List Info

Thread: two servers, one




two servers, one
user name
2006-07-15 19:23:28
Hello again,

as some of you might remember, I am struggling with
enhancing the
overall availability of my mail-services.

Now I have rented a second dedicated server (tiny, I know ;)
), and I
want to do this:

Identical setup on both machines.

mysql-db on server1, server2 reads/writes from/to the same
server via a
ssh-tunnel (my experiments with mysql4-replication didn't
make me feel
better ...).

So Maia, policyd, postfix, etc. on both machines work with
the same data.

Given the fact that both policyd-daemons use the same data,
I could set
both machines as MX-records, doing a Backup-MX-setup, which
would mean
some low-level loadbalancing (don't NEED it ...) and, more
important,
increased availability as there would be two machines
accepting mail.

-
For a start I would rsync nearly the whole server1 to
server2 regularly
(with specific include/exclude-lists).
In case of an emergency this would allow me to quickly
continue
operations by switching off the ssh-tunnel and starting the
mysqld on
server2 (which would at first release port 3306 and then use
it by the
local mysqld).

Pretty manual, I know, but at least a defined
emergency-procedure to get
into business again fast. And a way to reduce risk quickly,
after that I
plan to develop things further. I just don't want to have
server1 crash
while I search for the IDEAL solution .... you know.

-

What I don't yet have solved in detail, is the fact that

a) people fetch mail from server1
and
b) all the mail should be stored in the cyrus-imapd-boxes on
server1.

(/var/spool/imap would also be rsynced and could be restored
on server2
quickly).

-

So I want to teach server2 to deliver its scanned mail into
the
cyrus-boxes on server1, or more likely, to hand it over to
the postfix
on server1.

Now I have looked through my setup on the productive server1
and found
the changes made related to Maia, which I had done following

htt
ps://secure.renaissoft.com/maia/wiki/Addresses


So I assume that I will somehow have to setup another smtpd
on server1
and allow server2 to "transport" (postfix) its
mail to it, although this
also means that the aliasing-stuff has to be done at the
right point of
processing.

I am sure that I am close, but I am somewhat overworked
these days and
would really appreciate if someone could give me some
pointers or even
examples.

Thanks a lot, best regards,
Stefan

_______________________________________________
Maia-users mailing list
Maia-usersrenaissoft.com
http://www.renaissoft.com/mailman/listinfo/maia-users
two servers, one
user name
2006-07-16 03:14:47
Stefan G. Weichinger wrote:

>[stuff about having failover servers both hosting full
mail stack deleted]
>
>I am sure that I am close, but I am somewhat overworked
these days and
>would really appreciate if someone could give me some
pointers or even
>examples.
>  
>
This is not exactly what you are asking, but here is what we
did:
Two servers, one "frontend" and one
"backend". The frontend server 
hosted a full maia install with its own postfix, mysql,
apache, etc. 
This server is what accepted incoming port 25 (smtp)
requests. The 
backend server hosted another smtp mta (postfix in your
case) as well as 
the imap server. Incoming port 143 requests (imap) go to
this server. 
Maia on the frontend server is set up to do imap auth
against the 
backend server for user logins. This was quite easy. SMTP
auth on the 
frontend server can be set up to query whatever user db your
imap server 
uses directly or whatever, this is a postfix issue.
When mail comes in it gets accepted and scanned on the
frontend server 
and then passed to the backend server over smtp directed by
the postfix 
transports table. Users read their mail via imap from the
backend server 
and login to the frontend server to train the maia spam
filter if they 
wish. Since both our servers are onsite we also have the
backend server 
set up to accept port 25 traffic from internal machines so
local 
delivery is a bit quicker and does not go through maia.
This is a pretty common config I think and has the following
benefits 
over your proposed config:
If the frontend server goes down for some reason, we can
bypass it by 
redirecting incoming port 25 directly to the backend server.
Reading of 
old mail is not interrupted for this failure instance
either.
If the backend server goes down incoming mail does not
bounce and if 
users need to read mail, in a pinch they can do so from the
ham cache 
via the maia interface.

I understand this doesn't give full redundancy for all
services but 
works well in our case and is how I chose to set things up.
If we need 
to expand out it is much simpler to scale filtering and
backend storage 
separately. It also massively simplifies management and
upgrades.

