Обсуждение: Possible changes to pg_restore

Greg Sabino Mullane recently shared a script to split schema-only dump
files into a table creation file and a trigger/index/constraint file, for
restore performance reasons.
(http://blog.endpoint.com/2010/01/splitting-postgres-pgdump-into-pre-and.html)

Would it be possible and worthwhile to add functionality to pg_restore so
that with one switch triggers/indexes/constraints could be ignored, and
with another switch only triggers/indexes/constraints would be restored?

This would then remove the requirement to split the file.

<hywel@hmallett.co.uk> writes:
> Would it be possible and worthwhile to add functionality to pg_restore so
> that with one switch triggers/indexes/constraints could be ignored, and
> with another switch only triggers/indexes/constraints would be restored?

You should probably go back and look at the archives from the last
go-round on that.  I believe it was during the 8.4 devel cycle.
Somebody (might have been Simon, but not sure) submitted a patch for
three-part output from pg_dump, it was discussed and sent back for
revision, and the revision never materialized :-(.  But IIRC there
was reasonably clear consensus on what the feature ought to look like.

            regards, tom lane

Tom Lane wrote:
> <hywel@hmallett.co.uk> writes:
>   
>> Would it be possible and worthwhile to add functionality to pg_restore so
>> that with one switch triggers/indexes/constraints could be ignored, and
>> with another switch only triggers/indexes/constraints would be restored?
>>     
>
> You should probably go back and look at the archives from the last
> go-round on that.  I believe it was during the 8.4 devel cycle.
> Somebody (might have been Simon, but not sure) submitted a patch for
> three-part output from pg_dump, it was discussed and sent back for
> revision, and the revision never materialized :-(.  But IIRC there
> was reasonably clear consensus on what the feature ought to look like.
>
>             
>   

What is more, the changes done for parallel pg_restore actually did some 
of the requisite work of classifying members, as can be seen in this 
definition in pg_dump.h:
   typedef enum _teSection   {       SECTION_NONE = 1,           /* COMMENTs, ACLs, etc; can be   anywhere */
SECTION_PRE_DATA,          /* stuff to be processed before data */       SECTION_DATA,               /* TABLE DATA,
BLOBS,BLOB COMMENTS */       SECTION_POST_DATA           /* stuff to be processed after data */   } teSection;
 


Of course, you'd need to figure our how to handle the SECTION_NONE 
entries, but other than that it should be pretty simple, I think.

cheers

andrew