Re: cfengine cronjob on pool of servers

[Christopher Browne]

> Christopher Browne wrote:
>>> Can cfengine do this? Nicelly, easy and without hacks please!
>>> 
>>> We need to schedule a script to run on ONLY ONE of the N hosts.
>>> The catch is that any host could be down at any moment in the future,
>>> so schedulling cronjob on one of the hosts is not a reliable option.
>>> Also the job cannot be run more that once (ie cannot be run from
>>> multiple hosts).
>> 
>> This isn't the sort of thing that cfengine is designed to provide
>> answers for.
>> 
>> What you actually need is some sort of central authority; part of the
>> point of cfengine is to not have that vulnerability.
>> 
>> If I had a database server around, I'd be inclined to use it to manage
>> this sort of "locking" problem.
>> 
>> All N hosts would *attempt* to run the cron job; they start by telling
>> the DB server "I am the only one that will be running this process."
>> Only one would succeed (let's say they all try to insert the desired
>> time and job name into a table with unique index on time/job); the
>> "winner" gets to continue and run the job.
>
> Simplisticly, that could probably be done by creating (lock) files.  The
> cfengine host that creates the file first wins.  Of course, this assumes an
> NFS setup and, thus, that is the vulnerability in the scheme.

I suggested the database as a mechanism because:

a) You probably have one around if you have so many servers :-)

b) Databases tend to have well-implemented locking mechanisms.

Come to think of it, cfengine uses Sleepycat DB, which has a lock
management subsystem.

<http://www.sleepycat.com/docs/ref/lock/intro.html>
-- 
output = ("cbbrowne" "@" "gmail.com")
http://linuxdatabases.info/info/slony.html
Rules of  the Evil Overlord  #225. "I will  explain to my  guards that
most people have their eyes in the front of their heads and thus while
searching  for someone  it makes  little sense  to draw  a  weapon and
slowly back down the hallway." <http://www.eviloverlord.com/>