As for your proposed configuration, one thing you might
consider is 
doing the imap auth for Maia and not moving the bayes db and
awl db into 
mysql. This way the maia mysql db would only contain
statistics and maia 
configuration and you could share out the bayes db and awl
db via nfs 
and you would have far less to worry about for mysql.

Hope this was useful to you (or others) in some way,
Chris
_______________________________________________
Maia-users mailing list
Maia-usersrenaissoft.com
http://www.renaissoft.com/mailman/listinfo/maia-users
two servers, one
user name
2006-07-17 18:47:19
Chris Black schrieb:

> When mail comes in it gets accepted and scanned on the
frontend server
> and then passed to the backend server over smtp
directed by the postfix
> transports table. Users read their mail via imap from
the backend server
> and login to the frontend server to train the maia spam
filter if they
> wish. Since both our servers are onsite we also have
the backend server
> set up to accept port 25 traffic from internal machines
so local
> delivery is a bit quicker and does not go through maia.

Your setup is different because you have different needs.
Your servers are both onsite, while my two servers are
dedicated servers
in some datacenters, and located X kilometers apart.

Also my usage of Maia is different, currently only I and my
fellow admin
use its interface for training and reviewing, setting
individual
filter-settings etc. No customer has been given
Maia-credentials yet
(most of them would simply not fully understand and
therefore not use it
...).

> I understand this doesn't give full redundancy for all
services but
> works well in our case and is how I chose to set things
up. If we need
> to expand out it is much simpler to scale filtering and
backend storage
> separately. It also massively simplifies management and
upgrades.

Ok. Although I don't see that much problems with my setup
also. But this
could be related that I have "designed" it along
my own paths of
understanding 

> As for your proposed configuration, one thing you might
consider is
> doing the imap auth for Maia and not moving the bayes
db and awl db into
> mysql. This way the maia mysql db would only contain
statistics and maia
> configuration and you could share out the bayes db and
awl db via nfs
> and you would have far less to worry about for mysql.

Hm, good thought, I have to consider this. Right now I have
set up
things as I described it, for two of my own domains, and it
works fine
so far.

I only get SA timeouts, even after disabling the auto-expiry
(as
mentioned on the amavisd-site and in some threads in this
list here).

I am using SpamAssassin version 3.1.3 on both of my servers,
maybe
they're too young or something ...

Are there any parameters to maybe increase these
timeout-values?

-

For now I testdrive this with my domains, and evaluate how
to go on.

Thanks, Chris,
greetings to you and the list,

Stefan


_______________________________________________
Maia-users mailing list
Maia-usersrenaissoft.com
http://www.renaissoft.com/mailman/listinfo/maia-users
two servers, one
user name
2006-07-18 13:33:26
How about using LVS (http://www.linuxvir
tualserver.org)?  This way,  
you have a virtual IP that cyrus listens to for IMAP traffic
and  
postfix listens to for final delivery.  Each machine would
still have  
it's own IP that could be listed as separate MX records and
have  
Postfix/Maia listening for incoming mail.  You data would be
rsync'ed  
from Server1 to Server2 at regular intervals.

When both machines are up, incoming mail would be
load-balanced  
between the two, the messages would be scanned and then
handed to the  
virtual IP for final delivery.  Server1 would accept the
mail and  
deliver it the user's mailbox.  Incoming IMAP connections
would also  
get directed to Server1.  If Server1 fails, Server2 would
take over  
the Virtual IP and start accepting IMAP traffic. Postfix
would also  
be reconfigured to listen on the Virtual IP and be
restarted.

The big problem with this setup (or anything that uses
rsync) is file  
consistency.  If a message comes in 2 minutes before Server1
goes  
down, it will not exist on Server2 when the IP switchover
takes  
place.  If Server1's disk is dead, that message is *gone*,
but even  
if the drive is OK, that message is lost. Once Server1 comes
back on- 
line, you'll have to sync it with Server2 and once you do
that,  
everything that was on Server1 will be replaced.  If your
dataset is  
small enough, you could have a very small window between
rsyncs and  
this might not be a huge problem for you.  You could also
use DRBD to  
keep the servers in perfect sync or GFS if you have shared
storage.

