[Mageia-webteam] [Mageia-sysadm] Forum VM needs

[Michael scherer]

On Sat, Jan 15, 2011 at 11:46:38AM +0100, Maât wrote:
> Le 13/01/2011 13:29, Michael Scherer a écrit :
> 
> ----8<----
> > In any case, postgresql to mysql conversion doesn't look doable.
> >
> It could be done using things like TOS

Url ? 

> > But we are not ready, see my answer about gitorious. So here is a
> > proposal in 4 steps :
> >
> > Now :
> > - we do a git checkout every 5 minutes ( just need the url, and
> > testing )
> Ouch

Why ouch ?

5 minutes is too often, not often enough ?
( this is configurable, as this is based on cron )

I would prefer a pull based approach rather than push, because 
1) this is already written
2) this is more secure
 
But it may not be good for deployment.

> >>> Do you need modifications in phpBB sources ?
> >>>
> >> yes... some coming from (more or less) official patches (called "mods") some coming from ash 
> >> and i and perhaps other php devs
> > /me close eyes when reading "(more of less) official patches".
> >
> That's the way phpBB people deals with "mods" : authors (not phpBB Team) publish them on the mods 
> dedicated forum, and you are free to use (or not) them without warranty at all some are widely 
> used and well known to be stable...
> 
> That's why integration and testing stages are so important

Because software distribution is crappy ?
That's indeed a good reason ( now, do not be asthonished if we joke on php... )
 
> >>> If that's only for configuration files, we can host them on puppet svn
> >>> and deploy them with puppet.
> >>>
> >> if you want to play with puppet for production and having us not touching it with our dirty hands (^^) i'm perfectly ok with 
> >> that... but i neeed a integration and testing forum with cascaded by git cloning 
> >> (and i need to sync database of test from production to simulate version migration then cancel and replay as much a i need) 
> > Could we do it with a master/slave replication without much trouble ?
> > ( and possibly, one that would automatically recreate the db once it is
> > dropped )
> >
> > Or we can do snapshot using lvm, but I think this can be quite fragile.
> >
> Nope... master/slave replication will not fit my needs : i need 2 separate bases. The production i won't play 
> with it.
> 
> For testing : once i updated the code to the new version there can be SQL modifications : new tables, alter tables, 
> insert or update for rows...
> 
> This will be triggered from admin board of phpBB... with a master/slave this would be forbidden or would brake 
> replication

That's the idea, stop the replication for testing purposes. Hence my note about recreating it once we 
drop the slave.

We start with a clean db, we replicate.
Once we want to test the upgrade, we stop replication, we test.
We drop the db, we resync if needed

Once we finished the test, we drop the tables, and we restart the synchronisation.
 
> But we could perfectly use the last (nightly ,) production backup to populate the test base when i need to go 
> back to "clean production state"

This requires use to do backup, and so have at least a backup process
But we can also do it. I just think that replication would be less heavy 
on IO than dump/reload when needed ( and would be easier to automate ).

I wonder if we could distribute sql dump of the forum on mirror ( like we could for bugs ). I see
no obvious use for that at the moment, but liberating data can lead to unexpected usages, and would
be a interesting move ( or a at least, a interesting thought experiment, since there is private messages
involved, so this would requires cleaning  ) 

-- 
Michael Scherer