Обсуждение: pg_restore failing with "ERROR: out of memory"

Aaron Brown <abrown@bzzagent.com> writes:
> I�m attempting to do something that should be a trivially simple task.  I
> want to do a data only dump from my production data in the public schema and
> restore it on another machine.

Does it really need to be data-only?  A regular schema+data dump usually
restores a lot faster.

Your immediate problem is probably that it's running out of memory for
pending foreign-key triggers.  Even if it didn't run out of memory, the
ensuing one-tuple-at-a-time checks would take forever.  You'd be better
off dropping the FK constraint, loading the data, and re-creating the
constraint.

There's further discussion of bulk-loading tricks in the manual:
http://www.postgresql.org/docs/8.2/static/populate.html

            regards, tom lane

Aaron Brown <abrown@bzzagent.com> writes:
> I’m attempting to do something that should be a trivially simple task.  I
> want to do a data only dump from my production data in the public schema and
> restore it on another machine.

Does it really need to be data-only?  A regular schema+data dump usually
restores a lot faster.

Your immediate problem is probably that it's running out of memory for
pending foreign-key triggers.  Even if it didn't run out of memory, the
ensuing one-tuple-at-a-time checks would take forever.  You'd be better
off dropping the FK constraint, loading the data, and re-creating the
constraint.

There's further discussion of bulk-loading tricks in the manual:
http://www.postgresql.org/docs/8.2/static/populate.html

                        regards, tom lane

Aaron Brown <abrown@bzzagent.com> writes:
> Yes, it kind of needs to be data only since I am pulling from a slonized
> database.  My experience has been if you don�t load the schema first with a
> schema from slony1_extract_schema.sh, we end up with all the slony triggers
> and crap in the dump.  If there is a better way of doing this, I�m
> definitely all ears.

Well, you'd need to take that up on the Slony lists, but I'd say that
Slony definitely ought to provide a better export mechanism than forcing
you to use a standard data-only dump.  At minimum they should teach that
script to emit two separate files: one to load before the data and one
after.  Building the indexes and FK constraints on already-loaded data
is a heck of a lot faster.

            regards, tom lane

Aaron Brown <abrown@bzzagent.com> writes:
> Yes, it kind of needs to be data only since I am pulling from a slonized
> database.  My experience has been if you don’t load the schema first with a
> schema from slony1_extract_schema.sh, we end up with all the slony triggers
> and crap in the dump.  If there is a better way of doing this, I’m
> definitely all ears.

Well, you'd need to take that up on the Slony lists, but I'd say that
Slony definitely ought to provide a better export mechanism than forcing
you to use a standard data-only dump.  At minimum they should teach that
script to emit two separate files: one to load before the data and one
after.  Building the indexes and FK constraints on already-loaded data
is a heck of a lot faster.

                        regards, tom lane