Why the retry reflex gets learned on requests where it cannot do harm, and what changes as soon as the call writes something.
22 February 2027
A retry assumes the request had no effect before it failed. For a GET that assumption holds: nothing was written, so the second attempt reads the same row and the caller waits a few hundred milliseconds longer. The assumption is nowhere in the code. client.get(url) and client.post(url, body) are the same shape at the call site, and neither carries a note about what happens if it runs twice. From week 2 we remove one part of the assumption at a time — whether the call writes, how long the retry waits, how many other clients are retrying alongside you — and the same three lines stop being correct.
Outline#
- what the course covers, and why it is built around twelve specific failures
- why retrying a read works, and what that teaches by accident
- what changes in week 2
Teaching team
- Marisol Quaye — convenor