Today will require extra coffee, after last night's efforts.

I think I need to pause and actually use the tools before trying to make them better...

To that end I've sent some mail to try and get my access to cloud resources repaired and am getting a reasonable Fedora VM snapshot in place.

After various bits of messing around I finally decided on using a vagrant setup for devstack. I hadn't tried this one in the past because it is dual node and I didn't think I had the horsepower for that. Seems I do and eventually I was able to get it working, modulo some relatively minor issues:

• Lots of bandwidth needed to get started.
• In my environment /etc/hosts on the two nodes needed to be aware of each other's names, otherwise the compute node was not visible to the controller.
• A recent bug in devstack meant that there was no auth URL, causing some metric to not be available (until fixed).
• Using QEMU means that some metrics are simply unavailable as support isn't in place (not sure where).

Todo

Specific

• Decode Vagrantfile and puppet. I'm never going to grok the system unless I can see how it is assembled. Looking at the Python code doesn't help that.¹
• Figure out a metric or two which will allow experimenting with alarms. Measurements reference.
• Do some queries.
• Do some queries over the API.

General

• Find out of grenade test failures are expected on Fedora.
• Investigate the Support ceilometer services patch for devstack.

Questions

• Sample publishing interval seems quite low in devstack. Is this default? Looks like ten minutes?
• With that publishing time, some alarms may be too slow?