On 26.10.2020 04:13, Michael Paquier wrote:
On Fri, Oct 23, 2020 at 08:00:08AM +0900, Michael Paquier wrote:
Yeah, we could try to make the logic a bit more complicated like
that. However, for any code path relying on a page read without any
locking insurance, we cannot really have a lot of trust in any of the
fields assigned to the page as this could just be random corruption
garbage, and the only thing I am ready to trust here a checksum
mismatch check, because that's the only field on the page that's
linked to its full contents on the 8k page. This also keeps the code
simpler.
A small update here. I have extracted the refactored part for
PageIsVerified() and committed it as that's independently useful.
This makes the patch proposed here simpler on HEAD, leading to the
attached.
--
Michael
Thank you for committing the first part.
In case you need a second opinion on the remaining patch, it still looks
good to me.
--
Anastasia Lubennikova
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company
