draft
lore:python-async/retry-selective-exceptions/0007 (v1.0.0)
Catching generic Exception in async retry loops swallows non-retryable/fatal errors, causing wasted retries, latency spikes, and silent failures; restrict retry catches strictly to allowlisted exception types (e.g., `except retryable as e:` or explicitly re-raise).
Verbatim Symptoms (Primary Search Key)
- Fatal errors (e.g., auth failure, bad request, validation error) are retried repeatedly instead of failing fast
- Non-retryable exceptions take multiple retry delays before propagating
- Side-effects repeated on permanent failure conditions
Problem Statement
A common mistake when writing retry loops is using `except Exception as e:` which catches every exception indiscriminately. Permanent/fatal errors (like ValueError, TypeError, AuthenticationError) should never be retried, as retrying them wastes resources, causes unnecessary delay, and may repeat non-idempotent side effects.
Correct Procedure
- Pass a tuple of allowed retryable exception classes (e.g., `retryable=(TransientNetworkError, TimeoutError)`) to the retry helper.
- Use `except retryable as e:` in the retry block so non-matching exceptions bypass the catch and propagate immediately.
- If using a dynamic exception tuple, ensure it is passed as a tuple to `except` or checked with `isinstance(e, retryable)` and re-raised immediately if False.
Anti-Patterns
- Using `except Exception:` inside retry loops without re-raising non-retryable errors.
- Assuming all async exceptions are transient network glitches.
- Catching BaseException, which intercepts CancelledError and KeyboardInterrupt.
Documented Dead Ends
- Catching Exception and sleeping before checking status: Delays propagation of fatal errors and causes redundant executions. (~25 mins wasted)