internal/proxy/engine.go

@@ -31,6 +31,7 @@ type Engine struct {
 	cache     *cache.Redis
 	store     *storage.S3
 	cas       *storage.CAS
+	circuit   *CircuitBreaker
 	accessLog chan database.AccessLogEntry
 }
 
@@ -40,6 +41,7 @@ func NewEngine(db *database.DB, c *cache.Redis, s *storage.S3) *Engine {
 		cache:     c,
 		store:     s,
 		cas:       storage.NewCAS(s),
+		circuit:   NewCircuitBreaker(c),
 		accessLog: make(chan database.AccessLogEntry, accessLogBufferSize),
 	}
 	go e.runAccessLogWriter()
@@ -154,10 +156,26 @@ func (e *Engine) Fetch(ctx context.Context, remote models.Remote, path string, p
 		fwdHeaders = clientHeaders[0]
 	}
 
+	// Short-circuit upstream calls when the remote's breaker is open: serve
+	// stale from the store if we have it, otherwise fail fast rather than
+	// hammering a known-bad upstream.
+	if e.circuit.IsOpen(ctx, remote.Name) {
+		if stale, serr := e.serveFromStore(ctx, remote, path); serr == nil {
+			slog.Warn("circuit open, serving stale", "remote", remote.Name, "path", path)
+			stale.Source = "cache"
+			e.logAccess(remote.Name, path, true, stale.Size, 0)
+			return stale, nil
+		}
+		return nil, &ProxyError{Status: http.StatusServiceUnavailable, Message: "upstream circuit open"}
+	}
+
 	start := time.Now()
 	result, err := e.fetchFromUpstream(ctx, remote, path, prov, class, ttl, fwdHeaders)
 	upstreamMS := int(time.Since(start).Milliseconds())
 	if err != nil {
+		if isNetworkError(err) {
+			e.circuit.RecordFailure(ctx, remote.Name)
+		}
 		if remote.StaleOnError && isNetworkError(err) {
 			_ = e.cache.SetTTL(ctx, remote.Name, path, ttl)
 			stale, serr := e.serveFromStore(ctx, remote, path)
@@ -171,6 +189,7 @@ func (e *Engine) Fetch(ctx context.Context, remote models.Remote, path string, p
 		return nil, err
 	}
 
+	e.circuit.RecordSuccess(ctx, remote.Name)
 	e.logAccess(remote.Name, path, false, result.Size, upstreamMS)
 	return result, nil
 }