-Eric

#########################################################
Eric Pierce, RHCE                  Phone: (813) 974-8868
Academic Computing                 Fax:   (813) 974-1799
University of South Florida        Email: epierceusf.edu



On Jul 15, 2006, at 3:23 PM, Stefan G. Weichinger wrote:

>
> Hello again,
>
> as some of you might remember, I am struggling with
enhancing the
> overall availability of my mail-services.
>
> Now I have rented a second dedicated server (tiny, I
know ;) ), and I
> want to do this:
>
> Identical setup on both machines.
>
> mysql-db on server1, server2 reads/writes from/to the
same server  
> via a
> ssh-tunnel (my experiments with mysql4-replication
didn't make me feel
> better ...).
>
> So Maia, policyd, postfix, etc. on both machines work
with the same  
> data.
>
> Given the fact that both policyd-daemons use the same
data, I could  
> set
> both machines as MX-records, doing a Backup-MX-setup,
which would mean
> some low-level loadbalancing (don't NEED it ...) and,
more important,
> increased availability as there would be two machines
accepting mail.
>
> -
> For a start I would rsync nearly the whole server1 to
server2  
> regularly
> (with specific include/exclude-lists).
> In case of an emergency this would allow me to quickly
continue
> operations by switching off the ssh-tunnel and starting
the mysqld on
> server2 (which would at first release port 3306 and
then use it by the
> local mysqld).
>
> Pretty manual, I know, but at least a defined
emergency-procedure  
> to get
> into business again fast. And a way to reduce risk
quickly, after  
> that I
> plan to develop things further. I just don't want to
have server1  
> crash
> while I search for the IDEAL solution .... you know.
>
> -
>
> What I don't yet have solved in detail, is the fact
that
>
> a) people fetch mail from server1
> and
> b) all the mail should be stored in the
cyrus-imapd-boxes on server1.
>
> (/var/spool/imap would also be rsynced and could be
restored on  
> server2
> quickly).
>
> -
>
> So I want to teach server2 to deliver its scanned mail
into the
> cyrus-boxes on server1, or more likely, to hand it over
to the postfix
> on server1.
>
> Now I have looked through my setup on the productive
server1 and found
> the changes made related to Maia, which I had done
following
>
> htt
ps://secure.renaissoft.com/maia/wiki/Addresses
>
>
> So I assume that I will somehow have to setup another
smtpd on server1
> and allow server2 to "transport" (postfix)
its mail to it, although  
> this
> also means that the aliasing-stuff has to be done at
the right  
> point of
> processing.
>
> I am sure that I am close, but I am somewhat overworked
these days and
> would really appreciate if someone could give me some
pointers or even
> examples.
>
> Thanks a lot, best regards,
> Stefan
>
> _______________________________________________
> Maia-users mailing list
> Maia-usersrenaissoft.com
> http://www.renaissoft.com/mailman/listinfo/maia-users

_______________________________________________
Maia-users mailing list
Maia-usersrenaissoft.com
http://www.renaissoft.com/mailman/listinfo/maia-users
two servers, one
user name
2006-07-18 13:33:26
How about using LVS (http://www.linuxvir
tualserver.org)?  This way,  
you have a virtual IP that cyrus listens to for IMAP traffic
and  
postfix listens to for final delivery.  Each machine would
still have  
it's own IP that could be listed as separate MX records and
have  
Postfix/Maia listening for incoming mail.  You data would be
rsync'ed  
from Server1 to Server2 at regular intervals.

When both machines are up, incoming mail would be
load-balanced  
between the two, the messages would be scanned and then
handed to the  
virtual IP for final delivery.  Server1 would accept the
mail and  
deliver it the user's mailbox.  Incoming IMAP connections
would also  
get directed to Server1.  If Server1 fails, Server2 would
take over  
the Virtual IP and start accepting IMAP traffic. Postfix
would also  
be reconfigured to listen on the Virtual IP and be
restarted.

The big problem with this setup (or anything that uses
rsync) is file  
consistency.  If a message comes in 2 minutes before Server1
goes  
down, it will not exist on Server2 when the IP switchover
takes  
place.  If Server1's disk is dead, that message is *gone*,
but even  
if the drive is OK, that message is lost. Once Server1 comes
back on- 
line, you'll have to sync it with Server2 and once you do
that,  
everything that was on Server1 will be replaced.  If your
dataset is  
small enough, you could have a very small window between
rsyncs and  
this might not be a huge problem for you.  You could also
use DRBD to  
keep the servers in perfect sync or GFS if you have shared
storage.

