My server panics occasionally (bad RAM that I haven't replaced yet - it's ECC but one bank occasionally hits an uncorrectable multi-bit error), which of course leaves various state files laying around on disk without the processes to match them.
In particular, fetchmail(1) is causing me problems; there's a ~/.fetchmail.pid file that's supposed to contain the PID of the running fetchmail daemon. It has to run (well, I *want* it to run) as me, not as root, so starting it from init isn't a good solution. I'd also like it to be recover gracefully if I have to kill it (or it crashes due to a bug), also contraindicating the obvious rc.local solution.
So far, the best solution I've come up with is a crontab entry (for me, not root):
0 * * * * if [ -f /home/athompso/.fetchmail.pid ];then read FP</home/athompso/.fetchmail.pid;if [ -d /proc/$FP -a -r /proc/$FP/exe ];then BIN=$(stat -c %N /proc/$FP/exe);BIN="${BIN#*-> }";BIN="${BIN#`}";BIN="${BIN%'}";BIN=$(basename "${BIN}");if [ "$BIN" == "fetchmail" ];then exit 0;fi;fi;fi;fetchmail
(or, more readably:) if [ -f /home/athompso/.fetchmail.pid ] then read FP < /home/athompso/.fetchmail.pid if [ -d /proc/$FP -a -r /proc/$FP/exe ] then BIN=$(stat -c %N /proc/$FP/exe) BIN="${BIN#*-> }" BIN="${BIN#`}" BIN="${BIN%'}" BIN=$(basename "${BIN}") if [ "$BIN" == "fetchmail" ] then exit 0 fi fi fi fetchmail
I particularly hate the method of extracting the imagename for a given PID, but haven't come up with anything better yet. Suggestions gratefully appreciated. Or even a better way of doing this altogether; this feels like a terrible hack - it works, but doesn't seem elegant.
FYI: I decided against using sed(1) to parse the output because doing it this way saves two fork(2)s and one exec(2). I know I don't need to worry about efficiency on a dual-Xeon server, but I'm trying to stay in the habit because I'm also working on 200MHz ARM5 platforms where it does matter.
(And who ever thought putting the output of stat(1) in typographical-style quotes would be a good idea???)
Thanks for any suggestions, -Adam
On 06/13/2010 02:36 PM, athompso@athompso.net wrote:
So far, the best solution I've come up with is a crontab entry (for me, not root):
0 * * * * if [ -f /home/athompso/.fetchmail.pid ];then read FP</home/athompso/.fetchmail.pid;if [ -d /proc/$FP -a -r /proc/$FP/exe ];then BIN=$(stat -c %N /proc/$FP/exe);BIN="${BIN#*-> }";BIN="${BIN#`}";BIN="${BIN%'}";BIN=$(basename "${BIN}");if [ "$BIN" == "fetchmail" ];then exit 0;fi;fi;fi;fetchmail
I would put the script into a file and run that from cron, it must be hard to maintain a crontab with more than a couple of entries like that :-)
(or, more readably:) if [ -f /home/athompso/.fetchmail.pid ] then read FP< /home/athompso/.fetchmail.pid if [ -d /proc/$FP -a -r /proc/$FP/exe ] then BIN=$(stat -c %N /proc/$FP/exe)
I'd use readlink(1) here.
BIN="${BIN#*-> }" BIN="${BIN#\`}" BIN="${BIN%\'}"
Which should get rid of the above.
BIN=$(basename "${BIN}")
Why do basename? BIN=${BIN##*/} saves a fork.
if [ "$BIN" == "fetchmail" ] then exit 0 fi fi fi fetchmail
My script might look something like (typed in mail):
#!/bin/sh pidfile=/home/athompso/.fetchmail.pid BIN= test -f $pidfile && read FP < $pidfile test -d /proc/$FP && BIN=$(readlink /proc/$FP/exe) case "${BIN}" in */fetchmail) ;; *) fetchmail ;; esac
FYI: I decided against using sed(1) to parse the output because doing it this way saves two fork(2)s and one exec(2). I know I don't need to worry about efficiency on a dual-Xeon server, but I'm trying to stay in the habit because I'm also working on 200MHz ARM5 platforms where it does matter.
Yes, forking is expensive.
Peter
-
athompsoļ¼ athompso.net -
Peter O'Gorman