Edge Cases#
This page covers integration situations that usually cause regressions.
1. Exploratory probes produce noisy errors#
Problem:
helper functions that intentionally test many formats raise frequent exceptions.
Fix:
mark those with
@signal(exception_level="DEBUG"),keep assertive parsing paths at
ERROR.
An inner signal whose event is filtered does not consume the exception’s diagnostic observation. An eligible enclosing signal can still record that failure. Once a signal accepts an observation, enclosing signals avoid duplicate events; an accepted DEBUG observation is not promoted to ERROR by an outer wrapper. Application enablement, source silence and the current level threshold control acceptance. Coalescing and duplicate policies retain their own behavior.
2. Duplicate warnings through logging bridge#
Problem:
warning appears twice when logging/warnings interception is misconfigured.
Fix:
rely on SMonitor capture policy,
avoid custom parallel warning bridges unless required.
3. Contract mismatch between SIGNALS and emitted extra#
Problem:
SIGNALSrequires fields that emitters do not pass.
Fix:
in QA, enable
strict_signals=True,add targeted tests for each critical signal.
4. Silent emission failure#
Problem:
emission errors are swallowed and observability disappears.
Fix:
never use
except Exception: passin diagnostics paths,provide explicit fallback warning/log line with context.
5. Over-instrumentation#
Problem:
adding
@signaleverywhere increases overhead and noise.
Fix:
instrument API boundaries and orchestration,
skip tight loops and low-level utility hot paths.
6. Conflicting runtime configuration#
Problem:
tests, notebooks, and scripts override profile differently.
Fix:
be explicit in each execution context,
remember precedence: runtime > env >
_smonitor.py.
Next#
Continue with Troubleshooting.