-Eric

#########################################################
Eric Pierce, RHCE                  Phone: (813) 974-8868
Academic Computing                 Fax:   (813) 974-1799
University of South Florida        Email: epierceusf.edu



On Jul 15, 2006, at 3:23 PM, Stefan G. Weichinger wrote:

>
> Hello again,
>
> as some of you might remember, I am struggling with
enhancing the
> overall availability of my mail-services.
>
> Now I have rented a second dedicated server (tiny, I
know ;) ), and I
> want to do this:
>
> Identical setup on both machines.
>
> mysql-db on server1, server2 reads/writes from/to the
same server  
> via a
> ssh-tunnel (my experiments with mysql4-replication
didn't make me feel
> better ...).
>
> So Maia, policyd, postfix, etc. on both machines work
with the same  
> data.
>
> Given the fact that both policyd-daemons use the same
data, I could  
> set
> both machines as MX-records, doing a Backup-MX-setup,
which would mean
> some low-level loadbalancing (don't NEED it ...) and,
more important,
> increased availability as there would be two machines
accepting mail.
>
> -
> For a start I would rsync nearly the whole server1 to
server2  
> regularly
> (with specific include/exclude-lists).
> In case of an emergency this would allow me to quickly
continue
> operations by switching off the ssh-tunnel and starting
the mysqld on
> server2 (which would at first release port 3306 and
then use it by the
> local mysqld).
>
> Pretty manual, I know, but at least a defined
emergency-procedure  
> to get
> into business again fast. And a way to reduce risk
quickly, after  
> that I
> plan to develop things further. I just don't want to
have server1  
> crash
> while I search for the IDEAL solution .... you know.
>
> -
>
> What I don't yet have solved in detail, is the fact
that
>
> a) people fetch mail from server1
> and
> b) all the mail should be stored in the
cyrus-imapd-boxes on server1.
>
> (/var/spool/imap would also be rsynced and could be
restored on  
> server2
> quickly).
>
> -
>
> So I want to teach server2 to deliver its scanned mail
into the
> cyrus-boxes on server1, or more likely, to hand it over
to the postfix
> on server1.
>
> Now I have looked through my setup on the productive
server1 and found
> the changes made related to Maia, which I had done
following
>
> htt
ps://secure.renaissoft.com/maia/wiki/Addresses
>
>
> So I assume that I will somehow have to setup another
smtpd on server1
> and allow server2 to "transport" (postfix)
its mail to it, although  
> this
> also means that the aliasing-stuff has to be done at
the right  
> point of
> processing.
>
> I am sure that I am close, but I am somewhat overworked
these days and
> would really appreciate if someone could give me some
pointers or even
> examples.
>
> Thanks a lot, best regards,
> Stefan
>
> _______________________________________________
> Maia-users mailing list
> Maia-usersrenaissoft.com
> http://www.renaissoft.com/mailman/listinfo/maia-users

_______________________________________________
Maia-users mailing list
Maia-usersrenaissoft.com
http://www.renaissoft.com/mailman/listinfo/maia-users
two servers, one
user name
2006-07-18 13:33:26
How about using LVS (http://www.linuxvir
tualserver.org)?  This way,  
you have a virtual IP that cyrus listens to for IMAP traffic
and  
postfix listens to for final delivery.  Each machine would
still have  
it's own IP that could be listed as separate MX records and
have  
Postfix/Maia listening for incoming mail.  You data would be
rsync'ed  
from Server1 to Server2 at regular intervals.

When both machines are up, incoming mail would be
load-balanced  
between the two, the messages would be scanned and then
handed to the  
virtual IP for final delivery.  Server1 would accept the
mail and  
deliver it the user's mailbox.  Incoming IMAP connections
would also  
get directed to Server1.  If Server1 fails, Server2 would
take over  
the Virtual IP and start accepting IMAP traffic. Postfix
would also  
be reconfigured to listen on the Virtual IP and be
restarted.

The big problem with this setup (or anything that uses
rsync) is file  
consistency.  If a message comes in 2 minutes before Server1
goes  
down, it will not exist on Server2 when the IP switchover
takes  
place.  If Server1's disk is dead, that message is *gone*,
but even  
if the drive is OK, that message is lost. Once Server1 comes
back on- 
line, you'll have to sync it with Server2 and once you do
that,  
everything that was on Server1 will be replaced.  If your
dataset is  
small enough, you could have a very small window between
rsyncs and  
this might not be a huge problem for you.  You could also
use DRBD to  
keep the servers in perfect sync or GFS if you have shared
storage.

