flock-agent: GC orphaned allocations; retry birdc on socket-not-ready (#5)

flock-agent: GC orphaned allocations; retry birdc on socket-not-ready

pkg/agent/bird.go

@@ -8,6 +8,7 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
+	"strings"
 	"sync"
 	"time"
 
@@ -137,6 +138,15 @@ func (b *BirdManager) scheduleReload() {
 	})
 }
 
+// birdcMaxAttempts and birdcRetryDelay bound the startup-socket retry loop.
+// BIRD may not have bound its control socket yet when flock-agent first
+// tries to configure it; 20 × 500 ms = 10 s covers the typical BIRD
+// startup window without blocking indefinitely.
+const (
+	birdcMaxAttempts = 20
+	birdcRetryDelay  = 500 * time.Millisecond
+)
+
 func (b *BirdManager) reload() {
 	birdctl := b.BirdctlPath
 	if birdctl == "" {
@@ -146,18 +156,40 @@ func (b *BirdManager) reload() {
 	if socket == "" {
 		socket = "/run/flock/bird.ctl"
 	}
-	cmd := exec.Command(birdctl, "-s", socket, "configure")
-	out, err := cmd.CombinedOutput()
-	if err != nil {
-		// First-run case: bird may not be ready yet — retry on next change.
-		if errors.Is(err, exec.ErrNotFound) || os.IsNotExist(err) {
-			b.Logger.Warn("birdc not available", "err", err)
+
+	for attempt := 1; attempt <= birdcMaxAttempts; attempt++ {
+		if attempt > 1 {
+			time.Sleep(birdcRetryDelay)
+		}
+		// Socket absent → BIRD hasn't bound it yet; retry.
+		if _, err := os.Stat(socket); os.IsNotExist(err) {
+			b.Logger.Debug("birdc socket not ready, retrying",
+				"attempt", attempt, "socket", socket)
+			continue
+		}
+		cmd := exec.Command(birdctl, "-s", socket, "configure")
+		out, err := cmd.CombinedOutput()
+		if err == nil {
+			b.Logger.Info("birdc configure ok", "out", string(out))
 			return
 		}
-		b.Logger.Warn("birdc reload failed", "err", err, "out", string(out))
+		if errors.Is(err, exec.ErrNotFound) {
+			b.Logger.Warn("birdc not found", "err", err)
+			return
+		}
+		outStr := string(out)
+		// "Unable to connect" means BIRD exists but isn't listening yet.
+		if strings.Contains(outStr, "Unable to connect") {
+			b.Logger.Debug("birdc not ready, retrying",
+				"attempt", attempt, "err", err)
+			continue
+		}
+		// Any other failure (syntax error, etc.) is not retriable.
+		b.Logger.Warn("birdc reload failed", "err", err, "out", outStr)
 		return
 	}
-	b.Logger.Info("birdc configure ok", "out", string(out))
+	b.Logger.Error("birdc configure gave up after retries",
+		"socket", socket, "attempts", birdcMaxAttempts)
 }
 
 // SummaryRoutes installs blackhole kernel routes for each NodeConfig CIDR.