feat: Implement Phase 1 of kat-agent with leader election and init

internal/leader/election.go

@@ -0,0 +1,85 @@
+package leader
+
+import (
+	"context"
+	"log"
+	"time"
+
+	"git.dws.rip/dubey/kat/internal/store"
+)
+
+const (
+	// DefaultLeaseTTLSeconds is the default time-to-live for a leader's lease.
+	DefaultLeaseTTLSeconds = 15
+	// DefaultRetryPeriod is the time to wait before retrying to campaign for leadership.
+	DefaultRetryPeriod = 5 * time.Second
+)
+
+// LeadershipManager handles the lifecycle of campaigning for and maintaining leadership.
+type LeadershipManager struct {
+	Store           store.StateStore
+	LeaderID        string // Identifier for this candidate (e.g., node name)
+	LeaseTTLSeconds int64
+
+	OnElected  func(leadershipCtx context.Context) // Called when leadership is acquired
+	OnResigned func()                             // Called when leadership is lost or resigned
+}
+
+// NewLeadershipManager creates a new leadership manager.
+func NewLeadershipManager(st store.StateStore, leaderID string, onElected func(leadershipCtx context.Context), onResigned func()) *LeadershipManager {
+	return &LeadershipManager{
+		Store:           st,
+		LeaderID:        leaderID,
+		LeaseTTLSeconds: DefaultLeaseTTLSeconds,
+		OnElected:       onElected,
+		OnResigned:      onResigned,
+	}
+}
+
+// Run starts the leadership campaign loop.
+// It blocks until the provided context is cancelled.
+func (lm *LeadershipManager) Run(ctx context.Context) {
+	log.Printf("Starting leadership manager for %s", lm.LeaderID)
+	defer log.Printf("Leadership manager for %s stopped", lm.LeaderID)
+
+	for {
+		select {
+		case <-ctx.Done():
+			log.Printf("Parent context cancelled, stopping leadership campaign for %s.", lm.LeaderID)
+			// Attempt to resign if currently leading, though store.Close() might handle this too.
+			// This resign is best-effort as the app is shutting down.
+			resignCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+			lm.Store.Resign(resignCtx)
+			cancel()
+			return
+		default:
+		}
+
+		log.Printf("%s is campaigning for leadership...", lm.LeaderID)
+		leadershipCtx, err := lm.Store.Campaign(ctx, lm.LeaderID, lm.LeaseTTLSeconds)
+		if err != nil {
+			log.Printf("Error campaigning for leadership for %s: %v. Retrying in %v.", lm.LeaderID, err, DefaultRetryPeriod)
+			select {
+			case <-time.After(DefaultRetryPeriod):
+				continue
+			case <-ctx.Done():
+				return // Exit if parent context cancelled during retry wait
+			}
+		}
+
+		// Successfully became leader
+		log.Printf("%s is now the leader.", lm.LeaderID)
+		if lm.OnElected != nil {
+			lm.OnElected(leadershipCtx) // Pass the context that's cancelled on leadership loss
+		}
+
+		// Block until leadership is lost (leadershipCtx is cancelled)
+		<-leadershipCtx.Done()
+		log.Printf("%s has lost leadership.", lm.LeaderID)
+		if lm.OnResigned != nil {
+			lm.OnResigned()
+		}
+		// Loop will restart campaign unless parent ctx is done.
+		// Store.Resign() is implicitly called by the store when leadershipCtx is done or session expires.
+	}
+}