Remove idea of "instance" about myrrix-recommender HOT 4 CLOSED

No problem for me.

I don't know if it is really related, but have you ever considered changing the 
design such that a single CL and/or SL instance can manage multiple buckets? 
For the CL, this probably means watching multiple HDFS directories and managing 
multiple sets of MR jobs. For the SL, it would also mean that the request needs 
to contain information about which bucket you want to query.

Just a thought, don't know if it really make sense, but it could simplify 
deployment scenario's when there are multiple buckets.

As I dig into the scope of making this change, I find it may be easier to keep 
the existing scheme and fix the problem differently -- make sys/ "per 
instance". It turns out it is useful for a few small things. I'll post back 
when I know more.

Yeah I had thought that this was more simply managed by just using separate 
processes. That consumes a little more resource, and means managing several 
processes. But one way or the other it probably means managing several 
configurations. I had thought it simpler to make one instance deal with one 
model and encourage some isolation among instances.

Changed my mind, personally, after reading into the scope of the change 
further. There is some small use to instances after all. The magnitude of the 
change was more significant than I'd imagined too. Solving the issue I alluded 
to initially -- shared state in the "sys/" directory per bucket -- was easier 
than I'd expected. So, that's the outcome for now. Leave instances, fix the 
small existing problem directly.

I kind of like having "directories" under a single bucket personally, just from 
the perspective of keeping track of it all it's been pretty easy for me to run 
all instances under one bucket and just have a different naming convention for 
each logical instance (e.g. dev, test, staging, production). But honestly it 
doesn't make much of a difference to me, so whatever works best on the code 
side seems fine.