Re: [PATCH] [1/2] Improve check for daemon running

[Maynard J. <may...@us...>]

Andi Kleen wrote:
> I had some problems with opcontrol claiming a daemon was running
> when it wasn't. I think that was due to reused PIDs.
> This patch adds a common shell subfunction to check for the daemon
> running and always checks if its process name contains oprofiled.
[snip]
>  do_dump_data()
>  {
>  	# make sure that the daemon is not dead and gone
> -	if test -e "$SESSION_DIR/lock"; then
> -		OPROFILED_PID=`cat $SESSION_DIR/lock`
> -		if test ! -d "/proc/$OPROFILED_PID"; then
> -			echo "dump fail: daemon died during last run ?" >&2
> -			return 1;
> -		fi
> -	else
> +	if ! daemon_running ; then
> +		echo "dump fail: daemon died during last run ?" >&2
>  		return 1;
BUG!  If you stop the daemon and then happen to run any opcontrol command that does do_dump_data (deinit, stop, shutdown all do an implicit dump), you get this "dump fail" message.

I agree with the general idea of the change, but I think we'll wait on accepting this patch (once the "dump fail" problem is fixed) until after we release 0.9.5.  I'd rather have the community bang on it for a while before putting it in a release.

Thanks.
-Maynard
>  	fi
>