3.3 Serverless NEGs, Sessions, and Edge Controls

Serverless Network Endpoint Groups (Serverless NEGs)

A Network Endpoint Group (NEG) specifies a group of backend endpoints. While Zonal NEGs contain VM IP/port pairs or GKE Pod IPs, Serverless NEGs point directly to serverless compute engines:

• Cloud Run services
• App Engine apps (standard and flexible)
• Cloud Functions (HTTP-triggered 2nd gen / Cloud Run functions)

Multi-Region Cloud Run Active-Active Pattern

By attaching multiple regional Serverless NEGs to a single global Backend Service, the Global External ALB directs each client to the nearest Cloud Run region. Serverless NEGs do not support ordinary backend health checks or balancing-mode capacity settings. Where supported, explicitly configure outlier detection to reduce traffic to failing regions; it is not a guaranteed instantaneous regional failover or overflow mechanism. Test dependency failures as well as service failures.

## Create a Serverless NEG pointing to Cloud Run in us-central1
gcloud compute network-endpoint-groups create neg-cr-us \
    --region=us-central1 \
    --network-endpoint-type=serverless \
    --cloud-run-service=order-service

## Create a Serverless NEG pointing to Cloud Run in europe-west1
gcloud compute network-endpoint-groups create neg-cr-eu \
    --region=europe-west1 \
    --network-endpoint-type=serverless \
    --cloud-run-service=order-service

## Create a global Backend Service
gcloud compute backend-services create orders-backend-service \
    --global \
    --load-balancing-scheme=EXTERNAL_MANAGED \
    --protocol=HTTPS

## Attach both regional Serverless NEGs to the global Backend Service
gcloud compute backend-services add-backend orders-backend-service \
    --global \
    --network-endpoint-group=neg-cr-us \
    --network-endpoint-group-region=us-central1

gcloud compute backend-services add-backend orders-backend-service \
    --global \
    --network-endpoint-group=neg-cr-eu \
    --network-endpoint-group-region=europe-west1

---

Session Affinity Options

For supported VM/GKE backends, backend-service session affinity can improve locality. Serverless NEG backend services do not support load-balancer session affinity; configure Cloud Run's own best-effort affinity when needed. Store authoritative session state externally so failover and scaling do not lose it.

+--------------------------------------------------------------------------+
| Session Affinity Decision Matrix                                         |
+--------------------------------------------------------------------------+
|                                                                          |
| Are clients connecting through corporate NAT proxies or shared gateways? |
|        |                                                                 |
|        +--- YES ---> Use GENERATED_COOKIE                                |
|        |             (Sets 'GCLB' cookie; prevents IP hotspot skewing)   |
|        |                                                                 |
|        +--- NO ----> Do clients send a unique tracking header?           |
|                      |                                                   |
|                      +--- YES ---> Use HEADER_FIELD                      |
|                      |             (Hashes header, e.g., 'X-User-ID')    |
|                      |                                                   |
|                      +--- NO ----> Use CLIENT_IP                         |
|                                    (2-tuple hash of src/dest IP)         |
+--------------------------------------------------------------------------+

• NONE (Default): Requests are distributed across backends using round-robin or least-loaded algorithms. No stickiness.
• CLIENT_IP: Hashes the client's source IP address. Critical Pitfall: If thousands of users access the application from behind a single corporate enterprise NAT gateway or university proxy, all users share the same source IP. The load balancer hashes them to the exact same backend instance, creating an extreme performance bottleneck (hotspot) while other instances sit idle.
• GENERATED_COOKIE: The load balancer issues an HTTP session cookie (named GCLB) on the first response. Subsequent requests containing this cookie are routed to the same backend instance. If the instance fails or scale-down occurs, the cookie is regenerated for a healthy instance. Cookie-based affinity avoids grouping all users solely by a shared NAT address, but distribution remains best effort.
• HEADER_FIELD: Hashes the value of an arbitrary HTTP request header specified by the developer (e.g., X-Organization-Token or Authorization).

---

Edge Security Integration: Google Cloud Armor

Cloud Armor attaches directly to the Backend Services of External Application Load Balancers, inspecting traffic at the edge before packets can reach backend compute instances or Serverless NEGs.

• Layer 7 WAF & OWASP Top 10: Cloud Armor contains preconfigured rules for SQL injection (sqli-v33-stable), Cross-Site Scripting (xss-v33-stable), Local File Inclusion (lfi-v33-stable), and Remote Code Execution (rce-v33-stable).
• Adaptive Rate Limiting: Enforces maximum request thresholds per client IP or custom header over configurable time windows, with an automatic ban period (e.g., blocking an IP for 10 minutes if it exceeds 100 requests in 1 minute).
• Geo-Fencing: Evaluates client origin IP geolocations using ISO 3166-1 alpha-2 country codes:

  origin.region_code == 'CN' || origin.region_code == 'RU'
  

• Preview Mode: Allows developers to test new WAF rules against live production traffic without dropping packets; matched requests emit an audit log with the evaluated disposition.

---

Edge Caching: Cloud CDN

Cloud CDN leverages Google's globally distributed edge caching infrastructure to terminate client requests at the PoP, dramatically reducing origin server load and delivery latency.

• Enabling CDN: Can be enabled on supported backend services and backend buckets; verify compatibility with the selected backend and other features.
• Cache Modes:
  • CACHE_ALL_STATIC: Automatically caches static web assets (images, CSS, JS, fonts) using eligible content types and cacheability rules; it still respects origin directives such as private, no-store, and no-cache.
  • USE_ORIGIN_HEADERS: Strictly respects origin Cache-Control response headers (public, max-age, s-maxage, no-store).
  • FORCE_CACHE_ALL: Caches all responses regardless of origin headers (used with extreme caution).
• Cache Invalidation: If an urgent deployment modifies cached assets, invalidations are requested globally; allow for propagation via URL maps:

  gcloud compute url-maps invalidate-cdn-cache web-url-map \
      --path="/static/v2/*"
  

• Signed URLs & Signed Cookies: Protects private media streams and restricted downloads by verifying HMAC cryptographic signatures before serving cached assets.

Official references checked October 18, 2026: Google Cloud 1.