Re: Postgresql, PSN hack and table limits

Jasen Betts <jasen(at)xnet(dot)co(dot)nz> writes:
> On 2011-05-01, Mark Morgan Lloyd <markMLl(dot)pgsql-general(at)telemetry(dot)co(dot)uk> wrote:
>> Somebody is making a very specific claim that Postgres can support a
>> limited number of rows:
>>
>> "INPS (a data forensics team) said that there is 7 main Databases all
>> hosted at different data centers but linked over a type of 'cloud' Each
>> database uses PostGRESSQL which would mean the most amount of data each
>> database could hold with no stability issues is aproximitely equal to
>> that of 10,348,439 Rows" http://pastebin.com/MtX1MDdh
>>
>> Does anybody have any idea where they've got hold of this figure?

> the figure is within 1% of the maximun size for data stored in text
> (or bytea) column.

No it isn't; the max size per field is 1GB. Although actually
manipulating such field values will probably not work very well unless
you have a 64-bit machine, else you'll hit address-space issues.

I could believe that a specific application using specific fields in
a specific way in a 32-bit machine might start to hit "out of memory"
errors for field widths somewhere in the tens-of-MB range. But the
stated claim is about number of rows, not row width, and the exactness
and breadth of the claim is, well, ridiculous on its face.

I think INPS's level of knowledge about PG must be about as good as
their ability to spell it :-(

BTW, there *is* a hard limit of 32TB per table, arising from the limited
size of BlockNumber. But it's hard to believe that INPS's claim has
anything to do with that.

regards, tom lane