fix(workbench): bound session summary runtime query (#1952)

deploy/deploy.schema.json

@@ -392,6 +392,7 @@
       "type": "object",
       "properties": {
         "client": { "$ref": "#/$defs/workbenchRuntimeClient" },
+        "sessionsSummary": { "$ref": "#/$defs/workbenchRuntimeSessionsSummary" },
         "factsQuery": { "$ref": "#/$defs/workbenchRuntimeFactsQuery" },
         "postgres": { "$ref": "#/$defs/workbenchRuntimePostgres" },
         "cache": { "$ref": "#/$defs/workbenchRuntimeCache" }
@@ -414,6 +415,15 @@
       },
       "additionalProperties": false
     },
+    "workbenchRuntimeSessionsSummary": {
+      "type": "object",
+      "properties": {
+        "maxAttempts": { "type": "integer", "minimum": 1 },
+        "attemptTimeoutMs": { "type": "integer", "minimum": 1 },
+        "backoffMs": { "type": "integer", "minimum": 0 }
+      },
+      "additionalProperties": false
+    },
     "workbenchRuntimePostgres": {
       "type": "object",
       "properties": {

deploy/deploy.yaml

@@ -309,12 +309,16 @@ lanes:
     workbenchRuntime:
       client:
         timeoutMs: 3000
+      sessionsSummary:
+        maxAttempts: 1
+        attemptTimeoutMs: 2500
+        backoffMs: 0
       factsQuery:
         maxAttempts: 1
         backoffBaseMs: 0
         backoffMaxMs: 1
       postgres:
-        poolMax: 2
+        poolMax: 4
         queryTimeoutMs: 3000
         readyTimeoutMs: 2000
       cache:

internal/workbenchruntime/service.go

@@ -71,12 +71,15 @@ type sessionsSummaryRevision struct {
 }
 
 type Config struct {
-	Addr           string
+	Addr                         string
-	DatabaseURL    string
+	DatabaseURL                  string
-	QueryTimeout   time.Duration
+	QueryTimeout                 time.Duration
-	ReadinessDelay time.Duration
+	ReadinessDelay               time.Duration
-	PoolMax        int
+	PoolMax                      int
-	Cache          cacheConfig
+	SessionSummaryMaxAttempts    int
+	SessionSummaryAttemptTimeout time.Duration
+	SessionSummaryRetryBackoff   time.Duration
+	Cache                        cacheConfig
 }
 
 type Server struct {
@@ -1442,16 +1445,17 @@ func (s *Server) dbQueryAttrs(table string, argCount int, extra map[string]any)
 
 func (s *Server) querySessionSummaries(ctx context.Context, stage string, sqlText string, args []any) ([]any, error) {
 	var lastErr error
-	for attempt := 1; attempt <= sessionSummaryQueryMaxAttempts; attempt++ {
+	maxAttempts := positiveInt(s.config.SessionSummaryMaxAttempts, sessionSummaryQueryMaxAttempts)
-		attemptResult, err := s.querySessionSummariesAttempt(ctx, stage, sqlText, args, attempt, sessionSummaryQueryMaxAttempts)
+	for attempt := 1; attempt <= maxAttempts; attempt++ {
+		attemptResult, err := s.querySessionSummariesAttempt(ctx, stage, sqlText, args, attempt, maxAttempts)
 		if err == nil {
 			return attemptResult, nil
 		}
 		lastErr = err
-		if attempt == sessionSummaryQueryMaxAttempts || !retryableSessionSummaryQueryError(err) || ctx.Err() != nil {
+		if attempt == maxAttempts || !retryableSessionSummaryQueryError(err) || ctx.Err() != nil {
 			break
 		}
-		backoff := sessionSummaryRetryBackoff(attempt)
+		backoff := s.sessionSummaryRetryBackoff(attempt)
 		timer := time.NewTimer(backoff)
 		select {
 		case <-ctx.Done():
@@ -1465,7 +1469,7 @@ func (s *Server) querySessionSummaries(ctx context.Context, stage string, sqlTex
 
