Add -ingest-storage.kafka.target-consumer-lag-at-startup support #8579
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
pracucci
force-pushed
the
add-target-consumer-lag-at-startup
branch
2 times, most recently
from
a86cb02
to
c3e0fc3
Compare
pracucci
commented
Jul 2, 2024
| } | ||
| } | ||
|
|
||
| return boff.Err() | ||
| // TODO should be moved to dskit's backoff |
There was a problem hiding this comment.
Note to reviewers: I will open a PR to add context.Cause() support to dskit's backoff but I prefer to not block on it.
There was a problem hiding this comment.
See: grafana/dskit#538
dimitarvdimitrov
approved these changes
Jul 2, 2024
| boff.Wait() | ||
| continue | ||
| } | ||
|
|
||
| // Send a direct request to the Kafka backend to fetch the last produced offset. | ||
| // We intentionally don't use WaitNextFetchLastProducedOffset() to not introduce further | ||
| // latency. | ||
| lastProducedOffsetRequestedAt := time.Now() |
There was a problem hiding this comment.
that was a subtle bug previously 👍
pracucci
force-pushed
the
add-target-consumer-lag-at-startup
branch
from
3919855
to
d2bd338
Compare
dimitarvdimitrov
approved these changes
Jul 2, 2024
tacole02
approved these changes
Jul 2, 2024
Signed-off-by: Marco Pracucci <[email protected]>
Signed-off-by: Marco Pracucci <[email protected]>
Signed-off-by: Marco Pracucci <[email protected]>
Signed-off-by: Marco Pracucci <[email protected]>
pracucci
force-pushed
the
add-target-consumer-lag-at-startup
branch
from
0cf787f
to
6de7564
Compare
Signed-off-by: Marco Pracucci <[email protected]>
Co-authored-by: Taylor C <[email protected]>
Thanks @tacole02 for the review! |
pstibrany
approved these changes
Jul 3, 2024
| // written to the partition, we give the reader maxLag time to replay the backlog + ingest the new data | ||
| // written in the meanwhile. | ||
| func() (time.Duration, error) { | ||
| timedCtx, cancel := context.WithTimeoutCause(ctx, maxLag, errWaitTargetLagDeadlineExceeded) |
There was a problem hiding this comment.
TIL about context.WithTimeoutCause, nice.
|
Going to merge this. Will address the TODO once dskit PR is merged. |
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.
What this PR does
As soon as an ingester switches to
Running, the reported end-to-end latency spikes high. We can see the correlation here:Why? Because the ingesters are configured with
-ingest-storage.kafka.max-consumer-lag-at-startup=15s(default), which means an ingester is guaranteed to have an end-to-end latency of up to 15s when switched to Running but there's no guarantee that the latency will be significantly lower than that. Significantly reducing-ingest-storage.kafka.max-consumer-lag-at-startupsetting (e.g. to a couple of seconds) is tricky because if the ingester can't catch up to that low latency then it will never startup.In this PR I'm proposing to add a new config option called
-ingest-storage.kafka.target-consumer-lag-at-startupand making it a best-effort. How it works:-ingest-storage.kafka.max-consumer-lag-at-startup(as inmain)-ingest-storage.kafka.target-consumer-lag-at-startup(defaults 2s). The grace period is equal to-ingest-storage.kafka.max-consumer-lag-at-startup; the rationale here is that we expect at least a 1x replay speed from Kafka.This means that
-ingest-storage.kafka.target-consumer-lag-at-startupis a best-effort, and-ingest-storage.kafka.max-consumer-lag-at-startupis guaranteed (as inmain).I've tested it in a Mimir dev cluster and looks fixing the issue:
Which issue(s) this PR fixes or relates to
N/A
Checklist
CHANGELOG.mdupdated - the order of entries should be[CHANGE],[FEATURE],[ENHANCEMENT],[BUGFIX].about-versioning.mdupdated with experimental features.