Re: Cyrus HA Scalable Solution? Rsync

2004-06-09 Thread Kevin Baker
Thanks for you reply Jason.

Have you had any instances where you have needed the
failover?

Are you using IP takeover to the synced mailstore or just
using the rsync as a backup solution?

Rsync makes sense to me, but when I posted a few people
seems to say DRBD  was a better way to go. I think I'm
going to look into this too.

Thanks again.


-Kevin


 On 4:39:45 pm 05/25/04 Kevin Baker
 [EMAIL PROTECTED] wrote:
 We are testing a number of email configurations for a
 10,000+ user-base. Was hoping to get some thoughts on
 below:

 - Postfix
 - Cyrus-SASL
 - Mysql Auth

 We will likely start with 3 frontend servers and 3
 backend
 servers. Replicate MySQL across all servers auth,
 maildrop
 routing.

 We were thinking of doing some sort of rysync of the
 imap
 mailstore across the backend servers. Then Heartbeat on
 the backend servers with IP takeover to handle failover.

 The hope is that if a server goes down the mailstore
 will
 be sync'ed up on the server that takes over.

 Thought? This is obviously just a sketch... but I
 haven't
 seen a this done before as far as the failover solution
 with rsync and thought it might work pretty well.

 I have been doing this with an 18 gig mailstore that uses
 maildirs. After
 the first sync I can run rsync every 5 minutes and it only
 takes 35-40
 seconds to complete, even if I move/delete a few thousand
 messages at a
 time.

 \__ Jason Munro
  \__ [EMAIL PROTECTED]
   \__ http://hastymail.sourceforge.net/





---
Cyrus Home Page: http://asg.web.cmu.edu/cyrus
Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html


Re: Cyrus HA Scalable Solution? Rsync

2004-06-09 Thread Jason Munro
On 11:50:02 am 06/09/04 Kevin Baker [EMAIL PROTECTED] wrote:
 Thanks for you reply Jason.

 Have you had any instances where you have needed the
 failover?

Not yet :)

 Are you using IP takeover to the synced mailstore or just
 using the rsync as a backup solution?

Just as a backup for the time being. We will automate the take-over process
as much as possible, once I get to that point on my todo list :)

 Rsync makes sense to me, but when I posted a few people
 seems to say DRBD  was a better way to go. I think I'm
 going to look into this too.

 Thanks again.


\__ Jason Munro
 \__ [EMAIL PROTECTED]
  \__ http://hastymail.sourceforge.net/


---
Cyrus Home Page: http://asg.web.cmu.edu/cyrus
Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html


Re: Cyrus HA Scalable Solution? Rsync

2004-05-26 Thread Jason Munro
On 4:39:45 pm 05/25/04 Kevin Baker [EMAIL PROTECTED] wrote:
 We are testing a number of email configurations for a
 10,000+ user-base. Was hoping to get some thoughts on
 below:

 - Postfix
 - Cyrus-SASL
 - Mysql Auth

 We will likely start with 3 frontend servers and 3 backend
 servers. Replicate MySQL across all servers auth, maildrop
 routing.

 We were thinking of doing some sort of rysync of the imap
 mailstore across the backend servers. Then Heartbeat on
 the backend servers with IP takeover to handle failover.

 The hope is that if a server goes down the mailstore will
 be sync'ed up on the server that takes over.

 Thought? This is obviously just a sketch... but I haven't
 seen a this done before as far as the failover solution
 with rsync and thought it might work pretty well.

I have been doing this with an 18 gig mailstore that uses maildirs. After
the first sync I can run rsync every 5 minutes and it only takes 35-40
seconds to complete, even if I move/delete a few thousand messages at a
time.

