Harald Armin Massa wrote:
> - is pg_dump supposed to produce a "ordered dump" with --data-only?
> (within documentation there is only a warning that --table will not dump
> the depending ones)
> [than I have to create a extract to reproduce that "bug"]
>
> - or if it is simply not implemented and an enhancement request;
> (so I should file an enhancement-request and a "warning" -patch to the
> pg_dump docu, that pg_dump --dataonly MUST be used with
> --disable-trigger or similiar for backup-purposes)
>
> - or if it is even on a theoretical basis impossible to derive the
> correct order. [circular foreign keys came to my mind] -> I also should
> submit a patch to documentation
Even _if_ it's theoretically possible, it would probably be too slow,
i'd say. I'd be possible if you dependency-graph between your tables is
acyclic - then simpily reordering the dump-order would be sufficient.
But if you have some sort of cycle, you'd at least need to give up
table-wise restoration, and instead e.g. restore 5 rows from t1, then
2 rows from t2, now 3 rows from _t1_ _again_. And if a _record_ some
way depends on itself (So, the dependency graph between you _records_
even shows circles), then you'd sometimes need to insert a record, then
insert a record that depends on the first, and then update the first
record to now depend on the second. or something similar. And, with
not-null contraints in place, this might be impossible without violating
them :-(.
So, I'd say you should either create backups using --disable-triggers,
or use the custom-format - which let's you specify --disable-triggers,
and a few other options at restore, rather then at backup time.
greetings, Florian Pflug
