Open Stack

Clark and I discovered yesterday that Postgresql is no longer running in 
the gate. Turns out a set of subtle changes over in devstack & devstack 
gate meant we weren't setting things correctly, so we've been running 
for a couple of weeks without postgresql.

On the plus side we didn't break any previous tempest tests. On the down 
side, we merged a few more tempest negative tests that generate stack 
traces in Postgresql because we very often send bad data all the way 
down to the database.

MySQL happily will put anything in a where clause (regardless of type, 
size, smelliness), and return "not found". Postgresql is actually typed, 
and throws a DataError if you try to use string as an int, or too long 
of an int as and int. When we fixed this for get_instance(id) we added 
catching the DataError in the SQLA layer, translating to an InvalidID to 
pass it to the request layer, where it gets turned into a NotFound for 
the user.

Now that we're seeing it in another couple of places, we should probably 
think about more systematic ways to go about this.

Also, if Postgresql is really important to someone, we really do need a 
champion on it that's going to aggressively be trying to expose these 
issues. There are some quick fixes (more DataError -> InvalidID 
catches), but the right answer is probably a more systemic approach 
about preventing these kinds of requests to get down to the lower layers 
at all.

But, this still needs a champion and a doer to get us there. I was that 
for Grizzly to prove it was doable, and clean out a set of issues, but 
for Havana this is pretty low priority for me, so someone else is really 
needed here to keep postgresql functioning in the gate.

	-Sean

-- 
Sean Dague
http://dague.net