- SIGNAL HANDLING
- SCRIPTING CLIENT
- USAGE EXAMPLE
App::Procapult - Hand cranked process launcher
$ procapult -s ./ctrl
Then in another shell ...
$ socat - ./ctrl STATUS: stopped start sleep 3 STATUS: started 31563 sleep 3 STATUS: stopped start bash STATUS: started 31585 bash
And play with the bash in the first shell until you're bored then
stop STATUS: stopped die $
and with that, your procapult will expire in a puff of logic.
The idea for procapult is to have a process launcher that sits around doing nothing, until you tell it to start something, at which point it runs that until it exits or you tell it to stop it.
A procapult can, by design, only run one process at once - it's expected to be started in a screen/tmux/dtach window or an xterm, so the behaviour is as simple as possible.
To control your procapult, you make a unix socket connection to the control socket passed when you started it. Multiple clients are permitted at the same time, and if they step on each others' toes that's considered operator error on your part.
The protocol for the socket is so simple even I can understand it:
On connect, procapult sends its current status
When the status changes, procapult sends the new status
Status lines look like one of
STATUS: started 12345 some shell process STATUS: stopped
where 12345 is the pid of the process procapult is currently running
Valid commands are 'start', 'stop' and 'die'
'start some shell process' passes the string 'some shell process' to perl's exec()
'stop' causes procapult to send its process a SIGHUP
'die' causes procapult itself to commit harakiri
If your command is malformed or makes no sense, procapult sends an error line
Error lines look like
ERROR: some description of what went wrong
A successful command returns nothing, on the assumption that a status line will be along shortly to tell you what happened
That's all, folks.
procapult traps both INT and QUIT, because it's likely sat at the root of a terminal. So Ctrl-C and Ctrl-\ won't blow it up. If you actually want your procapult to fall down and go boom, you can either send it a SIGTERM, which incidentally is what 'kill 12345' will do anyway, or send it a die -
$ echo die | socat - /path/to/procapult/socket
You can also avoid needing to use socat (or your own unix socket logic) by using the built-in client:
# sends start, reads one line, prints, exits # $ procapult -s foo start some process name STATUS: started 12345 some process name $ # sends stop, reads one line, prints, exits # $ procapult -s foo stop STATUS: stopped $ # sends start, reads one line, exits if not started, reads until stop, exits # $ procapult -s foo run sleep 3 STATUS: started 12345 sleep 3 STATUS: stopped $ # sends die to kill the procapult, exits # $ procapult -s foo die $ # reads status, prints, exits # $ procapult -s foo status STATUS: stopped $ # reads status, prints, repeats until killed # $ procapult -s foo watch STATUS: stopped STATUS: started 12345 sleep 3 STATUS: stopped ...
The purpose for which this code was originally written was that I tend to run clusters of four xterms locally and connect them to matching server sessions. Which gets boring when my connection's a bit patchy. So what I can now do is -
# on the server # $ for i in tl tr bl br; do dtach -c ~/dtach/0$i -z bash; done
which starts four dtach sessions running bash (if you don't know dtach, think "screen for grumpy minimalists" and you won't be far wrong). Then on my machine I start my four xterms, and in each one start a procapult -
# in different terminals - # $ procapult -s ~/clus0/tl $ procapult -s ~/clus0/tr $ procapult -s ~/clus0/bl $ procapult -s ~/clus0/br
and then with that done, I can cause a full (re)connect simply with -
$ for i in tl tr bl br; do procapult -s ~/clus0/$i start ssh -t servername dtach -a dtach/0$i; done
noting that the -t is required to get a tty allocated even though we're not just letting ssh start a shell, and if any of the four haven't died then you'll just get an error from those, which procapult will duly print out and assume is now your problem. Obviously, if you care about noticing when something falls over, you wanted either 'run' instead of 'start' or to run 'status' or 'watch' as preferred.
While you can, in theory, email me, and I will, in theory, reply at some point, you're far better bugging me on #web-simple on irc.perl.org. I'm 'mst' on there, and my client is permanently connected, so while I might not reply until tomorrow if I've already called pubtime I should reply eventually.
mst - Matt S. Trout (cpan:MSTROUT) <firstname.lastname@example.org>
None yet - maybe this software is perfect! (ahahahahahahahahaha)
This library is free software and may be distributed under the same terms as perl itself.