lxinitd pid files

How lxinitd processes pid files.

If you configure a service with a pid file.

service /var/run/nginx.pid /sbin/nginx

lxinitd will ensure the service is restarted if it fails. This page details how the monitoring works.

fork()

Some services fork() a daemon process and exit when they boot, the lxinitd process does not know the pid to monitor.

A forked process is not a child process, so lxinitd can not necessarily wait for SIGCHLD signal to indicate it has exited.

These types of daemon write their pid to a file.

lxinitd periodically checks for the existence of the file until it knows the service's pid.

If the pid disappears from the ps list lxinitd restarts the process.

The existence of a service pid is checked every 3 seconds. This is called check().

re-parenting

When any process forks and the parent process terminates, the Linux kernel re-parents the orphaned process to pid 1. lxinitd may or may not be pid 1.

If re-parenting happens lxinitd will be notified with SIGCHLD when the process dies. This is called wait().

If lxinitd is pid 1, both wait() and check() notify lxinitd of the same process's death; there is a race condition that must be fixed.

The filesystem's atomic operations are used.

The first of wait() and check() to notice process has died deletes the pid file.

The second to notice the process has died, if the pid file is missing, does nothing.

If the pid file exists, check() verifies the pid is alive.

If the pid file exists, and check() fails, this indicates the process has failed again.

If an external process deletes the pid file, lxinitd will reboot the process if it knows the pid, but may not be able to reboot it a second time.