The call volume was not an attack. It was a retry loop with no backoff, shipped in a point release, running inside a DAW that reinstantiates plugins on every project load. It was entirely our fault for not making the failure mode obvious.
The immediate fix
We moved validation to the edge and made cached entitlements the default path. A validated licence now answers locally until its lease expires, which collapsed steady-state traffic by roughly two orders of magnitude.
The real fix
Rate limits protect us, not the customer. The durable fix was making the SDK refuse to hot-loop: exponential backoff is now built in and not configurable, because every configurable backoff eventually gets configured to zero.
- Edge-cached entitlement leases with a default 24 hour TTL
- Mandatory exponential backoff in the client
- Per-product quotas surfaced in the dashboard before they are hit
- Alerting on validation spikes, sent to the vendor and not just to us