\__ Jason Munro
 \__ [EMAIL PROTECTED]
  \__ http://hastymail.sourceforge.net/


---
Cyrus Home Page: http://asg.web.cmu.edu/cyrus
Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html


Re: Cyrus HA Scalable Solution? Rsync

2004-05-25 Thread Michael Loftis

--On Tuesday, May 25, 2004 14:39 -0700 Kevin Baker [EMAIL PROTECTED] 
wrote:


Thought? This is obviously just a sketch... but I haven't
seen a this done before as far as the failover solution
with rsync and thought it might work pretty well.
rsync sucks for large numbers of files/directories.  It has to build an 
in-memory tree before it even starts syncing.

what would be 'nice' to see is something built inside of cyrus to handle 
multiple backends but that's a pretty complicated bit of beast.  (no i'm 
not volunteering ;) )

--
GPG/PGP -- 0xE736BD7E 5144 6A2D 977A 6651 DFBE 1462 E351 88B9 E736 BD7E 
---
Cyrus Home Page: http://asg.web.cmu.edu/cyrus
Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html


Re: Cyrus HA Scalable Solution? Rsync

2004-05-25 Thread Kevin Baker
So I'm guessing I should look into the DRBD then.

http://www.drbd.org)

It's in the archive:
http://www.mail-archive.com/[EMAIL PROTECTED]/msg18820.html

Other than that I can't think of how to handle the fail-over.







 --On Tuesday, May 25, 2004 14:39 -0700 Kevin Baker
 [EMAIL PROTECTED]
 wrote:


 Thought? This is obviously just a sketch... but I
 haven't
 seen a this done before as far as the failover solution
 with rsync and thought it might work pretty well.

 rsync sucks for large numbers of files/directories.  It
 has to build an
 in-memory tree before it even starts syncing.

 what would be 'nice' to see is something built inside of
 cyrus to handle
 multiple backends but that's a pretty complicated bit of
 beast.  (no i'm
 not volunteering ;) )


 --
 GPG/PGP -- 0xE736BD7E 5144 6A2D 977A 6651 DFBE 1462 E351
 88B9 E736 BD7E
 ---
 Cyrus Home Page: http://asg.web.cmu.edu/cyrus
 Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
 List Archives/Info:
 http://asg.web.cmu.edu/cyrus/mailing-list.html



---
Cyrus Home Page: http://asg.web.cmu.edu/cyrus
Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html


Re: Cyrus HA Scalable Solution? Rsync

2004-05-25 Thread Andrew Morgan

On Tue, 25 May 2004, Michael Loftis wrote:



 --On Tuesday, May 25, 2004 14:39 -0700 Kevin Baker [EMAIL PROTECTED]
 wrote:


  Thought? This is obviously just a sketch... but I haven't
  seen a this done before as far as the failover solution
  with rsync and thought it might work pretty well.

 rsync sucks for large numbers of files/directories.  It has to build an
 in-memory tree before it even starts syncing.

 what would be 'nice' to see is something built inside of cyrus to handle
 multiple backends but that's a pretty complicated bit of beast.  (no i'm
 not volunteering ;) )

There has been some discussion lately about software solutions to provide
redundancy, but what about buying redundant hardware?  This may be cheaper
and more reliable in the long run anyways.

It is not hard to find fully redandunt disk arrays (usually in the context
of SANs, but a full SAN environment is not required).  A few examples I've
come across: Sun T3 Enterprise Pairs, Dell/EMC Cx300/500/700.

That pushes the point of failure out to the server (or backend server in a
Murder configuration).  If a server fails for some reason, you could have
a backup server available (also in the SAN, or manually connected at the
time of failure) that mounts the Cyrus partition and carries on.

An additional benefit is that these higher end disk arrays also typically
have much better performance.

With all of that said, I'm currently running 35,000 mailboxes on a single
Dell 2650 with an external SCSI disk array configured as RAID 0+1
(stripe/mirror).  Mail relaying is handled by separate servers, so all
this box does is IMAP and LMTP delivery.

We use Sun's T3 Enterprise Pairs for user home directories and have been
very happy with the performance and reliability (in conjunction with
Veritas Volume Manager and Veritas File System).  However, the Dell/EMC
solutions are much cheaper and appear to offer the same levels of
reliability.

Andy

---
Cyrus Home Page: http://asg.web.cmu.edu/cyrus
Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html


Re: Cyrus HA Scalable Solution? Rsync

2004-05-25 Thread Lee
We (my company) uses DRBD (http://drbd.cubit.at/) with heartbeat and cyrus quite
successfully. To distribute load we use multiple heartbeat/drbd backend
clusters. Each cluster is comprised of 2 machines connected together via
gigabit ethernet cards and serial links. Postfix references ldap (for you
mysql) to determine which backend cluster the user's mailbox resides on.
Perdition or Cyrus Murder can be used to proxy the user logging in to check
mail to the correct backend machine. This solution provides unlimited
scalability and pretty good redunancy.