 func (s *Server) querySessionSummariesAttempt(ctx context.Context, stage string, sqlText string, args []any, attempt int, maxAttempts int) ([]any, error) {
 	result := []any{}
-	attemptCtx, cancel := context.WithTimeout(ctx, sessionSummaryQueryAttemptTimeout)
+	attemptCtx, cancel := context.WithTimeout(ctx, positiveDuration(s.config.SessionSummaryAttemptTimeout, sessionSummaryQueryAttemptTimeout))
 	defer cancel()
 	attrs := s.dbQueryAttrs("workbench_sessions", len(args), map[string]any{
 		"db.index.expected": "idx_workbench_sessions_owner_updated|idx_workbench_sessions_updated",
@@ -1473,7 +1477,7 @@ func (s *Server) querySessionSummariesAttempt(ctx context.Context, stage string,
 		"retryMax":          int64(maxAttempts),
 	})
 	if attempt > 1 {
-		attrs["retryBackoffMs"] = sessionSummaryRetryBackoff(attempt - 1).Milliseconds()
+		attrs["retryBackoffMs"] = s.sessionSummaryRetryBackoff(attempt - 1).Milliseconds()
 	}
 	err := s.withOtelInternalSpan(attemptCtx, stage, attrs, func(spanCtx context.Context) error {
 		rows, err := s.db.QueryContext(spanCtx, sqlText, args...)
@@ -1513,11 +1517,26 @@ func (s *Server) querySessionSummariesAttempt(ctx context.Context, stage string,
 	return result, err
 }
 
-func sessionSummaryRetryBackoff(attempt int) time.Duration {
+func (s *Server) sessionSummaryRetryBackoff(attempt int) time.Duration {
+	base := positiveDuration(s.config.SessionSummaryRetryBackoff, sessionSummaryQueryRetryBackoff)
 	if attempt <= 1 {
-		return sessionSummaryQueryRetryBackoff
+		return base
 	}
-	return time.Duration(attempt) * sessionSummaryQueryRetryBackoff
+	return time.Duration(attempt) * base
+}
+
+func positiveDuration(value time.Duration, fallback time.Duration) time.Duration {
+	if value > 0 {
+		return value
+	}
+	return fallback
+}
+
+func positiveInt(value int, fallback int) int {
+	if value > 0 {
+		return value
+	}
+	return fallback
 }
 
 func retryableSessionSummaryQueryError(err error) bool {
@@ -2310,12 +2329,15 @@ func normalizeProjectionHealth(value string, projectionStatus string) string {
 func configFromEnv() Config {
 	port := first(os.Getenv("HWLAB_WORKBENCH_RUNTIME_PORT"), os.Getenv("PORT"), "6671")
 	return Config{
-		Addr:           ":" + port,
+		Addr:                         ":" + port,
-		DatabaseURL:    first(os.Getenv("HWLAB_WORKBENCH_RUNTIME_DB_URL"), os.Getenv("HWLAB_CLOUD_DB_URL")),
+		DatabaseURL:                  first(os.Getenv("HWLAB_WORKBENCH_RUNTIME_DB_URL"), os.Getenv("HWLAB_CLOUD_DB_URL")),
-		QueryTimeout:   time.Duration(envInt("HWLAB_WORKBENCH_RUNTIME_QUERY_TIMEOUT_MS", 10000)) * time.Millisecond,
+		QueryTimeout:                 time.Duration(envInt("HWLAB_WORKBENCH_RUNTIME_QUERY_TIMEOUT_MS", 10000)) * time.Millisecond,
-		ReadinessDelay: time.Duration(envInt("HWLAB_WORKBENCH_RUNTIME_READY_TIMEOUT_MS", 2000)) * time.Millisecond,
+		ReadinessDelay:               time.Duration(envInt("HWLAB_WORKBENCH_RUNTIME_READY_TIMEOUT_MS", 2000)) * time.Millisecond,
-		PoolMax:        envInt("HWLAB_WORKBENCH_RUNTIME_DB_POOL_MAX", 8),
+		PoolMax:                      envInt("HWLAB_WORKBENCH_RUNTIME_DB_POOL_MAX", 8),
-		Cache:          cacheConfigFromEnv(),
+		SessionSummaryMaxAttempts:    envInt("HWLAB_WORKBENCH_SESSIONS_SUMMARY_QUERY_MAX_ATTEMPTS", sessionSummaryQueryMaxAttempts),
+		SessionSummaryAttemptTimeout: time.Duration(envInt("HWLAB_WORKBENCH_SESSIONS_SUMMARY_QUERY_ATTEMPT_TIMEOUT_MS", int(sessionSummaryQueryAttemptTimeout/time.Millisecond))) * time.Millisecond,
+		SessionSummaryRetryBackoff:   time.Duration(envInt("HWLAB_WORKBENCH_SESSIONS_SUMMARY_QUERY_BACKOFF_MS", int(sessionSummaryQueryRetryBackoff/time.Millisecond))) * time.Millisecond,
+		Cache:                        cacheConfigFromEnv(),
 	}
 }
 

