prisma-risk / tsoracle / 26349625509

95%

Build Type

push

github

Committed by web-flow

Commit Message

fix(client): bound the connect+RPC pair by one per-attempt deadline (#238) (#271)

`attempt` wrapped both the connect phase (`pool.client`) and the RPC phase
(`get_ts`) in `timeout(budget, ..)` with the *same* `budget`, so a single
`(connect, get_ts)` pair could take up to `2 * per_attempt_deadline`. That
contradicts the documented contract (the deadline applies to the pair) and
lets a slow-connecting endpoint overrun `overall_deadline` before
`max_attempts` is exhausted.

Anchor one `pair_deadline = Instant::now() + budget` before the connect and
give `get_ts` only `pair_deadline.saturating_duration_since(Instant::now())`
— the time the connect left behind (floored at zero, so a fully-consumed
budget yields an immediate RPC timeout rather than a fresh one). The
RPC-timeout error and trace now report the remaining share rather than the
full budget.

A new test drives `attempt` with an injected connector that sleeps for most
of the budget before returning a live channel, points it at a server whose
`get_ts` hangs, and asserts the whole pair completes within ~1x the
per-attempt deadline rather than ~2x.

Coverage Stats

50 of 56 new or added lines in 1 file covered. (89.29%)

6 existing lines in 2 files now uncovered.

10496 of 11079 relevant lines covered (94.74%)

500789.79 hits per line