>> Sounds like a half-measure to me. If we really want to go down this
>> rabbit hole IMO SLRU should be moved to shared buffers as proposed
>> elsewhere [1].
> Thread that I cited stopped in 2018 for this exact reason. 5 years ago. Is this argument still valid?
Meanwhile checksums of buffer pages also reside on a page :)
I would love to have seen more progress on the set of threads that proposed
the page header and integration of SLRU into buffer cache. The changes were
large, and unfortunately as a result, it didn't get the detailed review
that it needed. The complex nature of the feature allowed for more branches
to be split from the main thread with alternative approaches. Athough this is
great to see, it did result in the set of core requirements around LSN and
checksum tracking via page headers to not get into PG 16.
What is being proposed now is the simple and core functionality of introducing
page headers to SLRU pages while continuing to be in the SLRU cache. This
allows the whole project to be iterative and reviewers to better reason about
the smaller set of changes being introduced into the codebase.
Once the set of on-disk changes are in, we can follow up on optimizations.
It may be moving to buffer cache or reviewing Dilip's approach in [1], we
will have the option to be flexible in our approach.
[1] https://www.postgresql.org/message-id/flat/CAFiTN-vzDvNz=ExGXz6gdyjtzGixKSqs0mKHMmaQ8sOSEFZ33A@mail.gmail.com
Shawn
