Presto Adapter (#1106, #1229, #1230)

[beckjake]

Resolves #1106, #1229, #1230

Implement a presto adapter. Presto required some massaging of the SQL adapter class as it has some novel properties - the local client does its own transaction+session management, the cursor doesn't have a status available, and queries do not appear to execute until results are fetched.

Working:

• Views/tables/seeds
• error handling
• configuration
• Linux tests, including integration tests against a local docker-compose setup on Linux.
• Transactions, apparently, though I suspect there are bugs here
• The various required macros
• A nice little script that generates a bunch of boilerplate for you

Not Working/incomplete:

• archives probably don't work and also don't raise an exception about it (see Presto: raise not implemented exceptions for unsupported functionality #1228)
• incremental models don't work and raise an exception about it
• I don't think the integration tests on Windows will work, and I think getting them to work on appveyor will be unpleasant, at best
• I can't come up with a nice way to test query cancellation and I'm not sure I trust it. ctrl+c does work locally but that doesn't mean queries are actually canceled.
• never tested against a real database
• only "no auth required" has been tested. I don't even know how to test kerberos!
• never tested with s3 backing instead of hdfs

Things to consider that might make plugin development easier:

• Should schema.py's column definitions get rolled into adapters and subclasses like connection management/adapter implementations? It seems wrong to have to add database-specific type information to core in support of a plugin!
• commit/rollback/begin transaction are not all defined in the same place. We canonically do rollback via a method but begin/commit transactions via queries. That seems odd and caught me by surprise when I was implementing this.

[drewbanin]

Some tiny comments here, will review more completely after poking around with this branch for a little bit :)

[drewbanin]

I think there were issues with this hack. For one, I think this splits on semicolons found inside of sql comments. Passable for the moment I think, but let's consider if it makes sense to handle this differently

[drewbanin]

I like the idea of pulling the logic out into the SQLAdapter. We need to do this basically everywhere

[drewbanin]

I was able to test this out on EMR Presto with both the Hive metastore, as well as with AWS Glue. In both cases, this code worked basically as expected! There is one caveat though -- you must set the following configs when launching your Presto cluster:

hive.metastore-cache-ttl=0s
hive.metastore-refresh-interval = 5s
hive.allow-drop-table=true
hive.allow-rename-table=true

Let's be sure to document these configs for our eventual release.

I do think that we need to do more stress testing here to feel really comfortable about deploying this code, but I feel pretty good about getting this merged and then making small tweaks as needed.

Between the quality of this code (it looks great!) and some initial testing in a realistic prod environment, I am happy to say LGTM, , 🚢, 🎉