Field Notes
Reading a vendor timeout as an architectural signal
A slow partner is not only an ops ticket. It tells you whether your boundary assumed synchronicity the world will not provide.
On-call notes full of upstream_timeout look like infrastructure weather. In an Integration Boundary Assessment we treat them as design feedback.
If your order flow cannot proceed without a synchronous tax or fraud call, a partner slowdown becomes a customer-visible outage. Sometimes that coupling is required by law or risk policy. Often it is leftover from a prototype that never grew a queue or a deferred path.
Questions we put on the whiteboard:
- Can the user-facing action complete with a pending state?
- Who reconciles when the partner eventually answers?
- Is the timeout tuned to the partner’s documented SLA or to a guess from 2019?
Timeouts will keep happening. The architectural choice is whether they cascade.