perfork

Perfork is a term used to describe a concurrency pattern in which a separate child process is created to handle each incoming task or connection. In this model, a parent process listens for requests and forks a new child to process the request, with the child typically handling the work and exiting upon completion. This approach contrasts with prefork, which maintains a fixed pool of worker processes, and with thread-per-request models, where a new thread handles each task.

In a typical perfork implementation, the parent accepts a connection or task, forks a child process, and the child processes the request, communicates results, and then terminates. The parent continues listening for new work and reap any terminated children to prevent zombies. Forking is usually performed via the Unix fork system call, and security or isolation can be enhanced by further spawning or exec’ing dedicated binaries. Resource limits, such as memory usage and the number of concurrent processes, are important considerations, and cgroups or other containment methods may be used to constrain perfork workloads.

Perfork offers strong process-level isolation, making it easier to confine faults or untrusted tasks. However, it can be expensive in terms of CPU and memory overhead, especially under high concurrency, due to repeated process creation and context switching. As a result, perfork is less common for high-traffic services, which often prefer thread pools or event-driven architectures, while it may still be attractive for security-sensitive workloads or environments with strict isolation requirements.

Related concepts include prefork, thread-per-request, process pools, and CGI-style spawning.