Support cluster_by on snowflake

[jon-rtr]

Issue #634 add cluster_by support for snowflake data warehouses.

[drewbanin]

We're currently experiencing a ton of pain around quoting on Snowflake. I think our default approach going forward will be to not quote identifiers in dbt, but instead ask the user to explicitly quote fields like this if desired. Do you buy that?

[drewbanin]

Another benefit of not quoting:

You made a great example of using cluster_by in the sample.dbt_project.yml file below:

                clusterby: ["date_day", "ad_group_id % 10"]

If dbt adds quotes, then I think the expression ad_group_id % 10 will look like an identifier, and Snowflake will do the wrong thing

[jon-rtr]

Yes, I buy that! We are definitely in the "don't quote identifiers and let the database do it's thing" camp. Allowing users to override that behavior on a case-by-case basis makes sense.

We have an edge case in our Vertica to Snowflake migration, where we manually created a handful of views that quote identifiers lower-cased versions of the table names. This is so that our pre-existing Tableau reports, which quote a lower-cased version of the table name, continue to function without change.

[drewbanin]

@jon-rtr I played around with this a little bit, and I think we still need to address the fact that Snowflake requires the table DDL to be specified along with create table ... as () when setting a cluster by.

For example, this works:

create table dbt_dbanin.debug (
    id int
) cluster by (id)
as (
    select 1 as id
);

but this does not:

create table dbt_dbanin.debug (
    id int
) cluster by (id)
as (
    select 1 as id
);

Do you know if it's possible to use cluster by without passing in table DDL?

[drewbanin]

@jon-rtr we're closing out some stale PRs -- did you have a chance to give this any more thought? I'm definitely keen to add support for cluster by on Snowflake, but I think there are some major aspects of dbt that need to change in order to make this happen. Let's discuss in #634 -- we can re-open this when we come to a good conclusion for implementation