On 2018-Sep-13, Tom Lane wrote:
> Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> > That's the problem all right. The solution is to drop all
> > index/constraint objects together in one performMultipleDeletions()
> > instead of performDeletion() one by one, as in the attached patch.
>
> Looks reasonable as far as it goes. Given that we no longer require
> any of this:
>
> - * Now we can drop the existing constraints and indexes --- constraints
> - * first, since some of them might depend on the indexes. In fact, we
> - * have to delete FOREIGN KEY constraints before UNIQUE constraints, but
> - * we already ordered the constraint list to ensure that would happen.
>
> can we make any simplifications in earlier steps? At the very least,
> look for comments related to this assumption.
Ah, I had looked, but not hard enough. In this new version I removed
some code in ATExecAlterColumnType that's now irrelevant. I tested this
by changing both lappend calls to lcons in that function; seems to
behave the same. (Also added more constraints to the test case.)
Another thing I found I can change is to move the add_object_address()
calls to the other loops scanning the same lists, so that we don't have
to walk each the two lists twice.
--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services