pg.ddx.io  pgbackrest/pgbackrest GitHub issues and pull requests  
help / color / mirror / Atom feed
From: dwsteele (@dwsteele) <noreply+dwsteele@github.com>
To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>
Subject: Re: [pgbackrest/pgbackrest] PR #2887: Direct I/O for Posix/CIFS storage.
Date: Tue, 17 Nov 2026 17:40:25 +0000
Message-ID: <gh-pgbackrest-pgbackrest-2887-c6955988599@github.com> (raw)
In-Reply-To: <gh-pgbackrest-pgbackrest-2887@github.com>
References: <gh-pgbackrest-pgbackrest-2887@github.com>

@fmbiete 

> I think I'm ok with those changes.

I'm still fine with adding alignment via statx(). I just wanted to keep the initial commit simple. Feel free to submit statx() as a followup if you like.

> The last bit as buffered IO is a tread-off

I really debated this but I think it is the right way to go. My thinking is we can default to use preadv2()/pwritev2() for the last block when it is present and avoid buffering entirely. I have a note to add that support but you are welcome to take that on if you like.

> but I have some doubts about those kubernetes-operators

I'm not sure how well this is going to work on k8s (or containers in general) based on my reading but I'm hoping preadv2()/pwritev2() also gives us a path forward there.

view thread (20+ messages)  latest in thread

Message-ID: <gh-pgbackrest-pgbackrest-2887-c6955988599@github.com>
Permalink:  ../gh-pgbackrest-pgbackrest-2887-c6955988599@github.com/

 · 


This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox