fix: relax v03 external postgres startup timeouts (#2190)

deploy/deploy.yaml

@@ -353,7 +353,7 @@ lanes:
         timeoutMs: 3000
       sessionsSummary:
         maxAttempts: 1
-        attemptTimeoutMs: 2500
+        attemptTimeoutMs: 10000
         backoffMs: 0
       factsQuery:
         maxAttempts: 1
@@ -361,8 +361,8 @@ lanes:
         backoffMaxMs: 1
       postgres:
         poolMax: 16
-        queryTimeoutMs: 3000
+        queryTimeoutMs: 10000
-        readyTimeoutMs: 2000
+        readyTimeoutMs: 10000
       cache:
         redis:
           enabled: true
@@ -473,6 +473,16 @@ lanes:
         artifactKind: go-service
         healthPath: /health/ready
         healthPort: 6671
+        healthProbe:
+          readiness:
+            timeoutSeconds: 10
+            failureThreshold: 5
+          liveness:
+            timeoutSeconds: 5
+            failureThreshold: 6
+          startup:
+            timeoutSeconds: 10
+            failureThreshold: 30
         componentPaths:
           - go.mod
           - go.sum
@@ -488,6 +498,16 @@ lanes:
         artifactKind: go-service
         healthPath: /health/ready
         healthPort: 6670
+        healthProbe:
+          readiness:
+            timeoutSeconds: 10
+            failureThreshold: 5
+          liveness:
+            timeoutSeconds: 5
+            failureThreshold: 6
+          startup:
+            timeoutSeconds: 10
+            failureThreshold: 30
         componentPaths:
           - go.mod
           - go.sum
@@ -503,6 +523,16 @@ lanes:
         artifactKind: bun-command
         healthPath: /health/ready
         healthPort: 6672
+        healthProbe:
+          readiness:
+            timeoutSeconds: 10
+            failureThreshold: 5
+          liveness:
+            timeoutSeconds: 5
+            failureThreshold: 6
+          startup:
+            timeoutSeconds: 10
+            failureThreshold: 30
         componentPaths:
           - cmd/hwlab-project-management/
           - internal/project-management/
@@ -751,6 +781,8 @@ lanes:
         env:
           HWLAB_USER_BILLING_DB_URL: secretRef:hwlab-cloud-api-v03-db/database-url
           HWLAB_USER_BILLING_DB_POOL_MAX: "2"
+          HWLAB_USER_BILLING_MIGRATION_TIMEOUT_SECONDS: "180"
+          HWLAB_USER_BILLING_BOOTSTRAP_TIMEOUT_SECONDS: "60"
           HWLAB_BOOTSTRAP_ADMIN_ID: usr_v03_admin
           HWLAB_BOOTSTRAP_ADMIN_USERNAME: admin
           HWLAB_BOOTSTRAP_ADMIN_DISPLAY_NAME: HWLAB v0.3 Admin
@@ -769,6 +801,7 @@ lanes:
         replicas: 1
         env:
           HWLAB_PROJECT_MANAGEMENT_DB_URL: secretRef:hwlab-cloud-api-v03-db/database-url
+          HWLAB_PROJECT_MANAGEMENT_DB_CONNECT_TIMEOUT_MS: "5000"
           HWLAB_PROJECT_MANAGEMENT_PORT: "6672"
           HWLAB_PROJECT_MANAGEMENT_SOURCE_ROOT: /workspace/hwlab-boot/repo
           HWLAB_PROJECT_MANAGEMENT_SOURCE_ID: hwlab-v03-mdtodo

internal/userbilling/service.go

@@ -46,6 +46,8 @@ type Config struct {
 	InternalToken              string
 	RequireInternalToken       bool
 	MigrateOnStart             bool
+	MigrationTimeout           time.Duration
+	BootstrapTimeout           time.Duration
 	StateAuthority             string
 	RuntimeNamespace           string
 	ExternalDatabaseLabel      string
@@ -332,6 +334,8 @@ func LoadConfig(env []string) Config {
 		InternalToken:              values["HWLAB_USER_BILLING_INTERNAL_TOKEN"],
 		RequireInternalToken:       values["HWLAB_USER_BILLING_REQUIRE_INTERNAL_TOKEN"] == "1",
 		MigrateOnStart:             values["HWLAB_USER_BILLING_MIGRATE_ON_START"] != "0",
+		MigrationTimeout:           parseDurationSeconds(values["HWLAB_USER_BILLING_MIGRATION_TIMEOUT_SECONDS"], 20),
+		BootstrapTimeout:           parseDurationSeconds(values["HWLAB_USER_BILLING_BOOTSTRAP_TIMEOUT_SECONDS"], 10),
 		StateAuthority:             first(values["HWLAB_USER_BILLING_STATE_AUTHORITY"], "pk01-postgres"),
 		RuntimeNamespace:           first(values["POD_NAMESPACE"], values["HWLAB_NAMESPACE"], "hwlab-v03"),
 		ExternalDatabaseLabel:      first(values["HWLAB_USER_BILLING_EXTERNAL_DB_LABEL"], "pk01-external-postgres"),
@@ -357,13 +361,13 @@ func NewServer(ctx context.Context, cfg Config) (*Server, error) {
 		db.SetConnMaxLifetime(30 * time.Minute)
 		server.db = db
 		if cfg.MigrateOnStart {
-			migrateCtx, cancel := context.WithTimeout(ctx, 20*time.Second)
+			migrateCtx, cancel := context.WithTimeout(ctx, durationOrDefault(cfg.MigrationTimeout, 20*time.Second))
 			defer cancel()
 			if err := server.migrate(migrateCtx); err != nil {
 				return nil, fmt.Errorf("migrate user billing schema: %w", err)
 			}
 		}
-		bootstrapCtx, cancel := context.WithTimeout(ctx, 10*time.Second)
+		bootstrapCtx, cancel := context.WithTimeout(ctx, durationOrDefault(cfg.BootstrapTimeout, 10*time.Second))
 		defer cancel()
 		if err := server.ensureBootstrapAdmin(bootstrapCtx); err != nil {
 			return nil, fmt.Errorf("ensure bootstrap admin: %w", err)
@@ -2941,6 +2945,17 @@ func parseDurationMinutes(value string, fallbackMinutes int64) time.Duration {
 	return time.Duration(parseInt64(value, fallbackMinutes)) * time.Minute
 }
 
+func parseDurationSeconds(value string, fallbackSeconds int64) time.Duration {
+	return time.Duration(parseInt64(value, fallbackSeconds)) * time.Second
+}
+
+func durationOrDefault(value time.Duration, fallback time.Duration) time.Duration {
+	if value <= 0 {
+		return fallback
+	}
+	return value
+}
+
 func first(values ...string) string {
 	for _, value := range values {
 		if strings.TrimSpace(value) != "" {