-Eric

#########################################################
Eric Pierce, RHCE                  Phone: (813) 974-8868
Academic Computing                 Fax:   (813) 974-1799
University of South Florida        Email: epierceusf.edu



On Jul 15, 2006, at 3:23 PM, Stefan G. Weichinger wrote:

>
> Hello again,
>
> as some of you might remember, I am struggling with
enhancing the
> overall availability of my mail-services.
>
> Now I have rented a second dedicated server (tiny, I
know ;) ), and I
> want to do this:
>
> Identical setup on both machines.
>
> mysql-db on server1, server2 reads/writes from/to the
same server  
> via a
> ssh-tunnel (my experiments with mysql4-replication
didn't make me feel
> better ...).
>
> So Maia, policyd, postfix, etc. on both machines work
with the same  
> data.
>
> Given the fact that both policyd-daemons use the same
data, I could  
> set
> both machines as MX-records, doing a Backup-MX-setup,
which would mean
> some low-level loadbalancing (don't NEED it ...) and,
more important,
> increased availability as there would be two machines
accepting mail.
>
> -
> For a start I would rsync nearly the whole server1 to
server2  
> regularly
> (with specific include/exclude-lists).
> In case of an emergency this would allow me to quickly
continue
> operations by switching off the ssh-tunnel and starting
the mysqld on
> server2 (which would at first release port 3306 and
then use it by the
> local mysqld).
>
> Pretty manual, I know, but at least a defined
emergency-procedure  
> to get
> into business again fast. And a way to reduce risk
quickly, after  
> that I
> plan to develop things further. I just don't want to
have server1  
> crash
> while I search for the IDEAL solution .... you know.
>
> -
>
> What I don't yet have solved in detail, is the fact
that
>
> a) people fetch mail from server1
> and
> b) all the mail should be stored in the
cyrus-imapd-boxes on server1.
>
> (/var/spool/imap would also be rsynced and could be
restored on  
> server2
> quickly).
>
> -
>
> So I want to teach server2 to deliver its scanned mail
into the
> cyrus-boxes on server1, or more likely, to hand it over
to the postfix
> on server1.
>
> Now I have looked through my setup on the productive
server1 and found
> the changes made related to Maia, which I had done
following
>
> htt
ps://secure.renaissoft.com/maia/wiki/Addresses
>
>
> So I assume that I will somehow have to setup another
smtpd on server1
> and allow server2 to "transport" (postfix)
its mail to it, although  
> this
> also means that the aliasing-stuff has to be done at
the right  
> point of
> processing.
>
> I am sure that I am close, but I am somewhat overworked
these days and
> would really appreciate if someone could give me some
pointers or even
> examples.
>
> Thanks a lot, best regards,
> Stefan
>
> _______________________________________________
> Maia-users mailing list
> Maia-usersrenaissoft.com
> http://www.renaissoft.com/mailman/listinfo/maia-users

_______________________________________________
Maia-users mailing list
Maia-usersrenaissoft.com
http://www.renaissoft.com/mailman/listinfo/maia-users
two servers, one
user name
2006-07-18 13:33:26
How about using LVS (http://www.linuxvir
tualserver.org)?  This way,  
you have a virtual IP that cyrus listens to for IMAP traffic
and  
postfix listens to for final delivery.  Each machine would
still have  
it's own IP that could be listed as separate MX records and
have  
Postfix/Maia listening for incoming mail.  You data would be
rsync'ed  
from Server1 to Server2 at regular intervals.

When both machines are up, incoming mail would be
load-balanced  
between the two, the messages would be scanned and then
handed to the  
virtual IP for final delivery.  Server1 would accept the
mail and  
deliver it the user's mailbox.  Incoming IMAP connections
would also  
get directed to Server1.  If Server1 fails, Server2 would
take over  
the Virtual IP and start accepting IMAP traffic. Postfix
would also  
be reconfigured to listen on the Virtual IP and be
restarted.

