Date: Fri, 21 Dec 2018 09:42:57 +0100
From: "Jehan-Guillaume (ioguix) de Rorthais" <ioguix@free.fr>
To: Prince Pathria <prince.pathria@goevive.com>
Cc: soumitra bhandary <soumitra.bhandary@hotmail.com>,
 "pgsql-admin@lists.postgresql.org" <pgsql-admin@lists.postgresql.org>
Subject: Re: Multi master implementation
Message-ID: <20181221094257.51210267@firost>
In-Reply-To: 
 <CAON0Dax_j06D1+h1uOj_iv6W-T_T1M0T0GOk-OgzUc=xwpphjA@mail.gmail.com>
References: 
 <CAON0Daw1yYP0fB6rXzGok_UpoHxBRrE2iuth1e0bHm_9_-h_ig@mail.gmail.com>
	<MA1PR0101MB16873C1F9070F9E1BE39A685E0BF0@MA1PR0101MB1687.INDPRD01.PROD.OUTLOOK.COM>
	<CAON0Dax1V8qHwxsUKemunhz3x9Wp4K2siFu7Z2eqWcW3t2u-fA@mail.gmail.com>
	<MA1PR0101MB168752FE78A9FFF8B80A6DBCE0BF0@MA1PR0101MB1687.INDPRD01.PROD.OUTLOOK.COM>
	<CAON0Dax_j06D1+h1uOj_iv6W-T_T1M0T0GOk-OgzUc=xwpphjA@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Fri, 21 Dec 2018 12:43:57 +0530
Prince Pathria <prince.pathria@goevive.com> wrote:

> Thanks Soumitra, Postgres-XC isn't getting any updates since 2015. Don't
> know where it might get fail. I've also considered Postgres-XL which is in
> active development state. But ver. 10 is very new to be used for producti=
on.

I would not deploy Postgres-XC/XL over a WAN anyway!

> Curious to know how others are implementing multi master architecture!!

They are using Bucardo or BDR. Or they just don't.

Before sketching a very complex multi-master architecture, bench easier and
safer solution first. Put all your energy to define the simplest and
effective architecture you can, and bench it.

You might be surprised.

> On Thu, Dec 20, 2018 at 11:40 PM soumitra bhandary <
> soumitra.bhandary@hotmail.com> wrote: =20
>=20
> > Well Prince , I have never worked with multiple master to be working in
> > master slave replication . As you mentioned you have configured auto
> > failover is this using EFM ? If you are using EFM then it won=E2=80=99t=
 support
> > multiple master in the cluster . I have worked with BDR where all nodes=
 are
> > worked as Master node but as you know it has downside as well .
> >
> > Coming to your application operation , you can segregate all read reque=
st
> > to be diverted to the slave nodes , may be to one synchronous node and =
all
> > write operation to be diverted only in Master node. In tis manner your
> > application should be able to share read and write load . Anyways that =
you
> > would be the best person to take  a call on this depending on your
> > application .
> >
> > For multi master replication you can go through below link if that helps
> > you out subjected to the implementation and compatibility with your
> > application process . If any other member in this group has any experie=
nce
> > with multiple master with  multi slave replication (apart from BDR)
> > configuration experience , please share the same .
> >
> > https://wiki.postgresql.org/wiki/Postgres-XC
> >
> > Thanks
> > Soumitra
> >
> >
> >
> >
> > On 20-Dec-2018, at 12:21 PM, Prince Pathria <prince.pathria@goevive.com>
> > wrote:
> >
> > Hi Soumitra,
> >
> > Sorry, please consider it as another site instead of DR.
> >
> > Or if I want multiple masters in main site only i.e. [2masters+3slaves].
> > How should I implement that?
> >
> > Our devs are planning to push applications that might have a lot of wri=
te
> > operations. If I say read to write ratio, [65% reads]:[35% writes]. I do
> > have automatic failover configured[in single master arch] but don't want
> > that postgres crashes due to unable to handle writes since writes conta=
ins
> > complex queries and blobs as well which makes them heavy. App code chan=
ge
> > isn't an option for me.
> >
> > Need your suggestions please!
> >
> > Happy to help :)
> > Prince Pathria Systems Architect Intern Evive +91 9478670472 goevive.com
> >
> >
> > On Thu, Dec 20, 2018 at 7:30 AM soumitra bhandary < =20
> > soumitra.bhandary@hotmail.com> wrote: =20
> > =20
> >> Hi Prince,
> >>
> >> BDR is something which is not that much robust as conventional master
> >> slave replication.
> >>
> >> Why do you want master node in DR set up ?
> >> Do you want to enable simultaneous write operations at DR site as well=
??
> >> Normally DR setups remain in cascade replication from one of the stand=
 by
> >> node at production side .
> >>
> >> Sent from my iPhone
> >>
> >> On 19-Dec-2018, at 11:55 PM, Prince Pathria <prince.pathria@goevive.co=
m>
> >> wrote:
> >>
> >> Hi,
> >>
> >> I want to how people are implementing multi master architecture in
> >> Postgres. For now, I've come across BDR3 and bucardo for multimaster
> >> approach.
> >>
> >> Though both doesn't seem to be much promising to me.
> >>
> >> Goal: 2 master and their 3 slaves. [1 master + 2 slaves] in production
> >> site and [1 master + 1 slave]
> >>  in DR site.
> >>
> >> Can you guys share how you did it and what issues did you faced?
> >>
> >> Thanks in advance!
> >>
> >> Happy to help :)
> >> Prince Pathria Systems Architect Intern Evive +91 9478670472 goevive.c=
om
> >>
> >> =20
> > =20


--=20
Jehan-Guillaume de Rorthais
Dalibo