Move all non-equals query columns to end of key

[c-w]

If we want to query some columns in a cluster key by an equals relationship (e.g. externalsourceid = 'aljazeera') and others by a non-equals relationship (e.g. (tilex, tiley) <= (1, 1)) then all the equals columns have to come before all the non-equals columns, otherwise we get an error like this:

InvalidRequest: Error from server: code=2200 [Invalid query]
message="PRIMARY KEY column "pipelinekey" cannot be restric
ted (preceding column "tilex" is restricted by a non-EQ relation)"

[jcjimenez]

LGTM

[Smarker]

LGTM

[erikschlegel]

whats the motivation behind re-arranging the PKs for pipelinekey, externalsourceid?

[c-w]

@erikschlegel This PR is essentially just moving all the keys on which we do a range query (like <= or >=) to the end of the key definition. Any key for which we do a point query (i.e., =) has to come before the range queries for Cassandra to accept the query.

For example, this will not work:

CREATE TABLE foo (
  col1 text,
  col2 text,
  col3 int,
  col4 int,
  PRIMARY KEY ((col1), col3, col4, col2)
);

SELECT *
FROM foo
WHERE col1='a'
  AND col2='b'
  AND (col3, col4) <= (10, 10)
  AND (col3, col4) >= (0, 0);

However, the query would work with this table definition since all the point query columns (col1 and col2) come before all the range query columns (col3 and `col4) in the key definition:

CREATE TABLE foo (
  col1 text,
  col2 text,
  col3 int,
  col4 int,
  PRIMARY KEY ((col1), col2, col3, col4)
);

So the motivating for moving the pipelinekey and externalsourceid is to put them before columns like tilex or tiley so that we can range query on the latter while point querying on the former.

[erikschlegel]

LGTM