Skip to content

Stage 19: F5 XC Origins, Load Balancing & Security

Key takeaways: In XC the origin pool owns "where backends are and how to reach them," the health check owns "who is allowed to serve," the HTTP load balancer owns "which clients reach which pools and under what policy," and the app firewall is the WAF attached to the LB. A request survives six decisions between DNS and the private origin.

Origin pool — the routing contract

An origin pool answers three questions:

Question Field (Terraform shape) Lab value
Who are the backends? origin_servers → private_ip / k8s_service / public_ip / discovered endpoint 10.200.0.10
Reached how? inside_network / outside_network + site_locator inside, site gcp-xc-lab
On what port & policy? port, no_tls/use_tls, endpoint_selection, loadbalancer_algorithm 80, LOCAL_PREFERRED, LB_OVERRIDE

The inside_network flag is the security-relevant line — it means "deliver through the CE's SLI interface." Pointing at outside_network sends the REs' traffic down the public side, which is usually both wrong and unreachable.

endpoint_selection controls whether only the local site's endpoints serve (LOCAL_ONLY), local is preferred with remote failover (LOCAL_PREFERRED), or any site may serve (DISTRIBUTED).

Health checks

Independent objects referenced by pools. An HTTP check (GET /) verifies the application; a TCP check verifies the listener. Pools only send traffic to healthy endpoints — a dead origin produces 503/no-healthy-upstream responses at the RE before any client bytes reach your site. Match the check to what "healthy" means: a TCP check passes on a crashed-but-listening app; HTTP 200 on / is stronger.

HTTP load balancer — anatomy

Concern Field(s) Notes
Hostname domains[] Must be DNS-answerable: delegated to XC or CNAME'd to the LB's VIP hostname
Exposure advertise_on_public_default_vip / custom VIP / do_not_advertise Public VIP = reachable at every RE; do_not_advertise for internal testing
Routing default_route_pools (or routes for path/host rules) The pool(s) requests fall through to
Transport http / https / https_auto_cert Auto-cert needs XC-managed DNS to complete ACME
Stickiness/algo round_robin / least_active / source_ip_stickiness … Per-LB choice
Policy app_firewall ref, service_policies, rate_limit, *_challenge The security stack attached at the edge

A newer provider quirk worth knowing: many "feature off" choices (disable_api_discovery, no_challenge, disable_rate_limit, …) are required one-of groups — you must explicitly pick off/on rather than leaving them unset. That is why Lab 04's LB block lists so many disable_* lines.

The security stack — three enforcement layers

Layer Object Evaluates Typical lab use
Edge WAF volterra_app_firewall Signatures, threat campaigns, bot classes, attack types — returns a block page with support ID blocking mode with default detection
Service policy volterra_service_policy L7 allow/deny rules (path, method, headers, geo, ASN, IP reputation) Deny /admin, allow-list partners
Site network policy active_network_policies / enhanced firewall on the site L3/L4 rules inside the site fabric Restrict what the CE may reach SLI-side

WAF modes: monitoring (log only), blocking (log + block). Start a new app in blocking only if you accept false-positive tuning work; monitoring first is the safer rollout for real apps.

Numbered request path from client through DNS, RE, WAF, tunnel, and SLI to a private GCP origin

Where each failure shows up

Symptom Stage First check
Domain won't resolve 1 domains[] value, delegated DNS/CNAME
TLS/cert errors 2 https_auto_cert vs delegated DNS, or BYO cert
403 + support ID 3 Security event log; WAF mode/signatures
503 no healthy upstream 4/5 Site SLO/SLI UP? health check passing?
Timeouts to origin 5 GCP firewall rule, inside/outside flag on pool
Right page, wrong client IP in logs — enable_trust_client_ip_headers + app reads X-Forwarded-For

Same four questions, two new checkpoints

The learning-path loop — Resolution · Routing · Policy · Application — still applies. XC adds two policy checkpoints: the edge WAF (stage 3) and the tunnel/site health (stage 4). Symptoms name the layer: 403-support-ID is policy, 503-no-upstream is reachability.

Map it to the labs

  • Lab 02 clicks through pool → LB → WAF in the Console.
  • Lab 04 encodes all of it: volterra_healthcheck, volterra_origin_pool, volterra_app_firewall, volterra_http_loadbalancer.
  • Extend with volterra_discovery to let GCP labels/GKE services populate endpoints dynamically, and volterra_service_policy for allow/deny rules beyond the WAF.