Re: Parallel (concurrent) inserts?

On Fri, May 25, 2012 at 3:04 PM, Ivan Voras <ivoras(at)freebsd(dot)org> wrote:
> Hello,
>
> I'm wondering if there is ia document describing which guarantees (if
> any) PostgreSQL makes about concurrency for various operations? Speaking
> in general (i.e. IO can handle it, number of CPU cores and client
> threads is optimal), are fully concurrent operations (independant and
> non-blocking) possible for:

By "fully concurrent" do you mean that there is no detectable
sub-linear scaling at all? I'm pretty sure that no such guarantees
can be made.

> 1) An unindexed table?

For concurrent bulk loads, there was severe contention on generating
WAL between concurrent bulk loaders. That is greatly improved in the
upcoming 9.2 release.

> 2) A table with 1+ ordinary (default btree) indexes? (are there
> constraints on the structure and number of indexes?)

This will be limited by contention for generating WAL. (Unless
something else limits it first)

> 3) A table with 1+ unique indexes?

If more than one transaction attempts to insert the same value, one
has to block until the other either commits or rollbacks.

> 4) A table with other objects on it (foreign keys, check constraints, etc.)?

Same as above. If they try to do conflicting things, they don't
continue operating concurrently.

Cheers,

Jeff