Hello Tom,
from your mail from 25.10.2021:
>0005 implements your suggestion of accounting for TOAST data while
>scheduling parallel dumps. I realized while looking at that that
>there's a pre-existing bug, which this'd exacerbate: on machines
>with 32-bit off_t, dataLength can overflow. Admittedly such machines
>are just about extinct in the wild, but we do still trouble to support
>the case. So 0005 also includes code to check for overflow and clamp
>the result to INT_MAX blocks.
>Maybe we should back-patch 0005. OTOH, how likely is it that anyone
>is wrangling tables exceeding 16TB on a machine with 32-bit off_t?
>Or that poor parallel dump scheduling would be a real problem in
>such a case?
I noticed that you patched master with all the improvements in pg_dump.
Did you change your mind about backpatching patch 0005 to fix the toast size matter?
It would be rather helpfull for handling existent user data in active branches.
On the matter of 32bit versions I think they are used only in much more little instances.
BTW the 32 bit build of postgres on Windows does not work any more with more modern tool sets (tested with VS2019 and
VS2022)albeit not excluded explicitly in the docs. But no one complained yet (for a long time now...).
Thanks
Hans Buschmann
