
The global external Application Load Balancer is a proxy that runs on Google Front Ends spread across Google's edge locations. It uses a single anycast IP address, so each user connects to a nearby edge location, where the TCP and TLS handshakes finish. The request then travels to the backends over Google's private backbone. Ending connections close to the user removes several long round trips from each new connection, which is how this load balancer cuts latency for distant users even when the backends cannot move. For the Professional Cloud Network Engineer exam, the topics that come up most are how it distributes traffic across regions, how its backends should be organized, and how its URL map matches requests.
Because the load balancer advertises one anycast address from Google's edge, every user enters Google's network at a location near them, regardless of where the backends run. The connection setup happens at that edge, and only the request itself crosses the backbone to a backend region. A regional passthrough load balancer cannot do this. It only forwards packets to backends in its own region and never terminates the connection near the user, so it cannot remove those round trips for distant users.
To get proximity routing with automatic overflow between regions, Google recommends putting every regional managed instance group into a single backend service as separate backends. The load balancer can then consider all of them for each request. By default it prefers healthy backends closest to where the request entered Google's network. When the nearest group fails health checks or approaches the target set by its balancing mode, the load balancer begins sending new requests to the next closest region that has capacity. This distribution is best effort, so a small share of requests may land in another region.
The proximity and overflow choice happens among the backends of one backend service, which is why the single backend service matters. The URL map selects a backend service by host and path, not by where the user is, so regions that sit in different backend services never compete for the same request. A load balancer also only knows its own backends, so the same logic means one global load balancer and one frontend address in front of every region.
Cloud Run services are regional, so protecting a serverless application against a regional outage means running a copy of each service in more than one region. The load balancer reaches each regional copy through a serverless network endpoint group, with one NEG per region, and those NEGs sit behind backend services on the global external Application Load Balancer. Users get a single anycast IP address, and each request goes to the closest region.
Serverless NEGs do not support active health checks. Instead, the load balancer relies on outlier detection to notice when a regional deployment starts returning errors and to shift traffic to another region.
For a global stateless web front end, the Google-recommended design combines these pieces:
The load balancer fails over between regions on its own. The network tier matters here as well. Choosing the Standard Network Service Tier limits you to regional load balancing, which cannot provide a single global anycast front end with Cloud CDN.
The URL map of an Application Load Balancer decides which backend service or backend bucket handles each request, based on its host and path. Host and path rules can send shop.example.com/cart/* to one backend service and shop.example.com/account/* to another, and the same mechanism can give each API path its own backend service. Requests that match no rule go to the default backend.
Because the URL map serves the public frontend, adding a path rule to it that sends traffic to a private tier exposes that tier through the public frontend.
URL maps support two kinds of matching, and they differ in what patterns they accept.
Path rules are simple. They do not evaluate regular expressions, and they allow a wildcard only at the very end of the path, right after a slash. A path such as /*/reports/* is not valid as a path rule because the first wildcard sits in the middle.
Route rules add richer matching. With pathTemplateMatch, a named variable such as {lang=*} matches exactly one path segment, and a trailing ** matches everything below it. Regular expression matching is also available through regexMatch, but only inside route rules.
Path templates are useful when a variable segment sits in front of a fixed folder. A template such as /{team=*}/reports/** matches that folder under every team segment, including segments added later, so the URL map doesn't need an edit each time a new one appears, and each route rule can send its matches to the right backend. Routing adapts to the existing URL layout, so URLs that are already printed or hardcoded keep working without content having to move under new prefixes. For the Professional Cloud Network Engineer exam, a pattern that needs a variable segment in the middle of the path, or a regular expression, points to route rules rather than path rules.
Our Professional Cloud Network Engineer practice exams cover the global external Application Load Balancer alongside URL map routing and Cloud CDN, with detailed explanations that drill these distinctions.