So, we’ve been trying to get a nice stable config running for
deployment with SwitchTower, spinner spawner and reaper, and I have
to say, they are, as it says in the title, teh suxx0r.
I’m annoyed, and wrote this while annoyed
The problems are rife:
- Because lighttpd has no control over the fcgis, the restart
process is delicate. Lighty frequently cannot “see” new fcgis.
- Reaper does not successfully kill all dispatch.fcgi processes in a
horrifyingly large percentage of calls
- SwitchTower frequently hangs on the last task forever, apparently
waiting for some kind of output
- If you miss putting in your password, you get a trace dump from
the SSH lib, not "please enter your password
- The assumptions ST makes about process are very strange - I have
always used a production tag (in SVN and even back in CVS days.
Promoting a random revision number seems dumb and very hard to manage
outside of ST.
- If a single small thing goes wrong, ST falls on its sword, hides
the output of the task that had the problem but is otherwise
EXTREMELY verbose about nothing useful.
I was going to write a list of things I think need to happen with ST,
but you know, I’m not sure if I ever want what it offers, even on
large clusters, hacking ssh commands explicitly is just a loop
through a host list in a text file… Deployment isn’t complex when
you have tags.
Also, updating takes a hell of a lot less time than a full checkout,
and a rollback is as simple as copying one tag to another. 4 line
alex black, founder
the turing studio, inc.
2600 10th street, suite 635
berkeley, ca 94710