fix(ha): PR #101 HA implementation bug fixes

.github/workflows/benchmarks.yml

@@ -27,7 +27,22 @@ jobs:
             | grep -E '^(goos:|goarch:|pkg:|cpu:|Benchmark|PASS$|ok\s)' \
             | tee bench.txt
 
-      - name: Store Benchmark Result
+      - name: Store Benchmark Result (PR)
+        if: github.event_name == 'pull_request'
+        uses: benchmark-action/github-action-benchmark@v1
+        with:
+          name: Go Benchmarks
+          tool: 'go'
+          output-file-path: bench.txt
+          # On PRs, publishing to gh-pages is not allowed in all permission models.
+          auto-push: false
+          # Fail if performance drops by more than 50%
+          alert-threshold: '200%'
+          comment-on-alert: false
+          fail-on-alert: false
+
+      - name: Store Benchmark Result (main)
+        if: github.event_name == 'push' && github.ref == 'refs/heads/main'
         uses: benchmark-action/github-action-benchmark@v1
         with:
           name: Go Benchmarks

cmd/api/main.go

@@ -130,16 +130,28 @@ func run() error {
 	defer db.Close()
 	defer func() { _ = rdb.Close() }()
 
-	compute, storage, network, lbProxy, err := initBackends(deps, cfg, logger, db, rdb)
+	rawCompute, rawStorage, rawNetwork, rawLBProxy, err := initBackends(deps, cfg, logger, db, rdb)
 	if err != nil {
 		logger.Error("backend initialization failed", "error", err)
 		return err
 	}
 
+	// Wrap raw backends with resilience decorators (circuit breaker, bulkhead, timeouts).
+	compute := platform.NewResilientCompute(rawCompute, logger, platform.ResilientComputeOpts{})
+	storage := platform.NewResilientStorage(rawStorage, logger, platform.ResilientStorageOpts{})
+	network := platform.NewResilientNetwork(rawNetwork, logger, platform.ResilientNetworkOpts{})
+	lbProxy := platform.NewResilientLB(rawLBProxy, logger, platform.ResilientLBOpts{})
+
 	repos := deps.InitRepositories(db, rdb)
+
+	// Create leader elector for singleton worker coordination.
+	// When multiple worker replicas run, only one will hold leadership per key.
+	leaderElector := postgres.NewPgLeaderElector(db, logger)
+
 	svcs, workers, err := deps.InitServices(setup.ServiceConfig{
 		Config: cfg, Repos: repos, Compute: compute, Storage: storage,
 		Network: network, LBProxy: lbProxy, DB: db, RDB: rdb, Logger: logger,
+		LeaderElector: leaderElector,
 	})
 	if err != nil {
 		logger.Error("service initialization failed", "error", err)
@@ -154,52 +166,61 @@ func run() error {
 		r.Use(otelgin.Middleware("compute-api"))
 	}
 
-	runApplication(deps, cfg, logger, r, workers)
-	return nil
+	return runApplication(deps, cfg, logger, r, workers)
 }
 
-func runApplication(deps AppDeps, cfg *platform.Config, logger *slog.Logger, r *gin.Engine, workers *setup.Workers) {
-	role := os.Getenv("APP_ROLE")
+func runApplication(deps AppDeps, cfg *platform.Config, logger *slog.Logger, r *gin.Engine, workers *setup.Workers) error {
+	role := os.Getenv("ROLE")
 	if role == "" {
 		role = "all"
 	}
 
+	validRoles := map[string]bool{"api": true, "worker": true, "all": true}
+	if !validRoles[role] {
+		logger.Error("invalid ROLE value, must be one of: api, worker, all", "role", role)
+		return fmt.Errorf("invalid ROLE value %q, must be one of: api, worker, all", role)
+	}
+	logger.Info("starting with role", "role", role)
+
 	wg := &sync.WaitGroup{}
 	workerCtx, workerCancel := context.WithCancel(context.Background())
 
 	if role == "worker" || role == "all" {
 		runWorkers(workerCtx, wg, workers)
 	}
 
-	srv := deps.NewHTTPServer(":"+cfg.Port, r)
-
+	var srv *http.Server
 	if role == "api" || role == "all" {
+		srv = deps.NewHTTPServer(":"+cfg.Port, r)
 		go func() {
 			logger.Info("starting compute-api", "port", cfg.Port)
 			if err := deps.StartHTTPServer(srv); err != nil && !stdlib_errors.Is(err, http.ErrServerClosed) {
 				logger.Error("failed to start server", "error", err)
 			}
 		}()
 	} else {
-		logger.Info("running in worker-only mode")
+		logger.Info("running in worker-only mode, HTTP server disabled")
 	}
 
 	quit := make(chan os.Signal, 1)
 	deps.NotifySignals(quit, syscall.SIGINT, syscall.SIGTERM)
 	<-quit
 
-	logger.Info("shutting down server...")
+	logger.Info("shutting down...")
 
 	ctx, cancel := context.WithTimeout(context.Background(), defaultShutdownTimeout)
 	defer cancel()
 
-	if err := deps.ShutdownHTTPServer(ctx, srv); err != nil {
-		logger.Error("server forced to shutdown", "error", err)
+	if srv != nil {
+		if err := deps.ShutdownHTTPServer(ctx, srv); err != nil {
+			logger.Error("server forced to shutdown", "error", err)
+		}
 	}
 
 	workerCancel()
 	wg.Wait()
-	logger.Info("server exited")
+	logger.Info("shutdown complete")
+	return nil
 }
 
 type runner interface {

cmd/api/main_test.go

@@ -158,7 +158,10 @@ func TestRunApplicationApiRoleStartsAndShutsDown(t *testing.T) {
 		}()
 	}
 
