"Marc G. Fournier" <scrappy@hub.org> writes:
> Did I ever send in a bug report about pg_dump 'crashing' while dumping a
> database where one of the tables gets drop'd while the pg_dump is running?
> Not the easiest thing to reproduce, mind you, cause its a matter of that
> one in a million timing thing ... but, if you run pg_dump against a
> database where one of the tables yet to be dump gets drop'd, the pg_dump
> will crash, as opposed to just skipping it and continue with those tables
> that still exist ...
I'd be inclined to fix this by having pg_dump issue a LOCK IN ACCESS
SHARE MODE against each table as it reads the table name from pg_class.
Not by allowing tables to disappear from under us after the dump starts.
The idea of pg_dump is to produce a consistent snapshot, no?
Even that is not *perfectly* secure since the locking phase will take
more than zero time, but it seems close enough.
regards, tom lane