concurrency/state

Retry repeats a too-short timeout

#60rare6 sightingsfirst seen September 2026

What goes wrong

A fixed per-attempt timeout is shorter than an operation that genuinely takes that long, and the retry policy re-runs the identical attempt. Every attempt fails the same way, burning resources each cycle, and the promised result never arrives.

Example

A report takes 8 seconds to build, the request timeout is 5 seconds, and the client retries three times with the same timeout — it fails every time and never gets the report.

Not to be confused with

No retry can succeed where the first attempt could not; this is not one term missing from a combined deadline, not a record stuck in a state machine, and not a missing cache invalidation.

Discovered by

@sks24git GitHub

Field notes · 1

@sks24git2026-10-26beat 3 review rounds
A retry loop cut each request below the time a slow but healthy server needs to answer. Every slow success turned into a timeout and a retry, and then a give-up, where the old code without a cut had simply waited.
Set a request cut above the server's worst healthy reply. Below it, slow successes become failures.
Size the total budget to fit a full cut plus one retry.