Signal Handling

When it receives a SIGHUP, it sends a SIGTERM to each of its children, and then resumes its normal operations.

When it receives a SIGTERM, it sends a SIGTERM to each of its children, sets an alarm(3) "die timeout" handler, and waits for each of its children to die. If all children die before this timeout, process manager exits with return status 0. If all children do not die by the time the "die timeout" occurs, the process manager sends a SIGKILL to each of the remaining children, and exists with return status 1.

FCGI::Engine::ProcManager uses POSIX::sigaction() to override the default SA_RESTART policy used for perl's %SIG behavior. Specifically, the process manager never uses SA_RESTART, while the child FastCGI servers turn off SA_RESTART around the accept loop, but re-enstate it otherwise.

The desired (and implemented) effect is to give a request as big a chance as possible to succeed and to delay their exits until after their request, while allowing the FastCGI servers waiting for new requests to die right away.