DR original loopback HTTP timeout and recovery workshop,2026-10-04
Run:python3 content/labs/micro-timeouts/run.py --output /tmp/dr-micro-timeouts.json
Choose a new output path; existing reports are not overwritten.
Recorded CPython3.13.1,SQLite3.51.2,Darwin arm64. Thirteen groups passed twice.
Two simultaneous listeners:teaching gateway and dependency. Original requests
use loopback only. A response is held AFTER the local SQLite commit. The gateway
actually times out while the effect remains. An unsafe retry duplicates it.
A protected retry consults a record committed with the effect and returns the
same effectId while the original response is still pending. Different units
under the same key return409. New keys or different fictional client scopes can
represent new intents. JSON field ordering does not alter the parsed intent.
A clean listener restart retains the same database in the same Python process.
Controlled ledger deletion allows a late duplicate; no automatic TTL exists.
Four listeners across the restart are closed, request threads are joined,
SQLite connections closed and the temporary directory removed. No external host,
authenticated client, distributed transaction, power-loss/crash test, external
effect, automatic retry/backoff/jitter, benchmark, OS process restart or human
workshop. This is teaching code, not a production proxy or security boundary.
