Обсуждение: problem with pg_restore

Hi there,
   I am a newbie, trying to move a pg db from PostgreSQL 7.13  on RedHat 7.3 to PostgreSQL 7.42 on FreeBSD 4.9.  The db has one table with 32,000 rows, each containing a blob.  The blobs range in size from 2K to about 300K.  The entire backup .tar file is just over 3GB.

   I first tried to restore using the folloing command:
   pg_restore -d athena backup.tar

   It ran for a while (creating all the table defs) and the I got the following messages:
   pg_restore: ERROR:   literal carriage return found in data.

   So, I had some bad carriage returns.  I found an answer at (http://forums.devshed.com/archive/t-99865), which suggested using sed to fix the carriage returns.  I did that, creating backup2.tar.  The problem is, when I now run the following command (after recreating the db athena):
pg_restore -d athena backup2.tar

   I get this reply _immediately_:
   pg_restore: [archiver] out of memory

    This happens so fast that I'm pretty sure the restore hasn't even *started*.  So, I'm not sure how it has run out of memory.  This is on a P4 2.8 GHz system with a 2GB swap and 1GB RAM.

   I can restore all the other tables individually with backups to sql and imports, but the blob table is the most important table in the db.  I've spent about 8 hours trolling through Google and mailing lists.  Any suggestions as to how I can find out more information on my problem?

Thanks in advance for your help,
Chris McCormick

PS - If I can't get this working myself, I'd be willing to pay someone to figure it out.  What would be a reasonable amount to expect to pay for this kind of troubleshooting?

Chris McCormick <cmccormick@mailsnare.net> writes:
>    It ran for a while (creating all the table defs) and the I got the
> following messages:
>    pg_restore: ERROR:   literal carriage return found in data.

>    So, I had some bad carriage returns.  I found an answer at
> (http://forums.devshed.com/archive/t-99865), which suggested using sed
> to fix the carriage returns.  I did that, creating backup2.tar.

You can't do that to a tar (or custom) format dump file, because you'll
clobber the tar metadata, which is full of raw-binary numbers, not to
mention absolute file lengths that will be wrong after sed hacks the data.

The basic problem here is that COPY's handling of embedded CRs changed
somewhere around 7.2, and the backwards compatibility kluge that was in
place for awhile isn't there anymore in 7.4.

Probably the least painful solution is to repeat the dump, adding the -d
switch (or whichever one it is that selects dump-as-INSERTs instead of
dump-as-COPY).  This will take much longer to restore :-( but a manual
solution that allows COPY will probably take even longer in total.

If you can't repeat the dump because you don't have the original 7.1
installation anymore, you might be forced into a serial upgrade: install
7.2, reload, dump, install 7.4, reload.  I think you can skip 7.3,
or skip 7.2 and use 7.3 as the intermediate step --- IIRC both 7.2 and
7.3 contain the backwards-compatibility kluge for the old COPY data
format.

            regards, tom lane