fix(tracing): fail open temporal span activities

[danielmillerp]

Summary

• make Temporal adk.tracing.start_span fail open and return None when the tracing activity fails
• make Temporal adk.tracing.end_span fail open and return the original span when the tracing activity fails
• add regression coverage for Temporal tracing activity timeout/failure behavior

Tests

• rye run pytest tests/lib/adk/test_tracing_module.py -n 0 -p no:cacheprovider
• rye run ruff check --no-cache src/agentex/lib/adk/_modules/tracing.py tests/lib/adk/test_tracing_module.py

Greptile Summary

This PR makes the Temporal-path start_span and end_span tracing activities fail open rather than crashing the workflow when they time out or error. It also fixes StateMachine.reset_to_initial_state to guard the span usage against the newly-possible None return, matching the pattern already present in step().

• start_span now catches ActivityError/TemporalTimeoutError (re-raising cancellations) and returns None; end_span returns the original span unchanged on the same errors; both emit a bounded agentex.tracing.temporal_span_activity.dropped metric.
• reset_to_initial_state initializes span = None and gates the span.output assignment and end_span call behind span is not None, preventing an AttributeError when tracing fails open during a reset.
• Six new unit tests cover the temporal fail-open path, propagation of unexpected errors, propagation of cancellation signals, the context-manager short-circuit, and an integration-style state_machine regression test.

Confidence Score: 5/5

Safe to merge — the fail-open logic is tightly scoped to Temporal activity errors, cancellations re-propagate correctly, and all direct callers of start_span now guard the None return before using the span.

The change is a focused defensive fix with no alterations to the happy path. The null-return contract is consistently handled across every direct caller (step() already guarded, reset_to_initial_state is fixed here, the span context manager uses if span:). Cancellation propagation is explicitly preserved. New tests cover the key branches including the regression path in state_machine.

No files require special attention.

Important Files Changed

Filename	Overview
src/agentex/lib/adk/_modules/tracing.py	Adds fail-open exception handling for Temporal activity errors in start_span and end_span, with cancellation re-propagation and a bounded dropped-metric counter.
src/agentex/lib/sdk/state_machine/state_machine.py	Fixes reset_to_initial_state to guard span usage with span is not None, matching the existing pattern in step(); no logic changes beyond the null guard.
tests/lib/adk/test_tracing_module.py	Adds six new tests covering start/end fail-open, context-manager short-circuit, unexpected-error propagation, and cancellation propagation for both start and end paths.
tests/lib/test_state_machine.py	New regression test confirming reset_to_initial_state completes successfully and skips end_span when start_span returns None.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[start_span / end_span called] --> B{in_temporal_workflow?}
    B -- No --> C[Call _tracing_service directly\nraises on error]
    B -- Yes --> D[execute_activity via ActivityHelpers]
    D -- Success --> E[Return Span]
    D -- ActivityError / TemporalTimeoutError --> F{is_cancelled_exception?}
    F -- Yes --> G[Re-raise — workflow cancellation\nmust propagate]
    F -- No --> H[workflow.logger.warning + metric]
    H --> I{start_span or end_span?}
    I -- start_span --> J[Return None\nfail open]
    I -- end_span --> K[Return original span\nfail open]
    J --> L[Callers guard span is not None\nbefore mutating or calling end_span]
    K --> M[Caller receives original span\nworkflow continues]

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A[start_span / end_span called] --> B{in_temporal_workflow?}
    B -- No --> C[Call _tracing_service directly\nraises on error]
    B -- Yes --> D[execute_activity via ActivityHelpers]
    D -- Success --> E[Return Span]
    D -- ActivityError / TemporalTimeoutError --> F{is_cancelled_exception?}
    F -- Yes --> G[Re-raise — workflow cancellation\nmust propagate]
    F -- No --> H[workflow.logger.warning + metric]
    H --> I{start_span or end_span?}
    I -- start_span --> J[Return None\nfail open]
    I -- end_span --> K[Return original span\nfail open]
    J --> L[Callers guard span is not None\nbefore mutating or calling end_span]
    K --> M[Caller receives original span\nworkflow continues]

Loading

_{Reviews (2): Last reviewed commit: "fix(tracing): fail open temporal span ac..." | Re-trigger Greptile}

[danielmillerp]

Addressed the Greptile nullable-span finding in fa5d4a9.

The issue was that adk.tracing.start_span(...) can now fail open and return None, but StateMachine.reset_to_initial_state() was still mutating span.output unconditionally. Updated reset to mirror step(): initialize span = None and only set output / call end_span when a span was actually created.

Added regression coverage in tests/lib/test_state_machine.py for the reset path where start_span returns None.

[danielmillerp]

Added the span-drop metric nit in e90a03e.

On Temporal tracing activity fail-open, we now emit replay-safe workflow metrics via workflow.metric_meter():

agentex.tracing.temporal_span_activity.dropped

with event_type=start or event_type=end. Cancellation still propagates and is not counted as a dropped span.