run-one

This is a third-party reimplementation of Dustin Kirkland's run-one utility, with various bugfixes, better portability, and some semantic differences.

Requirements

A working /bin/sh, flock or lockf, sha256 or sha256sum, and pgrep/pkill.

Tested on FreeBSD 11 and Debian jessie.

Installation

/usr/local is the default PREFIX, provided to illustrate:

% make && sudo make install PREFIX=/usr/local

Linux users will want to use bmake, as punishment for all the times BSD users had to use gmake.

There is a deinstall target.

Usage

run-one will exit with EX_TEMPFAIL if another instance of the given command is running:

% run-one sleep 60 &
[1] 3544
% run-one sleep 60
zsh: exit 75    run-one sleep 60

run-this-one will kill the running command first:

% run-one sleep 60 &
[1] 16658
% run-this-one sleep 60
[1]  + exit 70    run-one sleep 60

run-one-constantly will run in a loop until interrupted, sleeping a second between successful runs, and doubling it up to a limit of 60 seconds between unsuccessful:

% run-one-constantly echo moo
moo
moo
moo

run-one-until-success does the same, but exits the first time it succeeds. Both commands log failures, so be mindful of your sysadmin:

% run-one-until-success false &
[1] 92276
% tail -f /var/log/messages |grep run-one
.. run-one-until-success[92276]: last run failed (rc=1); sleeping 2 seconds
.. run-one-until-success[92276]: last run failed (rc=1); sleeping 4 seconds
.. run-one-until-success[92276]: last run failed (rc=1); sleeping 8 seconds

run-one-until-failure does the opposite, and will never log failures.

% run-one-until-failure false
zsh: exit 1     run-one-until-failure false

Commands can of course be mixed:

% run-one-constantly true &
[1] 85332
% run-one true
zsh: exit 75    run-one true
% run-this-one true
[1]  + exit 70    run-one-constantly true

Notable differences with the original run-one

Runs on FreeBSD (and hopefully others).
keep-one-running is not provided.
run-one-constantly and run-one-until-* lock the supervision loop, rather than the command itself, so are safer to use in cronjobs etc.
Arguments are hashed more carefully, so run-one frob "foo bar" and run-one frob foo bar are correctly treated as distinct commands.
run-this-one only considers other instances of flock/lockf, instead of treating the argument list as a pattern to pass to pkill.
run-this-one only attempts to kill the children under lockf, not lockf itself, to allow them to exit cleanly.
run-this-one will time out after 60s if the children of the existing command refuse to die.
Uses $XDG_CACHE_HOME instead of ~/.cache or /dev/shm
Is less anal about ownership. If you want to have a group share a lock directory, have at it.

Author

This version of run-one is written by Thomas Hurst. Do not confuse it with the original.

Korkman / run-one