kube/certs: use Let's Encrypt's recommended retry schedule (#20292)

Replace the doubling backoff (1m, 2m, 4m, ...) with LE's recommended
1m, 10m, 100m, daily. The old schedule burned retry attempts inside
the rate-limit window without speeding recovery.

Updates #20288
Updates #19895

Signed-off-by: chaosinthecrd <tom@tmlabs.co.uk>

kube/certs/certs.go

@@ -94,6 +94,16 @@ func (cm *CertManager) EnsureCertLoops(ctx context.Context, sc *ipn.ServeConfig)
 	return nil
 }
 
+// retrySchedule is the wait between successive failed issuance attempts,
+// following LE's recommended schedule.
+// https://letsencrypt.org/docs/integration-guide/#retrying-failures
+var retrySchedule = []time.Duration{
+	1 * time.Minute,
+	10 * time.Minute,
+	100 * time.Minute,
+	24 * time.Hour,
+}
+
 // runCertLoop:
 // - calls localAPI certificate endpoint to ensure that certs are issued for the
 // given domain name
@@ -103,13 +113,8 @@ func (cm *CertManager) EnsureCertLoops(ctx context.Context, sc *ipn.ServeConfig)
 // Note that renewal check also happens when the node receives an HTTPS request and it is possible that certs get
 // renewed at that point. Renewal here is needed to prevent the shared certs from expiry in edge cases where the 'write'
 // replica does not get any HTTPS requests.
-// https://letsencrypt.org/docs/integration-guide/#retrying-failures
 func (cm *CertManager) runCertLoop(ctx context.Context, domain string) {
-	const (
+	const normalInterval = 24 * time.Hour // regular renewal check
-		normalInterval   = 24 * time.Hour  // regular renewal check
-		initialRetry     = 1 * time.Minute // initial backoff after a failure
-		maxRetryInterval = 24 * time.Hour  // max backoff period
-	)
 
 	if err := cm.waitForCertDomain(ctx, domain); err != nil {
 		// Best-effort, log and continue with the issuing loop.
@@ -153,15 +158,11 @@ func (cm *CertManager) runCertLoop(ctx context.Context, domain string) {
 				nextInterval = normalInterval
 			} else {
 				retryCount++
-				// Calculate backoff: initialRetry * 2^(retryCount-1)
+				idx := retryCount - 1
-				// For retryCount=1: 1min * 2^0 = 1min
+				if idx >= len(retrySchedule) {
-				// For retryCount=2: 1min * 2^1 = 2min
+					idx = len(retrySchedule) - 1
-				// For retryCount=3: 1min * 2^2 = 4min
-				backoff := initialRetry * time.Duration(1<<(retryCount-1))
-				if backoff > maxRetryInterval {
-					backoff = maxRetryInterval
 				}
-				nextInterval = backoff
+				nextInterval = retrySchedule[idx]
 				cm.logf("Error refreshing certificate for %s (retry %d): %v. Will retry in %v\n",
 					domain, retryCount, err, nextInterval)
 			}