MIME-Version: 1.0
References: <PN0PR01MB9069922B90966633C8183C5AC5A12@PN0PR01MB9069.INDPRD01.PROD.OUTLOOK.COM>
 <PN0PR01MB90693B1DF528D64769D81B91C5A12@PN0PR01MB9069.INDPRD01.PROD.OUTLOOK.COM>
 <0BD5E273-8E2C-4EDD-AB13-37CC3309E30E@thebuild.com>
In-Reply-To: <0BD5E273-8E2C-4EDD-AB13-37CC3309E30E@thebuild.com>
From: Ron Johnson <ronljohnsonjr@gmail.com>
Date: Mon, 15 Jul 2024 15:55:19 -0400
Message-ID: <CANzqJaBA=iJ+vadW=d1b=12KCM5Aj_RDYHVZ13GDwo2jwdzR+g@mail.gmail.com>
Subject: Re: PostgreSQL Active-Active Clustering
To: "pgsql-general@lists.postgresql.org" <pgsql-general@lists.postgresql.org>
Content-Type: multipart/alternative; boundary="00000000000091a55e061d4e9c77"
Archived-At: <https://www.postgresql.org/message-id/CANzqJaBA%3DiJ%2BvadW%3Dd1b%3D12KCM5Aj_RDYHVZ13GDwo2jwdzR%2Bg%40mail.gmail.com>
Precedence: bulk

--00000000000091a55e061d4e9c77
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Mon, Jul 15, 2024 at 3:28=E2=80=AFPM Christophe Pettus <xof@thebuild.com=
> wrote:

>
> > On Jul 15, 2024, at 12:06, Sarkar, Subhadeep <subhadeepsarkar@kpmg.com>
> wrote:
> >
>
[snip]

> >   =E2=80=A2 In the Community edition of PostgreSQL is it possible to se=
tup  a
> cluster where all the nodes are able to concurrently read-write the
> underlying database image using NATIVE features (i.e. without using any
> extensions or external components or usage of Kubernetes/Dockers).
>
[snip]

> No product, either commercial or open-source, provides the last one
> (read-write shared storage), although there are commercial products that
> provide for a shared-storage model single-writer, multiple-reader model
> (for example, Amazon Aurora).
>

This "lack of products" puzzles me, because DEC was doing this with VAX
(then Alpha and Itanium) clusters 40 years ago via a Distributed Lock
Manager integrated deep into VMS.  Their Rdb and (CODASYL) DBMS products
used those functions extensively.

(In the late 1990s, they sold the DLM code to Oracle, which is where RAC
comes from.)

It was shared-disk, multiple-writer, because the DLM allowed for locking at
the row level.  Thus, a half dozen cluster nodes could hold write locks on
different rows on the same data page.

--00000000000091a55e061d4e9c77
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr"><div dir=3D"ltr">On Mon, Jul 15, 2024 at 3:28=E2=80=AFPM C=
hristophe Pettus &lt;<a href=3D"mailto:xof@thebuild.com">xof@thebuild.com</=
a>&gt; wrote:<br></div><div class=3D"gmail_quote"><blockquote class=3D"gmai=
l_quote" style=3D"margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,20=
4,204);padding-left:1ex"><br>&gt; On Jul 15, 2024, at 12:06, Sarkar, Subhad=
eep &lt;<a href=3D"mailto:subhadeepsarkar@kpmg.com" target=3D"_blank">subha=
deepsarkar@kpmg.com</a>&gt; wrote:<br>
&gt; <br>
</blockquote><div>[snip]=C2=A0</div><blockquote class=3D"gmail_quote" style=
=3D"margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding=
-left:1ex">
&gt;=C2=A0 =C2=A0=E2=80=A2 In the Community edition of PostgreSQL is it pos=
sible to setup=C2=A0 a cluster where all the nodes are able to concurrently=
 read-write the underlying database image using NATIVE features (i.e. witho=
ut using any extensions or external components or usage of Kubernetes/Docke=
rs).<br>
</blockquote><div>[snip]=C2=A0</div><blockquote class=3D"gmail_quote" style=
=3D"margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding=
-left:1ex">
No product, either commercial or open-source, provides the last one (read-w=
rite shared storage), although there are commercial products that provide f=
or a shared-storage model single-writer, multiple-reader model (for example=
, Amazon Aurora).<br></blockquote><div><br></div><div>This &quot;lack of pr=
oducts&quot; puzzles me, because DEC was doing this with VAX (then Alpha an=
d Itanium) clusters 40 years ago via a Distributed Lock Manager integrated =
deep into VMS.=C2=A0 Their Rdb and (CODASYL) DBMS products used those funct=
ions extensively.</div><div><br></div><div>(In the late 1990s, they sold th=
e DLM code to Oracle, which is where RAC comes from.)</div><div><br></div><=
div>It was shared-disk, multiple-writer, because the DLM allowed for lockin=
g at the row level.=C2=A0 Thus, a half dozen cluster nodes could hold write=
 locks on different rows on the same data page.</div><div><br></div></div><=
/div>

--00000000000091a55e061d4e9c77--