Mailing List Archive

> We have some mysql servers that we would like to function as a=20
> cluster. The
> mysql servers house a database that a cluster of webservers will =
access to
> display product information, accept orders, etc.
>=20
> >From the reading I have done on mysql, it would appear that =
(according to
> the mysql website) they don't really support "clustering". They=20
> don't REALLY
> seem to support replication either, other than there is some=20
> notes about how
> to replicate one way via moving the transaction log around.

Database clustering is a whole new ballgame, really. We're currently =
putting together a system that will use the basic MySQL replication to =
keep a pair of servers syncronised, and read from either using =
Wackamole, which is about as good as it gets currently.

I believe Ben Laurie is working on a product called Splash to address =
some of this, but it is in the early stages of development, AIUI.

Regards,

Geoff Campbell
Internetworking Ltd.
+44(0)1267-253747

Here's how I do it:

MySQL replication really does work, especially in the newer 3.23 releases.
"Clustering" works via daisy-chained replication.

I have two MySQL servers with virtual IP addresses. (I could manage these
two addresses with wackamole, but at the moment I'm using a hardware load
balancer for that). Each is configured as both master and slave to the
other, and each is configured with the "log-slave-updates" directive,
which allows daisy chaining. I have seen larger clusters in a lab, though
I decided for only two in production. Mine handle about 50 selects per
sec. and one insert or update every two seconds.

If you don't have any tables with auto_increment keys, you can direct SQL
inserts and updates to either server; otherwise it's best to direct all of
the updates to a single host.

John Duncan wrote....
> "Clustering" works via daisy-chained replication.
That is great info, THANKS!

[... snip ...]

> If you don't have any tables with auto_increment keys, you can direct SQL
> inserts and updates to either server; otherwise it's best to direct all of
> the updates to a single host.
Unfortunately, the database programmers told me that use auto_increment keys
all over the place :\ So, can you possibly expand on "best to direct all of
the updates to a single host" a little? If we target all the updates to one
wackamole IP address, that would be problematic from what you say I think.
Any further elucidation you can provide would be most appreciated.

Jay West

---
[This E-mail scanned for viruses by Declude Virus]

Geoff Campbell wrote:
>> We have some mysql servers that we would like to function as a
>> cluster. The mysql servers house a database that a cluster of
>> webservers will access to display product information, accept
>> orders, etc.
>>
>>> From the reading I have done on mysql, it would appear that
>>> (according to
>> the mysql website) they don't really support "clustering". They
>> don't REALLY seem to support replication either, other than there
>> is some notes about how to replicate one way via moving the
>> transaction log around.
>
>
> Database clustering is a whole new ballgame, really. We're currently
> putting together a system that will use the basic MySQL replication
> to keep a pair of servers syncronised, and read from either using
> Wackamole, which is about as good as it gets currently.
>
> I believe Ben Laurie is working on a product called Splash to address
> some of this, but it is in the early stages of development, AIUI.

Splash works. But it isn't designed to do heavyweight SQL clustering,
just key/value clustering.

Cheers,

Ben.

--
http://www.apache-ssl.org/ben.html http://www.thebunker.net/

Available for contract work.

"There is no limit to what a man can do or how far he can go if he
doesn't mind who gets the credit." - Robert Woodruff

John David Duncan wrote:

>Here's how I do it:
>
>MySQL replication really does work, especially in the newer 3.23 releases.
>"Clustering" works via daisy-chained replication.
>
I would say MySQL replication "works as documented"... and you better
read the documents before you assume someone else's definition of
replication is the same as yours or mine. IMNSHO MySQL doesn't
replicate at all... Rather it does an excellent job of distributing
transactions.

One quick example. If I insert a row on machine A and machine B runs
out of space when trying to apply that insert, it will not be inserted
on B and will be on A and you have two accessible DBs without a
consistent view. The transaction never should have taken place unless
it could be committed on all machines in the "replication pool". MySQL
provides nothing approaching this AFAIK.

MySQL allows applying ones transaction log to another database and
cascading these into a daisy-chain -- where atomicity, consistency and
integrity are maintained in all this is beyond me :-)

>If you don't have any tables with auto_increment keys, you can direct SQL
>inserts and updates to either server; otherwise it's best to direct all of
>the updates to a single host.
>
>
My main point is that wackamole make no claims or efforts to provide
application level replication. It will never be the "mission" of
wackamole. Spread, on the other hand, is an excellent subsystem for
building such a technology. Spread could be used to implement ACID
compliant replicated databases.

This discussion belongs on the MySQL or Spread lists (Spread-users IMHO
so Ihave CCed it).

If you are in desperate need of such a replication technology, I would
suggest talking to the guys are Spread Concepts, LLC. Shoot Yair an email.

--
Theo Schlossnagle
1024D/82844984/95FD 30F1 489E 4613 F22E 491A 7E88 364C 8284 4984
2047R/33131B65/71 F7 95 64 49 76 5D BA 3D 90 B9 9F BE 27 24 E7

Theo wrote...
> My main point is that wackamole make no claims or efforts to provide
> application level replication. It will never be the "mission" of
> wackamole. Spread, on the other hand, is an excellent subsystem for
> building such a technology. Spread could be used to implement ACID
> compliant replicated databases.
>
> This discussion belongs on the MySQL or Spread lists (Spread-users IMHO
> so Ihave CCed it).
I totally agree. I knew this was more of a mysql question, but figured there
might be people using wackamole in some fashion with mysql.

> If you are in desperate need of such a replication technology, I would
> suggest talking to the guys are Spread Concepts, LLC. Shoot Yair an
email.
I would love to see such a thing, as I'm sure many mysql users. I'll fire an
email off to Yair. I'm kinda suprised this hasn't been kicked around
already.

Thanks folks!

Jay West

---
[This E-mail scanned for viruses by Declude Virus]