fix for error-ed out transactions on redshift

[drewbanin]

Problem

On Redshift (and Postgres?) a single invalid freshness query (ie. for an invalid table) would result in a dead transaction. Subsequent queries used the same connection (and transaction?) and so would fail with with the following error:

Database Error in source my_table (models/schema.yml)
  relation "public.my_table" does not exist

Database Error in source my_table2 (models/schema.yml)
  current transaction is aborted, commands ignored until end of transaction block

Database Error in source my_table3 (models/schema.yml)
  current transaction is aborted, commands ignored until end of transaction block

Solution

Clear the transaction before running freshness queries

Questions

Why are different freshness queries using the same transaction/handle? The fix applied in this PR works well (though it is certainly heavy-handed). Is there a better way to accomplish this?

Changes

• Reset transactions before running freshness queries
• add a --threads argument (reasonable)

[beckjake]

Why are different freshness queries using the same transaction/handle?

I would guess that the freshness calculation queries do an implicit begin transaction at some point in the adapter sql running process, and never commit or roll back. In wilt chamberlain, this is probably better since we explicitly grab/rename our thread's connection. But maybe not, we should test against it!

The fix applied in this PR works well (though it is certainly heavy-handed). Is there a better way to accomplish this?

We could explicitly commit/rollback after calculating each source freshness instead of before., or do things outside a transaction entirely. I don't think either of those are really any better, and certainly not any easier.

This PR looks good to me!