fix: reduce connection timeouts for Kafka/Redis/Temporal to prevent CI test timeout

- Kafka: add 3s context timeout for DialContext (was using parent context with no timeout)
- Redis: set MaxRetries=1 (was default 5), reduce MinIdleConns to 2, reduce DialTimeout to 2s
- Temporal: use NewLazyClient + 3s CheckHealth instead of blocking Dial; close client on health check failure

Tests now complete in ~36s locally (was timing out at 60s in CI)

Co-Authored-By: Patrick Munis <pmunis@gmail.com>

Author: devin-ai-integration[bot]

services/gateway/internal/kafka/client.go

@@ -81,7 +81,9 @@ func (c *Client) connect() {
 		Async:        false,
 	}
 
-	conn, err := kafka.DialContext(c.ctx, "tcp", c.brokers)
+	dialCtx, dialCancel := context.WithTimeout(c.ctx, 3*time.Second)
+	defer dialCancel()
+	conn, err := kafka.DialContext(dialCtx, "tcp", c.brokers)
 	if err != nil {
 		log.Printf("[Kafka] WARN: Cannot reach %s: %v — running in fallback mode (in-memory dispatch)", c.brokers, err)
 		c.mu.Lock()

services/gateway/internal/redis/client.go

@@ -72,11 +72,12 @@ func (c *Client) connect() {
 		Addr:         c.url,
 		Password:     c.password,
 		DB:           0,
-		DialTimeout:  3 * time.Second,
+		DialTimeout:  2 * time.Second,
 		ReadTimeout:  2 * time.Second,
 		WriteTimeout: 2 * time.Second,
 		PoolSize:     20,
-		MinIdleConns: 5,
+		MinIdleConns: 2,
+		MaxRetries:   1, // Fail fast — don't retry 5 times on initial connect
 	}
 	rdb := goredis.NewClient(opts)
 

services/gateway/internal/temporal/client.go

@@ -57,13 +57,22 @@ func NewClient(host string) *Client {
 func (c *Client) connect() {
 	log.Printf("[Temporal] Connecting to %s", c.host)
 
-	// Create real Temporal SDK client
-	sdkClient, err := client.Dial(client.Options{
+	// Create real Temporal SDK client with short connection timeout
+	dialCtx, dialCancel := context.WithTimeout(c.ctx, 3*time.Second)
+	defer dialCancel()
+	sdkClient, err := client.NewLazyClient(client.Options{
 		HostPort:  c.host,
 		Namespace: "nexcom",
 	})
+	if err == nil {
+		// Verify connectivity with a quick health check
+		_, err = sdkClient.CheckHealth(dialCtx, &client.CheckHealthRequest{})
+	}
 	if err != nil {
 		log.Printf("[Temporal] WARN: Cannot reach %s: %v — running in fallback mode (in-memory workflows)", c.host, err)
+		if sdkClient != nil {
+			sdkClient.Close()
+		}
 		c.mu.Lock()
 		c.fallbackMode = true
 		c.connected = false