-	runApplication(deps, &platform.Config{Port: "0"}, logger, gin.New(), &setup.Workers{})
+	err := runApplication(deps, &platform.Config{Port: "0"}, logger, gin.New(), &setup.Workers{})
+	if err != nil {
+		t.Fatalf("runApplication returned error: %v", err)
+	}
 
 	select {
 	case <-shutdownCalled:
@@ -167,6 +170,106 @@ func TestRunApplicationApiRoleStartsAndShutsDown(t *testing.T) {
 	}
 }
 
+func TestRunApplicationWorkerRoleDoesNotStartHTTP(t *testing.T) {
+	logger := slog.New(slog.NewTextHandler(io.Discard, nil))
+	t.Setenv("ROLE", "worker")
+
+	deps := DefaultDeps()
+
+	deps.NewHTTPServer = func(string, http.Handler) *http.Server {
+		t.Fatalf("NewHTTPServer should not be called in worker-only mode")
+		return nil
+	}
+	deps.StartHTTPServer = func(*http.Server) error {
+		t.Fatalf("StartHTTPServer should not be called in worker-only mode")
+		return nil
+	}
+	deps.ShutdownHTTPServer = func(context.Context, *http.Server) error {
+		t.Fatalf("ShutdownHTTPServer should not be called in worker-only mode")
+		return nil
+	}
+	deps.NotifySignals = func(c chan<- os.Signal, _ ...os.Signal) {
+		go func() {
+			// Give workers a moment to start, then signal shutdown
+			time.Sleep(50 * time.Millisecond)
+			c <- syscall.SIGTERM
+		}()
+	}
+
+	err := runApplication(deps, &platform.Config{Port: "0"}, logger, gin.New(), &setup.Workers{})
+	if err != nil {
+		t.Fatalf("runApplication returned error: %v", err)
+	}
+	// If we reach here without t.Fatalf, the test passes — no HTTP server was touched.
+}
+
+func TestRunApplicationDefaultsToAllRole(t *testing.T) {
+	logger := slog.New(slog.NewTextHandler(io.Discard, nil))
+	t.Setenv("ROLE", "") // Explicitly empty to verify default
+
+	started := make(chan struct{})
+	shutdownCalled := make(chan struct{})
+	deps := DefaultDeps()
+
+	deps.NewHTTPServer = func(addr string, handler http.Handler) *http.Server {
+		return &http.Server{
+			Addr:              addr,
+			Handler:           handler,
+			ReadHeaderTimeout: 10 * time.Second,
+		}
+	}
+	deps.StartHTTPServer = func(*http.Server) error {
+		close(started)
+		return http.ErrServerClosed
+	}
+	deps.ShutdownHTTPServer = func(context.Context, *http.Server) error {
+		close(shutdownCalled)
+		return nil
+	}
+	deps.NotifySignals = func(c chan<- os.Signal, _ ...os.Signal) {
+		go func() {
+			<-started
+			c <- syscall.SIGTERM
+		}()
+	}
+
+	err := runApplication(deps, &platform.Config{Port: "0"}, logger, gin.New(), &setup.Workers{})
+	if err != nil {
+		t.Fatalf("runApplication returned error: %v", err)
+	}
+
+	select {
+	case <-shutdownCalled:
+	case <-time.After(time.Second):
+		t.Fatalf("expected server shutdown to be called when ROLE defaults to 'all'")
+	}
+}
+
+func TestRunApplicationInvalidRoleReturnsEarly(t *testing.T) {
+	logger := slog.New(slog.NewTextHandler(io.Discard, nil))
+	t.Setenv("ROLE", "invalid")
+
+	deps := DefaultDeps()
+
+	deps.NewHTTPServer = func(string, http.Handler) *http.Server {
+		t.Fatalf("NewHTTPServer should not be called for invalid role")
+		return nil
+	}
+	deps.StartHTTPServer = func(*http.Server) error {
+		t.Fatalf("StartHTTPServer should not be called for invalid role")
+		return nil
+	}
+	deps.NotifySignals = func(c chan<- os.Signal, _ ...os.Signal) {
+		t.Fatalf("NotifySignals should not be called for invalid role")
+	}
+
+	// Should return immediately without starting anything
+	err := runApplication(deps, &platform.Config{Port: "0"}, logger, gin.New(), &setup.Workers{})
+	if err == nil {
+		t.Fatalf("expected error for invalid role")
+	}
+}
+
 // Stub helpers below keep main.go testable without altering production behavior.
 
 type stubDB struct{ closed bool }

docs/swagger/docs.go

@@ -7862,6 +7862,12 @@ const docTemplate = `{
                         "name": "id",
                         "in": "path",
                         "required": true
+                    },
+                    {
+                        "type": "boolean",
+                        "description": "Force delete even with dependent resources (for async cleanup)",
+                        "name": "force",
+                        "in": "query"
                     }
                 ],
                 "responses": {

docs/swagger/swagger.json

@@ -7854,6 +7854,12 @@
                         "name": "id",
                         "in": "path",
                         "required": true
+                    },
+                    {
+                        "type": "boolean",
+                        "description": "Force delete even with dependent resources (for async cleanup)",
+                        "name": "force",
+                        "in": "query"
                     }
                 ],
                 "responses": {

docs/swagger/swagger.yaml

@@ -7420,6 +7420,10 @@ paths:
         name: id
         required: true
         type: string
+      - description: Force delete even with dependent resources (for async cleanup)
+        in: query
+        name: force
+        type: boolean
       produces:
       - application/json
       responses: