Cloud Load Balancer backend connection pooling

Problem

Distinct requests from distinct clients were proxied by Google Cloud Load Balancer (GCLB) to the backends through a same TCP session

Environment

  • Global external HTTP(S) load balancer

Solution

This is the intended behaviour known as 'Connection Pooling'.

Cause

Connection pooling in Google Front End (GFE) refers to the efficient management of incoming connections from clients to GFE servers. The goal of connection pooling is to reduce the overhead of establishing a new connection for each incoming request and improve the scalability of the system and reduce latency.

In GFE, connection pooling is implemented by reusing existing connections between clients and GFE servers. When a client requests a connection, the GFE checks if there is an available connection in the pool. If there is an available connection, the GFE uses that connection to handle the request. If there is no available connection, the GFE establishes a new connection to one of the GFE servers.