Test and document making the HaaS API server multi-node

[henn]

In theory, other than headnode stuff (which we eventually want to remove), it should be possible to make the HaaS API server multi-node as the DB is the point of synchronization.

This could be helpful in large environments where the API server needs to scale horizontally.

We should test and document this.

[zenhack]

In theory, other than headnode stuff (which we eventually want to remove), it should be possible to make HaaS multi-node as the DB is the point of synchronization. We should test and document this.

Could you please specify exactly what you mean by "multi-node"?

[henn]

Running multiple HaaS APIs (as opposed to the network action daemon)
under separate webservers.

[zenhack]

Running multiple HaaS APIs (as opposed to the network action
daemon) under separate webservers.

Ah, I see. Yeah, that should work, though we will have to remove the
headnode stuff or do something like OVS so it doesn't need to be on the
same node.

We also still need to add support for something other than sqlite

My inclination is: we need both of those things anyway, this will have
to wait until we have them.

-Ian

[henn]

I thought we supported mysql? Doesn't SQLAlchemy get us support for
several DBs?

[zenhack]

I thought we supported mysql? Doesn't SQLAlchemy get us support for several DBs?

Almost. It makes it a lot easier to do portability, but there are a few things you need to pay attention to. Off the top of my head, MySQL requires you specify the length of VARCHAR fields (which are exposed as String in SQLAlchemy), and we're not doing that. Postgresql doesn't require this, and for all I know that may just work, we should check.

…

-Ian