Repository navigation
Split internalServerErrorCount by wire-visibility - #3297
Merged
SophieGuo410 merged 6 commits intoAug 28, 2026
Merged
SophieGuo410 merged 6 commits into
SophieGuo410 merged 6 commits into
Conversation
nettyMetrics.internalServerErrorCount previously incremented unconditionally whenever a generic (non-RestServiceException, non-client-termination) exception was handled in NettyResponseChannel#getErrorResponse, even when the constructed 500 response could never actually be written to the client because response metadata (e.g. a 200) had already been committed for a streamed response. This conflated two different failure modes under one counter/alert: - a 500 that actually reached the client on the wire - a post-commit failure where the client instead sees a force-closed connection after a partially delivered 200 Add internalServerErrorAfterResponseCommittedCount to track the second case separately, so internalServerErrorCount now reflects only errors that were actually written to the wire. The sum of the two new counters equals what internalServerErrorCount counted before this change. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## master #3297 +/- ##
=============================================
- Coverage 64.24% 50.73% -13.52%
+ Complexity 10398 8707 -1691
=============================================
Files 840 939 +99
Lines 71755 80757 +9002
Branches 8611 9745 +1134
=============================================
- Hits 46099 40969 -5130
- Misses 23004 36388 +13384
- Partials 2652 3400 +748 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
… instead Replace the previous internalServerErrorCount / internalServerErrorAfterResponseCommittedCount split with the original, simpler behavior: internalServerErrorCount is incremented unconditionally for generic internal errors again, exactly as before this PR, so no existing alerting/dashboards keyed on it need to change. A 500 that could not be delivered to the client because response metadata (e.g. a 200) was already committed for a streamed GET is still a case worth tracking - but only distinctly for known offline (e.g. composite router secondary/parity-check) callers, since those already-committed-response drops for that traffic are expected and otherwise indistinguishable from a genuine client-facing incident. Add a new netty.server.offline.service.ids config listing the x-ambry-service-id values (as configured for those callers, e.g. in the composite router config) that identify offline traffic. When a request from one of those service IDs hits the already-committed-500 case, increment a new, additive NettyMetrics#offlineInternalServerErrorOnlyCount counter (OfflineInternalServerErrorOnlyCount.Count.rrd) instead of introducing a second alerting metric. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Read responseMetadataWriteInitiated after the CAS attempt in maybeSendErrorResponse(), not before. A concurrent writer (e.g. a router content-write callback) can commit response metadata on another thread between the earlier snapshot and our own CAS attempt, which could cause the offline metric to be silently skipped even though a response was already committed to the client. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
… case Adds offlineServiceUnavailableOnlyCount, incremented under the same conditions as offlineInternalServerErrorOnlyCount but for a genuine ServiceUnavailable (503) that never reached the wire because response metadata was already committed, for a request from a configured offline service ID. Host-level-throttled drops are excluded, matching how the existing serviceUnavailableErrorCount already excludes them from polluting SLO dashboards. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
…ed drops Mirrors offlineInternalServerErrorOnlyCount/offlineServiceUnavailableOnlyCount: tracks host-level-throttled 503 drops that never reached the wire because response metadata was already committed, for a request from a configured offline service ID. Kept as its own counter, matching how hostLevelThrottledCount is already tracked separately from serviceUnavailableErrorCount. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
beijxu
reviewed
Aug 28, 2026
….ids Previously an unconfigured (or trailing-comma) value produced a set containing a single empty string instead of a truly empty set, so the isEmpty() fast-path in isOfflineServiceRequest() was never hit by default. A stray space after a comma also produced an id that could never match via exact Set.contains(...). Trim each entry and drop empties so both cases are handled correctly. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
beijxu
approved these changes
Aug 28, 2026
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Summary
nettyMetrics.internalServerErrorCount previously incremented unconditionally whenever a generic (non-RestServiceException, non-client-termination) exception was handled in
NettyResponseChannel#getErrorResponse, even when the constructed 500 response could never actually be written to the client because response metadata (e.g. a 200) had already been committed for a streamed response.
This conflated two different failure modes under one counter/alert:
Add internalServerErrorAfterResponseCommittedCount to track the second case separately, so internalServerErrorCount now reflects only errors that were actually written to the wire. The sum of the two new counters equals what internalServerErrorCount counted before this change.
Testing Done
unit test