DigitalOcean API rate‑limit headers and connection‑pool exhaustion behavior

When integrating with the DigitalOcean API, a client must honor the X-RateLimit-Remaining and X-RateLimit-Reset headers to avoid exhausting its HTTP connection pool. The goal is to define a reliable back‑off approach that works for both the official SDKs and custom HTTP clients.

Constraints include a single global limit per OAuth token or API key, no per‑endpoint limits exposed by the control plane, and the possibility that rate‑limit policies change without notice. These factors make it difficult to design a static retry schedule.

What back‑off strategy is recommended when X-RateLimit-Remaining reaches zero? Does the platform provide any signal for per‑endpoint limits that could inform finer‑grained throttling? How should custom clients coordinate retries across multiple connection pools to prevent saturation?