psql and COPY BINARY

Andreas Pflug <pgadmin@pse-consulting.de> writes:

Examining why psql won't do sensible stuff with COPY BINARY, I realized
that psql still uses PQgetline, which is marked obsolete since 7.4.
Is this intentional or just a "never reviewed because it works"?

There wasn't any obvious bang for the buck in rewriting it.

regards, tom lane

Tom Lane wrote:

Andreas Pflug <pgadmin@pse-consulting.de> writes:

Examining why psql won't do sensible stuff with COPY BINARY, I realized
that psql still uses PQgetline, which is marked obsolete since 7.4.
Is this intentional or just a "never reviewed because it works"?

There wasn't any obvious bang for the buck in rewriting it.

Well a non-binary copy could take as much as 5 times as much as a binary
copy. I hit this when COPYing 1.5GB of data, getting a 6.6GB file. This
made the 100MBit LAN connection a bottleneck.

I might post a patch to support BINARY in psql some day.

Regards,
Andreas

Andreas Pflug <pgadmin@pse-consulting.de> writes:

Tom Lane wrote:

There wasn't any obvious bang for the buck in rewriting it.

Well a non-binary copy could take as much as 5 times as much as a binary
copy. I hit this when COPYing 1.5GB of data, getting a 6.6GB file. This
made the 100MBit LAN connection a bottleneck.

Or vice versa --- the binary format is *not* necessarily smaller than text.
As an example, an integer column that contains only small values (say 1
or 2 digits) will need 8 bytes as binary and only 2 or 3 as text.

Fixing psql to handle binary copy isn't an unreasonable thing to do,
but I can't get real excited about it either ...

regards, tom lane

Tom Lane wrote:

Andreas Pflug <pgadmin@pse-consulting.de> writes:

Tom Lane wrote:

There wasn't any obvious bang for the buck in rewriting it.

Well a non-binary copy could take as much as 5 times as much as a binary
copy. I hit this when COPYing 1.5GB of data, getting a 6.6GB file. This
made the 100MBit LAN connection a bottleneck.

Or vice versa --- the binary format is *not* necessarily smaller than text.
As an example, an integer column that contains only small values (say 1
or 2 digits) will need 8 bytes as binary and only 2 or 3 as text.

Fixing psql to handle binary copy isn't an unreasonable thing to do,
but I can't get real excited about it either ...

Having a choice can't be too bad.
A COMPRESSED option would be even better, but that's backend stuff (does
TOAST use an algorithm that's platform independent?). Would have reduced
the sample above to about 130MB.

Regards,
Andreas

I submitted a patch a while back to change that but I withdrew it
because I wasn't 100% confident I'd done it right.

Here is the link to it:

http://archives.postgresql.org/pgsql-patches/2005-03/msg00242.php

It's probably 99% there - just a bit of checking.

Chris

Andreas Pflug wrote:

Examining why psql won't do sensible stuff with COPY BINARY, I realized
that psql still uses PQgetline, which is marked obsolete since 7.4.
Is this intentional or just a "never reviewed because it works"?

There wasn't any obvious bang for the buck in rewriting it.

The obvious one (and why I submitted a patch - see previous email) was
for people who use the psql source code as sample code, which is what I
was doing at the time for PHP 5.1's PostgreSQL module.

Chris

Tom Lane wrote:

Andreas Pflug <pgadmin@pse-consulting.de> writes:

Examining why psql won't do sensible stuff with COPY BINARY, I realized
that psql still uses PQgetline, which is marked obsolete since 7.4.
Is this intentional or just a "never reviewed because it works"?

There wasn't any obvious bang for the buck in rewriting it.

We have a TODO:

o Update pg_dump and psql to use the new COPY libpq API
(Christopher)

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073