RE: step-by-step replication on MySQL 5.1.x

There are tradeoffs.

Writing cross-continent to get to the single master does take time. But not necessarily "too much" time. Having a single-source-of-truth is beneficial. I see 70ms between California and DC; 250ms to Beijing.

Some applications do a lot of processing on the master; we ship the data to a processing machine near the Master and let it do the work. This avoids most of the long hops.

I would try to avoid a two-tier setup of slaves -- if the relay goes down, it is complicated to repoint its slaves to a replacement relay.

Possibly another approach... Each continent has a master, its slaves, plus a slave in every other continent. Unfortunately, these are separate systems and would need separate instances of Mysql (diff port or diff machine).

> -----Original Message-----
> From: Raymond Rodgers [mailto:raymondr@twacomm.net]
> Sent: Wednesday, September 05, 2007 1:32 PM
> To: Rick James
> Cc: aslan@digirati.com.br; replication@lists.mysql.com
> Subject: RE: step-by-step replication on MySQL 5.1.x
>
> I'm not sure if this is Aslan's situation, and it's not mine
> either, but
> a hypothetical.
>
> What if an organization has offices on four separate continents, and
> needs to have a local master server on each continent, but
> also needs to
> be able to retrieve and update data from each of the other continents.
> Wouldn't a multi-master ring configuration be more
> appropriate in terms
> of access times and latency? Or would you suggest still as
> single master
> server to which all writes are directed, and local slaves on each
> continent (which could then server further slaves)?
>
> Just curious,
> Raymond
> On Wed, 2007-09-05 at 13:10 -0700, Rick James wrote:
> > Master-slave is no slower. And the failure cases are arguably more
> > manageable.
> >
> > > -----Original Message-----
> > > From: Aslan Carlos [mailto:aslan@digirati.com.br]
> > > Sent: Wednesday, September 05, 2007 11:21 AM
> > > To: Rick James
> > > Cc: replication@lists.mysql.com
> > > Subject: RE: step-by-step replication on MySQL 5.1.x
> > >
> > > Ring because I need multi-master and each one need has
> all information
> > > on the ring quickly, each server of ring is a master server on one
> > > farm.
> > >
> > > About the problem with field auto_increment, we'll use key with
> > > two-parts, (id , server_id ) as key.
> > >
> > > I'll try explain one problem that I find using mysql
> replication in
> > > ring.
> > >
> > > Multi-master ( 4 servers ) replication ring working
> perfectly, we are
> > > using a procedures each master to change the MASTER if they dies.
> > > If one master was removed because network problems, and
> this server
> > > removed receives some changes and the data in ring receives some
> > > changes. When the network's connection is fixed, this
> server removed
> > > cannot back to the ring because don't have the same data
> in masters on
> > > ring... understand ? I'll have inconsistency of data. :s
> > >
> > > Ring Work perfectly
> > > .___.
> > > | |
> > > |___|
> > > ' '
> > >
> > > One master dies, and your slave change to another master,
> this master
> > > removed receives one change in your local systems.. starts the
> > > problem . :(
> > > .__.
> > > | /
> > > |/
> > > ' '
> > >
> > >
> > > I'm looking for a solution in replication to do all data is
> > > readable quickly in all masters with multi-masters writable.
> > > They'll use fast connections
> > > like between 10Mbit/s to 150Mbit/s, with Link SLA of 99.97%.
> > >
> > > Thanks in advance.
> > > Aslan Carlos
> > >
> > >
> > >
> > >
> > > On Wed, 2007-09-05 at 09:36 -0700, Rick James wrote:
> > > > Why a ring? Several possible answers and problems...
> > > >
> > > > Three masters in a ring -- If one dies and cannot be
> > > restarted, you have a
> > > > replication mess that is hard to unravel.
> > > >
> > > > Multi-master (2 or 3) does NOT buy any additional
> > > performance in inserts --
> > > > all inserts have to go to all machines eventually.
> > > >
> > > > Multi-master has procedural hassles with auto_increment
> > > (use offset; don't
> > > > depend on monotonicity; etc) and other unique keys.
> > > >
> > > > For scaling readonly access, use plain slaves.
> > > >
> > > > Dual-master, with only one writable at a time, provides a
> > > high degree of
> > > > failover / BCP / recovery / etc. But you need slaves also.
> > > The machines
> > > > should be in geographically distinct places (don't want a
> > > flood/earthquake
> > > > to take out your whole system).
> > > >
> > > > > -----Original Message-----
> > > > > From: Aslan Carlos [mailto:aslan@digirati.com.br]
> > > > > Sent: Wednesday, September 05, 2007 9:13 AM
> > > > > To: Rick James
> > > > > Cc: replication@lists.mysql.com
> > > > > Subject: RE: step-by-step replication on MySQL 5.1.x
> > > > >
> > > > > I don't know, I want know this.
> > > > >
> > > > > I'm making a new project. I'll build a configuration of
> > > replication in
> > > > > circle using MySQL 5.1.x, ( called too as replication ring ).
> > > > > There many
> > > > > problems to do this configuration.
> > > > >
> > > > >
> > > > > Anyone here are using a replication ring or used one day?
> > > Did have any
> > > > > problem using this configuration?
> > > > >
> > > > >
> > > > > best regard,
> > > > > Aslan Carlos
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > On Wed, 2007-09-05 at 08:59 -0700, Rick James wrote:
> > > > > > Are there any differences? Maybe having to do with
> > > > > row-based replication,
> > > > > > but that is optional.
> > > > > >
> > > > > > > -----Original Message-----
> > > > > > > From: Aslan Carlos [mailto:aslan@digirati.com.br]
> > > > > > > Sent: Monday, September 03, 2007 1:19 PM
> > > > > > > To: replication@lists.mysql.com
> > > > > > > Subject: step-by-step replication on MySQL 5.1.x
> > > > > > >
> > > > > > > Hi folks,
> > > > > > >
> > > > > > > I'm looking for papers that describe step-by-step the
> > > > > replication
> > > > > > > process in the MySQL version 5.1.x, like was
> described in the
> > > > > > > book 'High
> > > > > > > performace MySQL', but there just talks the 3.23
> and 4.0.x.
> > > > > > >
> > > > > > >
> > > > > > > thanks in advance.
> > > > > > > --
> > > > > > > Aslan Carlos de M. Ramos
> > > > > > > Senior Network Administrator
> > > > > > > K8 Networks, Digirati Networks , Hostnet Web hosting
> > > > > > > Phone : +55(21)2233-5950 tie-line:29
> > > > > > > Mobile : +55(21)9989-4932
> > > > > > > Skype : aslancarlos
> > > > > > >
> > > > > > >
> > > > > >
> > > > >
> > > >
> > >
> >
> >
>
>

-- 
MySQL Replication Mailing List
For list archives: 
http://lists.mysql.com/replication
To unsubscribe:    
http://lists.mysql.com/replication?unsub=lists@pantek.com