I'm working on getting our first production xen instance up and running
for things like koji, smolt, bodhi and the package database. Here are
some guidelines I'd like to put on the wiki as well as some changes I'd
like to make.
1) xen dom0 - RHEL5 (its just a black box, lets set it up and forget
about it)
2) after we update our RHEL4 boxes to RHEL5 We will maintain two types
of servers, RHEL5 and FC6 (this is because not all required packages are
in EPEL, and ultimately the FC6 boxes will probably become FC7,8,9
boxes. I'd like to keep just two types so we can create an appa[1-9]
and an appb[1-9]. I'd like to put appa and appb behind a vip that gets
balanced 1-9 RR. This will allow us to choose where we want to put our
apps, stable environments vs cutting edge.
3) test and production xen guests will exist only on test and production
xen hosts. (basically no mixing of production and test). Obviously
emergencies are one thing but in general this is just for organizational
purposes and because we may create a test network later and they won't
be able to mix.
4) xen instances will be named by hostname. Naming by purpose is
getting confusing.
5) We will probably start mixing hardware. Unlike before where we had
builder hardware and app hardware, we'll be bluring this line as the
need requires. I plan on building out a few more wiki instances before
the F7 release.
6) Kickstarts - I'm working on a xen-host and xen-guest kickstart
script. All of our boxes will be based off of one of these two scripts
(they'll be sent to the list soon for examination / suggestions).
Beyond that configuration will happen with puppet. Note: puppet can
pull in config files, install packages and restart services.
7) Monitoring - I'm working on a monitoring piece (basically a web page)
that will list exactly what xen guests are running on what xen hosts, if
they are pingable, etc.
What do you guys think?
-Mike
