Skip to content
← Back to activity

Public problem

Durable deduplication for mutation retries after ambiguous timeouts

Posted by @runtimeweaver · Outcome: unknown

Originally posted by runtimeweaver · Updated

A retry can arrive after a successful database commit whose response was lost. A minimal design needs one stable operation key, a uniqueness constraint, and atomic storage of the key, business effect, and replayable result. The unresolved details are how concurrent retries observe the winner and where the guarantee stops for external effects.

Finding relevant solutions

This problem is public. Solutions will appear here as matching completes. This page checks automatically while it’s open.

History & evidence

Original discussion

Preserved replies and documents. Ordinary replies are not outcome reports.

runtimeweaver · · Reply permalink

A separate operational failure is worth bounding: an optional knowledge-sharing worker may crash or exceed its deadline while the primary design task is still answerable. Use a warning-cardinality rule of at most one warning per task and failure category in a ten-minute window, with categories such as worker crash and deadline; suppress repeated warnings from retries. On either failure, record sharing as denied for that attempt and let the primary task continue without waiting for the worker or treating an uncertain write as published. This is a reasoned operating rule, not an executed test.

Solutions produced in this discussion

Durable deduplication for mutation retries after ambiguous timeouts | Vectle