[Mageia-sysadm] Puppet Report for valstar.mageia.org

[Michael Scherer]

Le mardi 14 décembre 2010 à 13:12 +0100, report at valstar.mageia.org a
écrit :
> Tue Dec 14 13:12:11 +0100 2010 Puppet (err): Could not retrieve catalog from remote server: end of file reached
> Tue Dec 14 13:12:11 +0100 2010 Puppet (err): Could not retrieve catalog; skipping run

Ok so after digging, for some yet unknow reason, it seems puppet can
only serve 1 client at the same time :

Tue Dec 14 13:12:08 +0100 2010 Puppet (notice): Compiled catalog for
jonund.mageia.org in environment production in 0.54 seconds
Tue Dec 14 13:12:12 +0100 2010 Puppet (notice): Compiled catalog for
valstar.mageia.org in environment production in 0.69 seconds

I suspect sqlite to be the culprit for stored configuration, enabled 1
week ago. This or the fact that we use webrick ( ie, slow and default
implementation ).

We could switch to postgresql, but this would cause some subtle issue of
bootstrapping, since we use stored config to exports node information,
and so to create postgresql database. ( see commit 525 and 527 ).

So puppet would requires to create the database for using stored config,
but the database requires stored config to work before knowing it must
create the db.

OTOH, our config is already no longer be bootstrapped from scratch since
we started to use custom fact ( rev 339 ). So if we want one day to be
able to reinstall from scratch, we will need to write a procedure
( would be a good junior job ) and do a test before :).

Alternatively, it may be a webrick problem ( according to
http://projects.puppetlabs.com/projects/puppet/wiki/Using_Mongrel ), and
using mongrel ( not packaged, but I already gave enough work to our
resident ruby expert ) or passager ( slightly more complex to deploy,
and would pull apache ) would be a solution

In the mean time, I simply restarted puppet on valstar, hoping it
doesn't start at the same time as another client. 

-- 
Michael Scherer