PR #418 — payments: jittered exponential backoff for provider retries
Retries used to fire on a fixed 2-second beat, so a slow provider got a synchronised burst from every pod at once. This moves retry handling into the shared transport, spreads the waits out with jitter, caps them, and decides in one place what is even safe to repeat. Boxes are named in plain English. Click a ⤢ box to zoom into its parts, or a </> box to see the real code (L4). Flip Before ⇄ After or ▶ Play a request, then pass the check to mark yourself approval-ready.
BeforeAfter
NewModifiedUnchanged⤢ zoom into parts</> see the code
Hover a box for a one-line summary. Click a ⤢ box to zoom into its parts, or a </> box to open its code.
Approval check
Prove you understand the change before approving. Twelve questions, one at a time — the diagram is hidden once you start, so know it first. You need all twelve correct; miss any and you go back to the diagram and retake until you ace it.