oss_12_onending_frozen

[ocelotl]

Closes #11

Fixes SpanProcessor.on_ending mutability (Linear OSS-12, finding T1): an intermediate 'ending' state lets on_ending mutate the span (from the ending thread) before it is frozen; ConcurrentMultiSpanProcessor.on_ending now runs synchronously on the ending thread. Attributes/events/links added in on_ending are no longer dropped.

Linear issue: https://linear.app/dash0/issue/OSS-12/t1-spanprocessoronending-runs-after-the-span-is-frozen-mutability-must

[ocelotl]

🔒 Internal (dash0) — not for upstream.

• Linear: https://linear.app/dash0/issue/OSS-12/t1-spanprocessoronending-runs-after-the-span-is-frozen-mutability-must (finding T1)
• Closes: oss_12_onending_frozen #11 · Branch/worktree: oss_12_onending_frozen · Commit: 26417e1e1
• Validation: span-processor suite 28 passed; full SDK suite 832 passed; scope contained to opentelemetry-sdk + changelog.

[ocelotl]

📣 Public-facing draft — to be opened upstream in open-telemetry/opentelemetry-python. No internal references; copy verbatim.

Title: Allow SpanProcessor.on_ending to mutate the span before it is frozen

What

Introduce an intermediate "ending" state in Span.end(). The span records its end time and enters the ending state (attributes remain mutable) while on_ending runs; a finally block then finalizes it (freezes attributes) so it always ends read-only, even if a hook raises. A guard permits mutation during the ending window only from the ending thread. ConcurrentMultiSpanProcessor.on_ending now runs its hooks synchronously on the ending thread.

Why

Previously the span was frozen before on_ending ran, so any attribute/event/link a processor added in the hook was silently dropped — a spec violation (OnEnding must observe a still-mutable span). The thread-pool fan-out in ConcurrentMultiSpanProcessor was fundamentally incompatible with "mutable only from the ending thread" and also contradicted the documented "on_ending is synchronous" contract.

Testing

New tests prove an attribute, event, and link added inside on_ending survive into the exported span (for both the standard multi-span processor and the concurrent one), and that mutations attempted after end (outside on_ending) are still rejected. Full SDK suite passes (832).