Hi
I'm currently investigating how much work it'd be to implement arrays of
domains since I have a client who might be interested in sponsoring that
work.
The comments around the code handling ALTER DOMAIN ADD CONSTRAINT are
pretty clear about the lack of proper locking in that code - altering a
domain while simultaneously add a column with that domain as a type
might result in inconsistencies between the data in that column and the
domain's constraints after both transactions committed.
I do, however, suspect that ALTER TABLE is plagued by similar problems.
Currently, during the rewrite phase of ALTER TABLE,
find_composite_type_dependencies is used to verify that the table's row
type (or any type directly or indirectly depending on that type) is not
used as a column's type anywhere in the database.
But since this code does not take any permanent locks on the visited
types, it seems that adding such a column concurrently is not prevented. If the original ALTER TABLE changed a column's
type,data inserted into
the newly added column before the original ALTER TABLE committed will
have a type different from what the catalog says after the original
ALTER TABLE commits. Or at least so I think - I haven't yet tested that
theory...
I am aware that since a commit fest is currently running, now might not
be the best time to bring up this topic. Since I feared forgetting this
all together, I decided to still post now, though. I figured people
still have to option to ignore this for now if they're busy with getting
those patches committed.
best regards,
Florian Pflug
