. I.e. we would crash when we detected that we would consume an earlier offset compared to what we consumed before. In the GroupReadWorker we reset the monotonicity validation state on any errors because the next attempt at consuming is almost certain to read the second time offsets already consumed. As a result we wouldn't catch monotonicity issues. In retrospect, it would have been better to fail instantly when franz-go detects data loss. So this commit achieves exactly that.
8ac4000
to
7d45f83
Compare
See kgo PR for additional details redpanda-data/kgo-verifier#56
Includes the following changes: * redpanda-data/kgo-verifier#56