The big problem with this setup (or anything that uses
rsync) is file  
consistency.  If a message comes in 2 minutes before Server1
goes  
down, it will not exist on Server2 when the IP switchover
takes  
place.  If Server1's disk is dead, that message is *gone*,
but even  
if the drive is OK, that message is lost. Once Server1 comes
back on- 
line, you'll have to sync it with Server2 and once you do
that,  
everything that was on Server1 will be replaced.  If your
dataset is  
small enough, you could have a very small window between
rsyncs and  
this might not be a huge problem for you.  You could also
use DRBD to  
keep the servers in perfect sync or GFS if you have shared
storage.

-Eric

#########################################################
Eric Pierce, RHCE                  Phone: (813) 974-8868
Academic Computing                 Fax:   (813) 974-1799
University of South Florida        Email: epierceusf.edu



On Jul 15, 2006, at 3:23 PM, Stefan G. Weichinger wrote:

>
> Hello again,
>
> as some of you might remember, I am struggling with
enhancing the
> overall availability of my mail-services.
>
> Now I have rented a second dedicated server (tiny, I
know ;) ), and I
> want to do this:
>
> Identical setup on both machines.
>
> mysql-db on server1, server2 reads/writes from/to the
same server  
> via a
> ssh-tunnel (my experiments with mysql4-replication
didn't make me feel
> better ...).
>
> So Maia, policyd, postfix, etc. on both machines work
with the same  
> data.
>
> Given the fact that both policyd-daemons use the same
data, I could  
> set
> both machines as MX-records, doing a Backup-MX-setup,
which would mean
> some low-level loadbalancing (don't NEED it ...) and,
more important,
> increased availability as there would be two machines
accepting mail.
>
> -
> For a start I would rsync nearly the whole server1 to
server2  
> regularly
> (with specific include/exclude-lists).
> In case of an emergency this would allow me to quickly
continue
> operations by switching off the ssh-tunnel and starting
the mysqld on
> server2 (which would at first release port 3306 and
then use it by the
> local mysqld).
>
> Pretty manual, I know, but at least a defined
emergency-procedure  
> to get
> into business again fast. And a way to reduce risk
quickly, after  
> that I
> plan to develop things further. I just don't want to
have server1  
> crash
> while I search for the IDEAL solution .... you know.
>
> -
>
> What I don't yet have solved in detail, is the fact
that
>
> a) people fetch mail from server1
> and
> b) all the mail should be stored in the
cyrus-imapd-boxes on server1.
>
> (/var/spool/imap would also be rsynced and could be
restored on  
> server2
> quickly).
>
> -
>
> So I want to teach server2 to deliver its scanned mail
into the
> cyrus-boxes on server1, or more likely, to hand it over
to the postfix
> on server1.
>
> Now I have looked through my setup on the productive
server1 and found
> the changes made related to Maia, which I had done
following
>
> htt
ps://secure.renaissoft.com/maia/wiki/Addresses
>
>
> So I assume that I will somehow have to setup another
smtpd on server1
> and allow server2 to "transport" (postfix)
its mail to it, although  
> this
> also means that the aliasing-stuff has to be done at
the right  
> point of
> processing.
>
> I am sure that I am close, but I am somewhat overworked
these days and
> would really appreciate if someone could give me some
pointers or even
> examples.
>
> Thanks a lot, best regards,
> Stefan
>
> _______________________________________________
> Maia-users mailing list
> Maia-usersrenaissoft.com
> http://www.renaissoft.com/mailman/listinfo/maia-users

