Horton

Re: DSA failed to allocate memory

Dongming Liu <ldming101@gmail.com>

View thread View on postgr.es
From: Dongming Liu <ldming101@gmail.com>
To: Robert Haas <robertmhaas@gmail.com>
Cc: pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2022-03-18T07:30:49Z
Lists: pgsql-hackers

Attachments

>
> So it's OK for a segment to be in a bin that suggests that it has more
> consecutive free pages than it really does. But it's NOT ok for a
> segment to be in a bin that suggests it has fewer consecutive pages
> than it really does. If dsa_free() is putting things back into the
> wrong place, that's what we need to fix.


I'm trying to move segments into appropriate bins in dsa_free().
In 0001-Re-bin-segment-when-dsa-memory-is-freed.patch, I extract
the re-bin segment logic into a separate function called rebin_segment,
call it to move the segment to the appropriate bin when dsa memory is
freed. Otherwise, when allocating memory, due to the segment with
enough contiguous pages is in a smaller bin, a suitable segment
may not be found to allocate memory.

Fot test, I port the test_dsa patch from [1] and add an OOM case to
test memory allocation until OOM, free and then allocation, compare
the number of allocated memory before and after.

Any thoughts?

[1]
https://www.postgresql.org/message-id/CAEepm%3D3U7%2BRo7%3DECeQuAZoeFXs8iDVX56NXGCV7z3%3D%2BH%2BWd0Sw%40mail.gmail.com

Commits

  1. Re-bin segment when memory pages are freed.

  2. De-support floating-point timestamps.