DRBD is a good innexpensive solution. Its proved to be fast and pretty reliable.
I would recommend it if you are on a budget. If you have unlimited cash, a
kimberlite / SAN cluster might be another good option (havent tried it, but
have heard good things). 

Lee

Quoting Michael Loftis [EMAIL PROTECTED]:

 
 
 --On Tuesday, May 25, 2004 14:39 -0700 Kevin Baker [EMAIL PROTECTED] 
 wrote:
 
 
  Thought? This is obviously just a sketch... but I haven't
  seen a this done before as far as the failover solution
  with rsync and thought it might work pretty well.
 
 rsync sucks for large numbers of files/directories.  It has to build an 
 in-memory tree before it even starts syncing.
 
 what would be 'nice' to see is something built inside of cyrus to handle 
 multiple backends but that's a pretty complicated bit of beast.  (no i'm 
 not volunteering ;) )
 
 
 --
 GPG/PGP -- 0xE736BD7E 5144 6A2D 977A 6651 DFBE 1462 E351 88B9 E736 BD7E 
 ---
 Cyrus Home Page: http://asg.web.cmu.edu/cyrus
 Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
 List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
 


---
Cyrus Home Page: http://asg.web.cmu.edu/cyrus
Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html


Re: Cyrus HA Scalable Solution? Rsync

2004-05-25 Thread Kevin Baker
A number of servers with network storage would be ideal.

Unfortunately the initial rollout will require the use of
cheap managed hosting boxes. So we'll be restricted to
local drives.

I guess the hope is to develop a sort of *standard* way to
handle a highly scalable solution with failover with these
inexpensive racks.

The only real missing piece is the replication to a
failover server.

There seem to be a number of solutions at:
http://linas.org/linux

The question is of course what will work best.


Kevin







 On Tue, 25 May 2004, Michael Loftis wrote:



 --On Tuesday, May 25, 2004 14:39 -0700 Kevin Baker
 [EMAIL PROTECTED]
 wrote:


  Thought? This is obviously just a sketch... but I
 haven't
  seen a this done before as far as the failover
 solution
  with rsync and thought it might work pretty well.

 rsync sucks for large numbers of files/directories.  It
 has to build an
 in-memory tree before it even starts syncing.

 what would be 'nice' to see is something built inside of
 cyrus to handle
 multiple backends but that's a pretty complicated bit of
 beast.  (no i'm
 not volunteering ;) )

 There has been some discussion lately about software
 solutions to provide
 redundancy, but what about buying redundant hardware?
 This may be cheaper
 and more reliable in the long run anyways.

 It is not hard to find fully redandunt disk arrays
 (usually in the context
 of SANs, but a full SAN environment is not required).  A
 few examples I've
 come across: Sun T3 Enterprise Pairs, Dell/EMC
 Cx300/500/700.

 That pushes the point of failure out to the server (or
 backend server in a
 Murder configuration).  If a server fails for some reason,
 you could have
 a backup server available (also in the SAN, or manually
 connected at the
 time of failure) that mounts the Cyrus partition and
 carries on.

 An additional benefit is that these higher end disk arrays
 also typically
 have much better performance.

 With all of that said, I'm currently running 35,000
 mailboxes on a single
 Dell 2650 with an external SCSI disk array configured as
 RAID 0+1
 (stripe/mirror).  Mail relaying is handled by separate
 servers, so all
 this box does is IMAP and LMTP delivery.

 We use Sun's T3 Enterprise Pairs for user home directories
 and have been
 very happy with the performance and reliability (in
 conjunction with
 Veritas Volume Manager and Veritas File System).  However,
 the Dell/EMC
 solutions are much cheaper and appear to offer the same
 levels of
 reliability.

   Andy

 ---
 Cyrus Home Page: http://asg.web.cmu.edu/cyrus
 Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
 List Archives/Info:
 http://asg.web.cmu.edu/cyrus/mailing-list.html



---
Cyrus Home Page: http://asg.web.cmu.edu/cyrus
Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html