_______________________________________________
Maia-users mailing list
Maia-usersrenaissoft.com
http://www.renaissoft.com/mailman/listinfo/maia-users
two servers, one
user name
2006-07-27 16:44:36
Stefan G. Weichinger schrieb:
> Chris Black schrieb:
> 
>> When mail comes in it gets accepted and scanned on
the frontend server
>> and then passed to the backend server over smtp
directed by the postfix
>> transports table. Users read their mail via imap
from the backend server
>> and login to the frontend server to train the maia
spam filter if they
>> wish. Since both our servers are onsite we also
have the backend server
>> set up to accept port 25 traffic from internal
machines so local
>> delivery is a bit quicker and does not go through
maia.
> 
> Your setup is different because you have different
needs.
> Your servers are both onsite, while my two servers are
dedicated servers
> in some datacenters, and located X kilometers apart.
> 
> Also my usage of Maia is different, currently only I
and my fellow admin
> use its interface for training and reviewing, setting
individual
> filter-settings etc. No customer has been given
Maia-credentials yet
> (most of them would simply not fully understand and
therefore not use it
> ...).
> 
>> I understand this doesn't give full redundancy for
all services but
>> works well in our case and is how I chose to set
things up. If we need
>> to expand out it is much simpler to scale filtering
and backend storage
>> separately. It also massively simplifies management
and upgrades.
> 
> Ok. Although I don't see that much problems with my
setup also. But this
> could be related that I have "designed" it
along my own paths of
> understanding 
> 
>> As for your proposed configuration, one thing you
might consider is
>> doing the imap auth for Maia and not moving the
bayes db and awl db into
>> mysql. This way the maia mysql db would only
contain statistics and maia
>> configuration and you could share out the bayes db
and awl db via nfs
>> and you would have far less to worry about for
mysql.
> 
> Hm, good thought, I have to consider this. Right now I
have set up
> things as I described it, for two of my own domains,
and it works fine
> so far.
> 
> I only get SA timeouts, even after disabling the
auto-expiry (as
> mentioned on the amavisd-site and in some threads in
this list here).


I get stuff like this on both machines:

Jul 27 18:15:23 zion amavis[21373]: (21373-05) SA TIMED OUT,
backtrace:
at
/usr/lib/perl5/site_perl/5.8.7/Mail/SpamAssassin/BayesStore/
SQL.pm
line 819\n\teval {...} called at
/usr/lib/perl5/site_perl/5.8.7/Mail/SpamAssassin/BayesStore/
SQL.pm line
819\n\tMail::SpamAssassin::BayesStore::SQL::tok_get('Mail
::SpamAssassin::BayesStore::SQL=HASH(0xa8995e4)',
'5*VT\\x') called at
/usr/lib/perl5/site_perl/5.8.7/Mail/SpamAssassin/BayesStore/
SQL.pm line
2048\n\tMail::SpamAssassin::BayesStore::SQL::_put_tokens(
'Mail::SpamAssassin::BayesStore::SQL=HASH(0xa8995e4)',
'HASH(0xb5b3818)', 0, 1, 1154016877) called at
/usr/lib/perl5/site_perl/5.8.7/Mail/SpamAssassin/BayesStore/
SQL.pm line
963\n\tMail::SpamAssassin::BayesStore::SQL::multi_tok_coun
t_change('Mail::SpamAssassin::BayesStore::SQL=HASH(0xa8995e
4)',
0, 1, 'HASH(0xb5b3818)', 1154016877) called at
/usr/lib/perl5/site_perl/5.8.7/Mail/SpamAssassin/Bayes.pm
line
806\n\tMail::SpamAssassin::Bayes::learn_t...
Jul 27 18:15:33 zion amavis[21326]: (21326-05-2) SA TIMED
OUT,
backtrace: at
/usr/lib/perl5/site_perl/5.8.7/Mail/SpamAssassin/BayesStore/
SQL.pm line
2151\n\teval {...} called at
/usr/lib/perl5/site_perl/5.8.7/Mail/SpamAssassin/BayesStore/
SQL.pm line
2151\n\tMail::SpamAssassin::BayesStore::SQL::_put_tokens(
'Mail::SpamAssassin::BayesStore::SQL=HASH(0xa8995e4)',
'HASH(0xb6f50d4)', 0, 1, 1154016876) called at
/usr/lib/perl5/site_perl/5.8.7/Mail/SpamAssassin/BayesStore/
SQL.pm line
963\n\tMail::SpamAssassin::BayesStore::SQL::multi_tok_coun
t_change('Mail::SpamAssassin::BayesStore::SQL=HASH(0xa8995e
4)',
0, 1, 'HASH(0xb6f50d4)', 1154016876) called at
/usr/lib/perl5/site_perl/5.8.7/Mail/SpamAssassin/Bayes.pm
line
806\n\tMail::SpamAssassin::Bayes::learn_trapped('Mail::Sp
amAssassin::Bayes=HASH(0xa6fbe78)',
0, 'Mail::SpamAssassin::Message=HASH(0xb632adc)',
'HASH(0xb6299dc)',
'undef') called at
/usr/lib/perl5/site_perl/5.8.7/Mail/SpamAssassin/Bayes.pm
li...


