providerConnectRetries top-level constant
Bounded connect-stall retry budget (issue #1121): how many times a
request the connect watchdog killed with ZERO response bytes is re-sent
before the failure escalates to the run. 2 retries (3 total attempts):
the connect leg alone costs ≈ 3×(180s + 1s+2s backoff) ≈ 9 min against
an endpoint that black-holes. This leg sits BELOW TransientRetryStream
(3 attempts, 5s apart) and the roles failover ladder, so a hard-down
endpoint stacks ≈ 3 connect legs × 3 transient attempts ≈ 27 min per
role entry before failover (pre-PR: ≈ 9 min), and the ladder repeats
per queue entry — the recovery path of last resort stays the ladder's
own budgets, and environments that prefer faster failure shrink
providerTimeouts: connect: in ~/.fah/config.yaml (issue #1036),
which this retry inherits automatically. This budget is NOT the roles
failover ladder's — a connect stall never consumes a failover attempt.
Implementation
const providerConnectRetries = 2;