fix: retry OTLP gRPC responses with trailer status

[efegokdemir]

Problem

OkHttpGrpcSender.isRetryable only inspected the grpc-status response header. When a retryable status such as UNAVAILABLE was delivered in HTTP/2 trailers, the exporter reported the status but did not retry.

Change

Read grpc-status from trailers when it is absent from headers, matching the existing status-resolution path. An IOException while reading trailers remains non-retryable. Added a regression test and an Unreleased changelog entry.

Fixes #8843

Validation

• Baseline regression test on upstream/main: failed as expected (false instead of true).
• JAVA_HOME=/opt/homebrew/opt/openjdk@21 ./gradlew :exporters:sender:okhttp:test --tests io.opentelemetry.exporter.sender.okhttp.internal.OkHttpGrpcSenderTest: passed (49 tests).
• JAVA_HOME=/opt/homebrew/opt/openjdk@21 ./gradlew :exporters:sender:okhttp:spotlessCheck: passed.
• git diff --check: passed.
• :exporters:sender:okhttp:check: one unrelated existing timing-sensitive failure in shutdown_CompletableResultCodeShouldWaitForThreads; the changed regression and remaining tests passed.

Provenance

Codex assisted with repository analysis, implementation, regression testing, and validation. The submitter reviewed the complete diff and is responsible for the contribution.

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: efegokdemir / name: Efe Gökdemir (a7ee760, 65032eb, 7d5bbe9, 9f3bc14, b95dd4d, d0e3ead, d129565)

[opentelemetry-pr-dashboard]

Pull request dashboard status

Merged · refreshed 2026-10-08 22:29 UTC

Status above doesn't look right?

• Anything look wrong? Report it with what you expected; it helps us improve the dashboard.

[codecov]

Codecov Report

❌ Patch coverage is 91.45299% with 10 lines in your changes missing coverage. Please review.
✅ Project coverage is 90.64%. Comparing base (a13bc21) to head (023bcc0).
⚠️ Report is 55 commits behind head on main.

Files with missing lines	Patch %	Lines
...orter/sender/okhttp/internal/OkHttpGrpcSender.java	83.33%	5 Missing and 5 partials ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##               main    #8854      +/-   ##
============================================
- Coverage     91.44%   90.64%   -0.80%     
- Complexity    10667    10876     +209     
============================================
  Files          1007     1015       +8     
  Lines         28686    29712    +1026     
  Branches       3676     3874     +198     
============================================
+ Hits          26231    26932     +701     
- Misses         1657     1863     +206     
- Partials        798      917     +119     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[efegokdemir]

Fixed the patch-caused response-size regression in be1cea4. The new resolved-response retry path was retrying the sender-generated RESOURCE_EXHAUSTED response used when the local response-body limit is exceeded; that allowed the existing response-body-bound tests to observe a later successful attempt.

handleResponse now marks only responses decoded from the server as retryable candidates. Local size-limit, unsupported-encoding, and invalid-frame responses remain terminal failures.

Validation: git diff --check passed. The focused Gradle test could not run locally because this environment has no Java runtime; CI should validate the focused and matrix suites.

[efegokdemir]

I inspected the current required-status-check and failed matrix logs. The failures are broad across macOS, Ubuntu, and Windows and across JDK 8, 11, 17, 21, 25, and 27. The logs show unrelated SDK metrics stress-test timeouts (for example SynchronousInstrumentStressTest partialWriteStressTest) and the required-status-check itself is an intentional exit 1; security/API checks pass. I found no evidence that the PR changes caused these failures, so I made no unrelated code change.

[efegokdemir]

The retry regression was caused by OkHttp treating bodyless gRPC responses with an explicit trailer status as non-retryable. Updated the sender to preserve the decoded trailer status and mark only explicit server statuses retryable, while local size-limit/encoding failures remain non-retryable. Added focused coverage for an empty body with grpc-status: 14. New HEAD: 3f95082a. Local validation: :exporters:sender:okhttp:test PASS, :exporters:otlp:profiles:test --tests ...retryableError_tooManyAttempts PASS, spotlessCheck PASS, git diff --check PASS.

[otelbot]

Thank you for your contribution @efegokdemir! 🎉 We would like to hear from you about your experience contributing to OpenTelemetry by taking a few minutes to fill out this survey.