Ideas anyone?

Stefan
_______________________________________________
Maia-users mailing list
Maia-usersrenaissoft.com
http://www.renaissoft.com/mailman/listinfo/maia-users
two servers, one
user name
2006-07-27 16:57:52
Eric Pierce schrieb:
> How about using LVS (http://www.linuxvir
tualserver.org)?  This way,  
> you have a virtual IP that cyrus listens to for IMAP
traffic and  
> postfix listens to for final delivery.  Each machine
would still have  
> it's own IP that could be listed as separate MX
records and have  
> Postfix/Maia listening for incoming mail.  You data
would be rsync'ed  
> from Server1 to Server2 at regular intervals.
> 
> When both machines are up, incoming mail would be
load-balanced  
> between the two, the messages would be scanned and then
handed to the  
> virtual IP for final delivery.  Server1 would accept
the mail and  
> deliver it the user's mailbox.  Incoming IMAP
connections would also  
> get directed to Server1.  If Server1 fails, Server2
would take over  
> the Virtual IP and start accepting IMAP traffic.
Postfix would also  
> be reconfigured to listen on the Virtual IP and be
restarted.
> 
> The big problem with this setup (or anything that uses
rsync) is file  
> consistency.  If a message comes in 2 minutes before
Server1 goes  
> down, it will not exist on Server2 when the IP
switchover takes  
> place.  If Server1's disk is dead, that message is
*gone*, but even  
> if the drive is OK, that message is lost. Once Server1
comes back on- 
> line, you'll have to sync it with Server2 and once you
do that,  
> everything that was on Server1 will be replaced.  If
your dataset is  
> small enough, you could have a very small window
between rsyncs and  
> this might not be a huge problem for you.  You could
also use DRBD to  
> keep the servers in perfect sync or GFS if you have
shared storage.

Sorry for the delay, a motorcycle accident kept me afk for 8
days now
(everything ok again, thanks).

I had a look at DRBD etc. some months ago already, but the
problem I
couldn't solve was the lack of a common IP-range:

My two servers are NOT located in the same geographic
location and
therefore NOT in the same LAN-subnet. So I couldn't figure
out *how* to
share *which* common virtual IP between the two machines.

My machines are rented dedicated servers at two different
hosting
companies and located in two completely unrelated
data-centers. I have
two static IP-addresses for each of the boxes, but I don't
yet see how
to set up the cluster-IP that is needed for heartbeat, and
more
important to get the traffic routed correctly.

So I came to the described setup, it works Ok so far for two
of domains,
except for the SA-timeouts described in my other posting.

Thanks, Stefan
_______________________________________________
Maia-users mailing list
Maia-usersrenaissoft.com
http://www.renaissoft.com/mailman/listinfo/maia-users
two servers, one
user name
2006-07-27 19:53:27
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Stefan G. Weichinger wrote:

>
/usr/lib/perl5/site_perl/5.8.7/Mail/SpamAssassin/BayesStore/
SQL.pm line
>
819\n\tMail::SpamAssassin::BayesStore::SQL::tok_get('Mail
::SpamAssassin::BayesStore::SQL=HASH(0xa8995e4)',
> '5*VT\\x') called at
>
/usr/lib/perl5/site_perl/5.8.7/Mail/SpamAssassin/BayesStore/
SQL.pm line
>
2048\n\tMail::SpamAssassin::BayesStore::SQL::_put_tokens(
'Mail::SpamAssassin::BayesStore::SQL=HASH(0xa8995e4)',
> 
> 
> Ideas anyone?

This has been annoying the snot out of me.  The bayes
database is simply too
slow on updates.   No solution has been found yet.


- --
David Morton
Maia Mailguard                        - http://www.maiamailguard
.com
Morton Software Design and Consulting - http://www.dgrmm.net
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2.2 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org


iD8DBQFEyRm2Uy30ODPkzl0RAo1vAJ0TcCNVNL4AkffgB6o6PgKxi+B+DwCg
quQP
hJt34jjcwdRy9X6CKich/TI=
=9iF5
-----END PGP SIGNATURE-----
_______________________________________________
Maia-users mailing list
Maia-usersrenaissoft.com
http://www.renaissoft.com/mailman/listinfo/maia-users
[1-10] [11-12]

about | contact  Other archives ( Real Estate discussion Medical topics )