Marco Slot

Thanks, that resolves the issue.

> /me waits for this to blow up in our faces a year or three down the line. to be fair, catchup taking forever due to starvation blows up in...

In terms of [TPC-H](https://github.com/citusdata/citus/issues/1862), this PR adds support for Q13 and Q22.

> So, as far as I remember, Q21 is definitely not solved by de-correlation, but Q20 is still a candidate because the correlations are eq. operators. Q21 is already supported...

I was trying out this PR, but somehow I get the following output: ``` SELECT public.citus_stop_test_worker(:worker_2_port); WARNING: pg_ctl stop -D ../../worker.57638/data citus_stop_test_worker --------------------------------------------------------------------- 0 (1 row) SELECT * FROM ref;...

I think an alternative approach could be to have a layering that is more like DestReceiver -> Encoder (binary, text, ..) -> Byte sink (e.g. file, broadcast, ...). I think...

Long ago, the purpose of using a node string was probably to support sharding by an arbitrary expression, similar to how you can index on an expression. `pg_index.indexprs` is a...

Ì think part of the issue is that we consider almost anything a coordinated transaction, and coordinated transaction is the check we used in transaction_maangement.c on PREPARE TRANSACTION.

We should find a better way of dealing with PARAM_EXEC, since it can appear in other contexts as well (e.g. target list). In query pushdown planning we use the original...

Per separate discussion, we should refactor the lock that handles the `colocate_with :='default'` case where two concurrent create_distributed_table calls try to create the same co-location group because we started blocking...