* Tom Lane (tgl@sss.pgh.pa.us) wrote:
> Stephen Frost <sfrost@snowman.net> writes:
> > The issue is that pg_dump wants to lock the table against changes, which
> > is really to prevent the table to change between "we got the definition
> > of the table" and "pulling the records out of the table." It's not
> > immediately obvious, to me at least, that there's really any need to
> > lock the tables when doing a schema-only dump. Accesses to the catalogs
> > should be consistent across the lifetime of the transaction which
> > pg_dump is operating in and a schema-only dump isn't doing anything
> > else.
>
> This is the standard mistake about pg_dump, which is to imagine that it
> depends only on userspace operations while inspecting schema info. It
> doesn't; it makes use of things like ruleutils.c which operate on "latest
> available data" rules. Accordingly, no we're not going to skip taking
> the table locks. At least not without a ground-up rewrite of that whole
> mess, which as you know has been discussed multiple times without anything
> useful happening.
There's two different points here- the first is the whole discussion
around why pg_dump is depending on the backend for bits and pieces but
not everything, but the second is- aren't the accesses from ruleutils.c
now using an MVCC snapshot? Certainly there's a comment about that
happening for pg_get_constraintdef_worker(), and other parts appear to
go through SPI, but not everything does.
Of particular relevance to this appears to be trigger and index
handling, considering that the only thing pg_dump locks is relations
anyway, much of the rest isn't relevant.
Thanks,
Stephen