rc and smf

Bill Hacker wbh at conducive.org
Thu Feb 24 10:38:41 PST 2005


Dan Melomedman wrote:

Joerg Sonnenberger wrote:

On Wed, Feb 23, 2005 at 08:18:36PM -0500, George Georgalis wrote:

I've been meaning to give runit a spin. I've been very happy with
socklog (syslogd replacement) from smarden.org; daemontools and other
djb software too, despite the build on the target license requirement.
My first impression with Solaris Service Management Facility, is the
likely proprietary nature of it. While the license didn't exactly jump
out and bite me, I suspect the technology as well as the code is not
open. Is it?
I consider most of daemontools useless software, because it is build
on the wrong assumptions. A normal daemon should not need supervision,
just like a normal daemon can either write its own log files [if the
volume is high] or just use syslog.


It's built on the correct assumptions. You just don't understand all
possibilities why a service process can crash/exist, even if it's a
most perfect software ever written. Shit happens, and supervision will
attempt to restart the service for any reason it exits/killed. Including
human error, of course. What you've never killed the wrong process by
mistake? What if you set an rlimit, and the service reaches it, the
kernel will then kill the service. With supervision it will be
restarted, without it, the service will be down. Simple as that.
Programs have bugs, including memory leaks.
By and large in BSD-land shit does NOT 'happen'.

*BSD leans toward *correcting* misbehaving apps or replacing them.

Unreliable behaviour is not tolerated, even if it means not having the 
latest devices (SATA RAID controllers under FreeBSD 4.X).

Killed the wrong process? Your were at a prompt. Restart it.

If the battery in your car is always going flat, you replace it and 
check the alternator, regulator, and wiring.

You don't hire someone to follow you around with jumper cables.

Fix the *fault*, don't hire it a nurse.

Bill





More information about the Kernel mailing list