Dear Cagri,
While thinking this feature again, I had two points:
01.
Assuming the below schema definition, which only the subscriber has a partition:
```
pub - sub
tab tab
- tab_part1 // stores data from pub
- tab_part2 // stores data generated locally
```
In this case, REFRESH TABLE tab command truncate partition leaves. Data in
tab_part1 could be re-synced but tuples in tab_part2 were gone. Is it OK?
I felt several issues have been reported for the partition table, and this trend
would continue. How about separating the patch into two parts, 0001 is for normal
tables (reject for partition root) and 0002 adds the partition handling?
02.
Logical replication allowed to construct two(three)-way replication with the
subscription parameter origin=none. Can you evaluate the combination? I'm worry
because initial sync does not take care the origin of tuples.
Best regards,
Hayato Kuroda
FUJITSU LIMITED
