kill-hung-procs.sh

#!/bin/sh
#
# Look for any known-to-hang processes that have been running for longer than
# 2 hours, these are not caught by the ulimit -t set by pbulk as they are not
# using any CPU time.
#
# Run from cron and redirect output to a log
#

if [ "$1" = "-n" ]; then
	dry_run=true
else
	dry_run=false
fi

case "$(uname -s)" in
NetBSD)
	ps_args_pbulk="-o etime= -o pid= -o args= -U pbulk -x"
	ps_args_log="-ww -o user,pid,lstart,etime,args"
	process_restart=true
	;;
*)
	ps_args_pbulk="-o etime= -o pid= -o args= -U pbulk"
	ps_args_log="-fo user,pid,etime,args"
	process_restart=false
	;;
esac

kill_or_restart()
{
	pid=$1; shift

	if ${dry_run}; then
		if ${process_restart}; then
			echo "Would stop/start PID ${pid}"
		else
			echo "Would kill PID ${pid}"
		fi
		return
	fi

	#
	# On some OS it's enough to stop and restart processes to get them
	# running again (notably NetBSD with its broken libpthread).
	#
	if ${process_restart}; then
		kill -STOP ${pid}
		sleep 1
		kill -CONT ${pid}
	else
		kill -9 ${pid}
	fi
}

log_and_kill()
{
	pid=$1; shift

	# Output date and running command for the log.
	date '+%Y-%m-%d-%H:%M:%S'
	ps ${ps_args_log} -p ${pid}
	kill_or_restart ${pid}
}

ps ${ps_args_pbulk} | while read time pid cmd; do
	#
	# Match when the ETIME field is 2 hours or longer.  Account for OS
	# differences, most have leading 0, NetBSD does not.
	#
	# If anything has been running for over a day just kill it, it's
	# highly unlikely to be making forward progress.
	#
	case "${time}" in
	*-*:*:*)
		log_and_kill ${pid}
		;;
	0[2-9]:*:*|\
	[2-9]:*:*|\
	[1-9][0-9]:*:*)
		#
		# Only match either known fail processes or anything running
		# from within the work directory, skipping known false
		# positives such as Rust.
		#
		case "${cmd}" in
		*lang/rust*|*ghc94*)
			# Do nothing, likely legitimate long-running process.
			;;
		/Users/pbulk/*|\
		/home/pbulk/*|\
		*Configure*|\
		*cmake_autogen*|\
		mplayer|\
		*py-scipy*|\
		*" ./configure "*|\
		./*|../*)
			log_and_kill ${pid}
			;;
		esac
		;;
	esac
done