[Piuparts-devel] Thoughts on piattiNG

[Holger Levsen]

Hi,

with one disk failing (leaving us zero further faulty disk tolerance, though I 
made a backup of all logs last week ;) this has become more urgent.

Plus, with the new debian cluster in place, the hw is actually available.

On Dienstag, 26. März 2013, Dave Steele wrote:
> When do you expect the transition to happen for piuparts? How firm is
> the schedule?

DSA wants us to define our wishes *now*. And then the move will happen, in 
cooperation with us/me. 
 
> Is there a standard/preferred provisioning mechanism for defining the
> VMs (i.e. puppet, chef), or is it to be hand-rolled? 

DSA uses puppet.

> Is there any prep
> work that needs to be considered for this?

we should discuss whether we'd want to run from packages. I think I'd prefer 
this (by now), but then I know in the past DSA rather prefered local 
installation.

> Do you see separate VM slave instances in the short term, or is that a
> farther out thing?

No, that can/will/should happen now. I'm actually thinking about

2 amd64 slaves
1 i386 slave
1 armel slave running on that i386 slave in qemu

> What is the recovery strategy, should the master logs be lost?
> 1) No strategy. Start up another instance. It will catch up in a
> couple of months.
> 2) Resources can be lined up to release a horde of slaves to catch up,
> in about a week.
> 3) Implement a log/config backup process on a slave or other VM.
> Recovery in hours.

For now, I'd say a mix of 1-3 ;-)

> How much concern is there about configuration concurrency across
> master and slave VMs? What is the preferred strategy for keeping them
> in sync? 

One way could be to turn the (so far theoretical) disadvantage of  me being 
the single point of (maintainance) failure into a feature: I do the 
configuration and thus keep it consistent ;-)

But really, it's not that much: a package version (or a git tag) plus very few 
configuration files, which are in git at least, if not even part of the 
packages.

And btw I do think that Andreas should get access (and sudo rights) now.
(It's easy as Andreas is a DD.)

>      On several debian.org machines, services are currently located
>      in the /org directory. With the release of 'wheezy', DSA will no
>      longer support this directory and also begin removing existing
>      symlinks from /org to /srv. Please check which directory your
>      service is currently using and update your
>      scripts/configuration where neccesary.
> Do you expect to move to /srv, or use stock locations?

I couldn't care less which we think/pretend we use. DSA will put our stuff in 
/srv/$foo and if we want to use a link from $obscure_location to there we 
surely can.

IOW: lets stick with /srv/piuparts.debian.org

I'll send a reply to this (hopefully) with a URL for the "hw wishlist" to 
Santa Claus^w^wDSA... :-)


cheers,
	Holger
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 828 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.alioth.debian.org/pipermail/piuparts-devel/attachments/20130416/79d68fc4/attachment.pgp>