Re: lock contention on parallel COPY ?

On Fri, 2008-09-26 at 20:07 +0200, Stefan Kaltenbrunner wrote:
> Tom Lane wrote:
> > Simon Riggs <simon@2ndQuadrant.com> writes:
> >> Stefan Kaltenbrunner <stefan@kaltenbrunner.cc> writes:
> >>> samples  %        symbol name
> >>> 55526    16.5614  LWLockAcquire
> >>> 29721     8.8647  DoCopy
> >>> 26581     7.9281  CopyReadLine
> >>> 25105     7.4879  LWLockRelease
> >>> 15743     4.6956  PinBuffer
> >>> 14725     4.3919  heap_formtuple
> >
> >> Probably loading a table with a generated PK or loading data in
> >> ascending sequence, so its contending heavily for the rightmost edge of
> >> the index.
> >
> > No, given that DoCopy and CopyReadLine are right up there, I think we're
> > still looking at the COPY phase, not index building.
> >
> > The profile will probably change completely once index building
> > starts...
>
> yeah this profile is only showing the COPY phase ...

Try using this Postgres core patch. It's a updated version of my
fast_copy.v4.patch from Patches: Bulk Insert tuning 20 Mar 2008

--
 Simon Riggs           www.2ndQuadrant.com
 PostgreSQL Training, Services and Support