Timeouts, Retries & Backoff

Timeouts, retries, and backoff

Distributed calls fail; the professional questions are how long to wait and when to try again:

• Timeout every call — a missing timeout turns a slow dependency into a stopped service. HttpRequest.timeout(), Future.get(3, SECONDS), or orTimeout() on a CompletableFuture stage.
• Retry idempotent operations — GETs, yes; POST payments, only with idempotency keys.
• Exponential backoff with cap — 100ms, 200ms, 400ms… capped at a few seconds; retry storms make outages worse.
• Distinguish error classes — 4xx (your fault) usually isn't retryable; 5xx and timeouts usually are; 429 means slow down (respect Retry-After).

static String getWithRetry(HttpTransport t, HttpRequest req) throws Exception {
    int[] delays = {100, 200, 400};
    for (int attempt = 0; ; attempt++) {
        try {
            HttpResponse<String> res = t.send(req);
            if (res.statusCode() < 500) return res.body();  // don't retry 4xx
            if (attempt == delays.length) throw new IllegalStateException("gave up");
        } catch (java.io.IOException e) {
            if (attempt == delays.length) throw e;
        }
        Thread.sleep(delays[attempt]);
    }
}