scripts/gitops-render.mjs

@@ -941,6 +941,7 @@ function deployServicesForProfile(deploy, profile) {
         runtimeStoreEnvForProfile(deploy, profile, service.serviceId),
         workbenchRuntimeClientEnvForProfile(deploy, profile, service.serviceId),
         workbenchRuntimeFactsQueryEnvForProfile(deploy, profile, service.serviceId),
+        workbenchRuntimeSessionsSummaryEnvForProfile(deploy, profile, service.serviceId),
         workbenchRuntimePostgresEnvForProfile(deploy, profile, service.serviceId),
         workbenchRuntimeRedisEnvForProfile(deploy, profile, service.serviceId)
       );
@@ -957,6 +958,7 @@ function deployServicesForProfile(deploy, profile) {
         runtimeStoreEnvForProfile(deploy, profile, override.serviceId),
         workbenchRuntimeClientEnvForProfile(deploy, profile, override.serviceId),
         workbenchRuntimeFactsQueryEnvForProfile(deploy, profile, override.serviceId),
+        workbenchRuntimeSessionsSummaryEnvForProfile(deploy, profile, override.serviceId),
         workbenchRuntimePostgresEnvForProfile(deploy, profile, override.serviceId),
         workbenchRuntimeRedisEnvForProfile(deploy, profile, override.serviceId)
       )
@@ -1075,6 +1077,17 @@ function workbenchRuntimeFactsQueryEnvForProfile(deploy, profile, serviceId) {
   return env;
 }
 
+function workbenchRuntimeSessionsSummaryEnvForProfile(deploy, profile, serviceId) {
+  if (!isRuntimeLane(profile) || serviceId !== "hwlab-workbench-runtime") return {};
+  const sessionsSummary = runtimeLaneConfig(deploy, profile)?.workbenchRuntime?.sessionsSummary;
+  if (!sessionsSummary || typeof sessionsSummary !== "object" || Array.isArray(sessionsSummary)) return {};
+  const env = {};
+  if (Number.isInteger(sessionsSummary.maxAttempts)) env.HWLAB_WORKBENCH_SESSIONS_SUMMARY_QUERY_MAX_ATTEMPTS = String(sessionsSummary.maxAttempts);
+  if (Number.isInteger(sessionsSummary.attemptTimeoutMs)) env.HWLAB_WORKBENCH_SESSIONS_SUMMARY_QUERY_ATTEMPT_TIMEOUT_MS = String(sessionsSummary.attemptTimeoutMs);
+  if (Number.isInteger(sessionsSummary.backoffMs)) env.HWLAB_WORKBENCH_SESSIONS_SUMMARY_QUERY_BACKOFF_MS = String(sessionsSummary.backoffMs);
+  return env;
+}
+
 function workbenchRuntimePostgresEnvForProfile(deploy, profile, serviceId) {
   if (!isRuntimeLane(profile) || serviceId !== "hwlab-workbench-runtime") return {};
   const postgres = runtimeLaneConfig(deploy, profile)?.workbenchRuntime?.postgres;