Investigate pipe-file protocol (driver-upsd) use to kill old driver instances

[jimklimov]

Sometimes drivers get wedged, so restarting it requires not just a SIGTERM but a SIGKILL of some known PID. Part of the problem is learning that PID, especially if earlier attempts to restart failed and the file got deleted.

Some brainstorm ideas:

• Driver might tell its PID to UPSD when they connect, so the data server would remember and report it (constrained to some link e.g. localhost, credentials, etc?) - so a new driver starting can ask if a copy by same name exists and what its PID is.
• Extend driver-server protocol to let data server tell the driver to die, reconnect, etc. (if not wedged) - see also Investigate dual-process model like upsmon for drivers #1663
  • maybe this can help also with powerkill handling where we now stop a driver to run its new process with kill option
• Can a new driver instance figure out the PID by existing pipe file (like fuser it) and/or attach to same pipe to tell the older sibling to report the PID which we can send signals to (kill), or explicitly to die off (if not wedged)?