[zhongkechen]

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

Issue Link, if available

Reported OTel exception-attribution failures in child contexts, wait-for-condition, wait-for-callback, and chained invoke.

Description

Failed operation spans currently report DurableOperationException instead of the original stored error type, and Workflow/Invocation spans record additional exception events for the propagated failure. Both trace views now use ErrorObject.errorType for operation exception events, so a Java failure remains java.lang.RuntimeException and a callback rejection remains RejectedError. Workflow and Invocation retain their error status and description without exception events; attempt events remain intact.

Demo/Screenshots

Local regression tests reproduce all four operation scenarios and direct handler failures in both views. No cloud screenshots were captured.

Design

• Design note: not needed; this is a plugin-local bug fix with no public API change.
• Root cause: operation hooks intentionally expose stored failures through a DurableOperationException; recording that wrapper directly loses the stored type. Both plugins also record the propagated execution error on Workflow and Invocation.
• Simpler options considered: changing or reconstructing the SDK exception would affect caller/plugin contracts and cannot represent arbitrary remote-language types. A shared attribute override corrects telemetry at recording time while retaining native exception rendering and the existing timestamp policy.
• Net production source change: -1 line; public API members: unchanged. Four parent-span recordException calls are removed.

Assumptions

• Native OpenTelemetry exception recording honors additional exception attributes: verified against the real OpenTelemetry 1.67.0 provider in OperationExceptionEventTest.
• Checkpointed operation errors preserve their original type: verified through the real SDK and LocalDurableTestRunner for child contexts, conditions, callbacks, and invokes. Callback/invoke completions are supplied by the local backend.
• Live AWS/X-Ray validation of the updated SDK is unverified; the deployed conformance scenarios were not rerun.

Decisions not covered by the issue or the design note

• Keep the Java exception type when the stored type is absent, and preserve explicitly supplied empty types, consistent with existing error-detail handling.
• Keep exception messages, stack rendering, timestamps, caller exceptions, and checkpoint contents unchanged.
• Updated InvocationOtelPluginTest.failedInvocation_setsErrorOnBothWorkflowAndInvocationSpans: its old assertion required the extra Workflow exception event reported as a bug; it now checks that both parent spans have no events.
• Updated BackendOperationTimingTest.publicHandlerUsesBackendEndAndCompletedReplaySkipsHooks: its old assertion required the wrapper type; it now requires the stored type while retaining the timing and replay checks.
• No new conformance requirements or handlers were added. Event-type and parent-event assertions would be useful additions to the shared conformance suite.

Checklist

• I have filled out every section of the PR template
• I have thoroughly tested this change locally
• Every new branch has a test that fails without it
• Regression evidence: before the fix, 19 of 30 focused tests failed on wrapper types or parent-span events; after the fix, all 149 focused tests passed
• Every edited test expectation is listed above with the reason

Testing

Unit Tests

OperationExceptionEventTest covers Java and remote error types, absent/empty types, native timestamps, the epoch sentinel, and backend timestamps. Existing timing, metadata, virtual-context, and invocation tests pass.

Integration Tests

ExceptionAttributionIntegrationTest covers child-context, condition, callback, invoke, and direct-handler failures in both trace views, including preserved error status, original operation/attempt exception types, failure replay, and skipped completed user code.

Validation on JDK 21:

• mvn test: all modules passed; 3,689 tests reported, 34 skipped, zero failures/errors (including all 961 OTel tests).
• mvn -pl otel-plugin spotless:apply spotless:check: passed.
• git diff --check: passed.

The standard local suite leaves cloud tests disabled; live cloud conformance was not run.

Examples

Not applicable; no public usage change.