Retry failed GitHub release scans with backoff

A failed scan no longer advances last_synced_at; it schedules a retry from
60s with exponential backoff, honouring Retry-After/X-RateLimit-Reset.

internal/provider/syncretry.go

@@ -0,0 +1,75 @@
+package provider
+
+import (
+	"errors"
+	"fmt"
+	"net/http"
+	"strconv"
+	"time"
+)
+
+const (
+	syncRetryBase = time.Minute
+	syncRetryMax  = 10 * time.Minute
+)
+
+// UpstreamStatusError is a non-success upstream response. RetryAt is the
+// upstream's own retry hint (Retry-After, or X-RateLimit-Reset once the quota
+// is exhausted); zero when it gave none.
+type UpstreamStatusError struct {
+	URL     string
+	Status  int
+	RetryAt time.Time
+}
+
+func (e *UpstreamStatusError) Error() string {
+	return fmt.Sprintf("%s: status %d", e.URL, e.Status)
+}
+
+// NewUpstreamStatusError wraps a non-success response, capturing its retry hint.
+func NewUpstreamStatusError(url string, resp *http.Response) *UpstreamStatusError {
+	e := &UpstreamStatusError{URL: url, Status: resp.StatusCode}
+	if ra := resp.Header.Get("Retry-After"); ra != "" {
+		if secs, err := strconv.Atoi(ra); err == nil {
+			e.RetryAt = time.Now().Add(time.Duration(secs) * time.Second)
+		} else if t, err := http.ParseTime(ra); err == nil {
+			e.RetryAt = t
+		}
+	} else if resp.Header.Get("X-RateLimit-Remaining") == "0" {
+		if reset, err := strconv.ParseInt(resp.Header.Get("X-RateLimit-Reset"), 10, 64); err == nil {
+			e.RetryAt = time.Unix(reset, 0)
+		}
+	}
+	return e
+}
+
+// SyncResult is a background scan's outcome, recorded when its sync lease is
+// released. A failed scan keeps the prior sync time and ETag and schedules a
+// retry after Backoff, doubled per consecutive failure up to MaxBackoff, and
+// never earlier than RetryAt.
+type SyncResult struct {
+	Etag       string
+	Failed     bool
+	RetryAt    time.Time
+	Backoff    time.Duration
+	MaxBackoff time.Duration
+}
+
+// NewSyncResult builds the result for a scan against a remote with the given
+// mutable_ttl. The retry cap stays well below ttl, and an upstream hint is
+// clamped to ttl so a bogus reset can never stall the remote longer than a
+// normal sync interval would.
+func NewSyncResult(etag string, scanErr error, ttl time.Duration) SyncResult {
+	if scanErr == nil {
+		return SyncResult{Etag: etag}
+	}
+	res := SyncResult{Failed: true, Backoff: syncRetryBase, MaxBackoff: min(syncRetryMax, max(syncRetryBase, ttl/4))}
+	var se *UpstreamStatusError
+	if errors.As(scanErr, &se) && !se.RetryAt.IsZero() {
+		res.RetryAt = se.RetryAt
+		if limit := time.Now().Add(ttl); res.RetryAt.After(limit) {
+			res.RetryAt = limit
+		}
+	}
+	return res
+}