Content-Type: text/plain;
	charset=us-ascii
Mime-Version: 1.0 (Mac OS X Mail 16.0 \(3731.300.101.1.3\))
Subject: Re: Postgres on L2ARC
From: Scott Ribe <scott_ribe@elevated-dev.com>
In-Reply-To: 
 <CAHs7QM-SHvb8kwHXCat4WEP0RP7BzQv9aF-HUDP2qbKx2jB-pw@mail.gmail.com>
Date: Mon, 23 Jan 2023 20:39:45 -0700
Cc: pgsql-admin@lists.postgresql.org
Content-Transfer-Encoding: quoted-printable
Message-Id: <729B87B1-DE36-4876-9488-8F4C19DD3EF8@elevated-dev.com>
References: 
 <CAHs7QM-SHvb8kwHXCat4WEP0RP7BzQv9aF-HUDP2qbKx2jB-pw@mail.gmail.com>
To: Joseph Hammerman <joe.hammerman@datadoghq.com>
Archived-At: 
 <https://www.postgresql.org/message-id/729B87B1-DE36-4876-9488-8F4C19DD3EF8%40elevated-dev.com>
Precedence: bulk

> On Jan 23, 2023, at 6:18 PM, Joseph Hammerman =
<joe.hammerman@datadoghq.com> wrote:
>=20
> Has anyone built a Postgers architecture using L2ARC, and potentially =
TABLESPACES to mount the ZFS pools (dedicated one per table). I am =
imagining the ZFS pools would be backed by EBS volumes.

Yes. It worked fine, but much smaller and less busy than what you're =
probably planning for. Originally, Ubuntu 16 & older ZFS, PG data was on =
mirrorz SSD, with other SSD partitions for L2ARC & ZIL. Later, at a =
major update, I did some benchmarking, and got rid of the SSD dedicated =
to ZFS. Then the SSD became entirely ZIL & L2ARC, and PG shared L2ARC =
with other data.

So, to be clear, in this case the reason for having PG use L2ARC with =
spinning rust behind it was that the PG did not need dedicated SSD for =
performance...=