Re: speedup COPY TO for partitioned table.

You're right that order in find_all_inheritors is deterministic. But it's not always the same with the order of SELECT output. You can quickly see what I mean by running a slightly modified version of the example that you shared in your first email:

CREATE TABLE t3 (a INT, b int ) PARTITION BY RANGE (a);

-- change the order. first create t3_2 then t3_1
create table t3_2 partition of t3 for values from (11) to (15);
create table t3_1 partition of t3 for values from (1) to (11);
insert into t3 select g from generate_series(1, 3) g;
insert into t3 select g from generate_series(11, 11) g;

postgres=# COPY t3 TO STDOUT;
11      \N
1       \N
2       \N
3       \N
postgres=# COPY (SELECT * FROM t3) TO STDOUT;
1       \N
2       \N
3       \N
11      \N

Notice that "COPY t3 TO STDOUT" changes the order since the partition t3_2 has been created first, hence it has a smaller OID. On the other hand, SELECT sorts the partitions based on partition boundaries, not OIDs. That's why we should always see the same order regardless of the OIDs of partitions (you can see create_range_bounds() in partbounds.c if interested in more details). One thing that might be useful in the COPY case would be using a partition descriptor to access the correct order of partitions. I believe something like (PartitionDesc) partdesc->oid should give us the partition OIDs in order. 

Thanks,