(RADIATOR) HA Radiator under djb's daemontools

[Mariano Absatz]

Hi,

I'd like to share a "high availability" configuration I've been using whith 
Radiator for a couple of months. It uses a freely available software called 
"daemontools" (http://cr.yp.to/daemontools.html).

These are a set of tools that make the life of a server programmer easier. 
They are tiny pieces of software very simple and each one does JUST ONE THING 
(in the old Unix fashion).

They were developed by Dan J. Bernstein (http://cr.yp.to/djb.html) who I 
don't know personally, but, as evident from his messages in different mailing 
lists, suffers of an excess in personality. Nevertheless he's an excellent 
programmer and, if you don't try to argue with him, you can very happily use 
his software. He's mostly know as the developer of qmail 
(http://cr.yp.to/qmail.html) one of the good mail servers hanging around.

Enough intro... given that you have daemontools installed 
(http://cr.yp.to/daemontools/install.html), you should have a /service 
directory and a "svscan /service" process running.

Now you have to prepare a Radiator service directory. I use 
/app/Radiator/service, but you can use whatever you like.

This directory must have a "run" script executable by root that replaces 
itself (via exec) with Radiator running in the foreground, like the following 
one:

====================================================================
#!/bin/sh
# script to start Radiator with daemontools
#

exec 2>&1

exec /usr/bin/radiusd -config_file /etc/raddb/radius.cfg -foreground
====================================================================


Now, if you link -s this service directory to /service, svscan automatically 
starts "supervise" on this directory. eg: 
ln -s /app/Radiator/service /service/Radiator

supervise, executes the run script and, if the server stops, waits a few 
seconds, and starts it again.... forever and ever.

You also have a utility called svc to send signals to Radiator. If you say:
svc -h /service/Radiator (-h = hangup)
it would send a HUP signal.

If you say:
svc -t /service/Radiator (-t = terminate)
it would send a TERM signal and, if the server stops (which Radiator would), 
it'll start it again (effectively equivalent to a stop/start cycle).

If you say:
svc -d /service/Radiator (-d = down)
it would send a TERM signal but won't try to start it again.

svc -u /service/Radiator (-u = up)
will start Radiator again.

You also have 
svstat /service/Radiator 
which will tell you if Radiator is up or down, how it should normally be and, 
if it's up, for how long it's been up.

svok /service/Radiator
exists silently, but gives you a return value indicating if Radiator is 
running or not, which you can use from inside a shell script.

There are more options which I don't use, but you can check all of it in the 
daemontools on-line documentation.

In fact, I personally have more than one instance of Radiator on the same 
machine (I have up to 4 in one), I created one service directory for each 
instance and link the all of them to /service.


If you want, you can capture the standard output and standard error from 
Radiator (e.g. DBD usually sends error messages to stderr).

You must create a subdirectory of your service directory named "log", and put 
there an executable script named "run" which would run a logger.

daemontools include such a logger which is called multilog.

My "run" script looks like these 2 lines:
#!/bin/sh
exec multilog t ./main

this creates an auto-rotated log in /service/Radiator/log/main which includes 
every line sent by Radiator (or any included module) to standard output or 
standard error, preceded by a timestamp in tai format (see 
http://cr.yp.to/daemontools/tai64n.html).

If I want to check this log I do the following
tai64nlocal </service/Radiator/log/main/current | more

If I want to follow this output:
tail -f /service/Radiator/log/main/current | tai64nlocal

(tai64nlocal is used to translate the tai timestamps into humanly readable 
local times)


Hugh, Mike, if you want to add this explanation (however adapted) in the docs 
I don't have any problems (and I don't think Dan Bernstein would care either, 
see http://cr.yp.to/rights.html for his opinion on these kind of things).

http://www.open.com.au/radiator/ref.html#pgfId=361565

--
Mariano Absatz
El Baby
----------------------------------------------------------
My computer's sick. I think my modem is a carrier. 


===
Archive at http://www.open.com.au/archives/radiator/
Announcements on radiator-announce at open.com.au
To unsubscribe, email 'majordomo at open.com.au' with
'unsubscribe radiator' in the body of the message.