Обсуждение: Improving pg_dump performance when handling large numbers of LOBs

On Mon, Feb 5, 2024 at 2:01 PM Wyatt Tellis <wyatt.tellis@gmail.com> wrote:

Hi,

We've inherited a series of legacy PG 12 clusters that each contain a database that we need to migrate to a PG 15 cluster. Each database contains about 150 million large objects totaling about 250GB.

250*10^9 / (150*10^6) = 1667 bytes.  That's tiny.

Am I misunderstanding you?

I think it less about the size of the individual objects then the number (150 million) of them.

AFAIK that can't be handled by COPY, therefore they have to be transferred individually.

-- 
Adrian Klaver
adrian.klaver@aklaver.com

On Mon, Feb 05, 2024 at 09:17:53PM +0100, Andreas Joseph Krogh wrote:
> 
> På mandag 05. februar 2024 kl. 21:10:53, skrev Wyatt Tellis <
> wyatt.tellis@gmail.com <mailto:wyatt.tellis@gmail.com>>:
> 
> No, we don't have the ability to make schema changes and the schema in the 
> PG15 copy needs to match what's in the PG 12 versions
> 
> Well then, I guess it boils down to how badly the ones in charge wants this 
> migration…
> 
> --
> Andreas Joseph Krogh
> CTO / Partner - Visena AS
> Mobile: +47 909 56 963
> andreas@visena.com <mailto:andreas@visena.com>
> www.visena.com <https://www.visena.com>
>  <https://www.visena.com>

Hi,

Can you use logical replication? Then you aren't under the gun to make
a fast copy.

Regards,
Ken

Wyatt Tellis <wyatt.tellis@gmail.com> writes:
> Yes, the LOBs themselves are tiny, but there are a lot of them (~150
> million) which seem to be slowing down pg_dump.  Note, we did not
> design/build this system and agree that use of LOBs for this purpose was
> not necessary.

I don't know of anything much you can do at the user level.  But there
is a patchset under development to improve pg_dump's behavior with
tons of blobs:

https://www.postgresql.org/message-id/842242.1706287466@sss.pgh.pa.us

That's intended for v17, and I'm not sure how tough it might be to
adapt to v15 pg_dump, but maybe you could make that happen.  In any
case, more eyeballs reviewing that patchset would be welcome.

            regards, tom lane

On 2/5/24 12:32, Ken Marshall wrote:
> On Mon, Feb 05, 2024 at 09:17:53PM +0100, Andreas Joseph Krogh wrote:
>>
>> På mandag 05. februar 2024 kl. 21:10:53, skrev Wyatt Tellis <
>> wyatt.tellis@gmail.com <mailto:wyatt.tellis@gmail.com>>:
>>
>> No, we don't have the ability to make schema changes and the schema in the
>> PG15 copy needs to match what's in the PG 12 versions
>>
>> Well then, I guess it boils down to how badly the ones in charge wants this
>> migration…
>>
>> --
>> Andreas Joseph Krogh
>> CTO / Partner - Visena AS
>> Mobile: +47 909 56 963
>> andreas@visena.com <mailto:andreas@visena.com>
>> www.visena.com <https://www.visena.com>
>>   <https://www.visena.com>
> 
> Hi,
> 
> Can you use logical replication? Then you aren't under the gun to make
> a fast copy.

https://www.postgresql.org/docs/current/logical-replication-restrictions.html

"Large objects (see Chapter 35) are not replicated. There is no 
workaround for that, other than storing data in normal tables."


> 
> Regards,
> Ken
> 
> 

-- 
Adrian Klaver
adrian.klaver@aklaver.com