Hi! On Mon, May 10, 2010 at 6:47 AM, spambox <[email protected]> wrote: > > > On Mon, 10 May 2010 01:00:29 -0430 > Jose Ildefonso Camargo Tolosa <[email protected]> wrote: > >> Hi! >> >> Yes, I know, this is kinda off-topic, but looks interesting. > > Thank you for your interest! > >> > |-------------| >> > | postfix1 | very big storage >> > | LDAP1 |-------------|******************** >> > | server1 | | * >> > |-------------| | /home1 * >> > | | * >> > | |-------------------* >> > |-------------| | * >> > | postfix2 | | * >> > | LDAP2 |-------------| /home2 * >> > | server2 | | * >> > |-------------| | * >> > | |-------------------* >> > | | * >> > |-------------| | /home3 * >> > | postfix3 | | * >> > | LDAP3 |-------------|******************** >> > | server3 | >> > |-------------| >> > >> > >> > >> > > >> Ok, so, load balancing here, with different homes for each server, so, >> each server have its own "set of users", right? > > Yes, each LDAP have list of his users. For each LDAP entry we have a record > telling us, who is the Mail server for that user. So if the balancer send > user "test" to mail server1 and from the LDAP we see that "test" user belong > to server1 then "test" request is served, ELSE if user "test" has mail server > set on "mailserver3" for example, the request is redirect to mailserver3 that > will handle his request.
Cool, I have implemented a similar approach, but on postfix level, ie, each smtp server can receive mail for any user, and then look for the smtp server on a ldap attribute (I use maildrop attr for that). I use this to have a "distributed email system", so that each user have its mail "locally" on the locality he/she works. If one server is unavailable, the system will keep the mail queued until the corresponding server is back, so: no HA here. > > >> > >> > if one server goes down, the one next to him will Up his ip address and >> > mount his partition on the storage untill the server restarts correctly. >> >> Now, this is more typical of HA clusters, what gets me confused is >> what I asked on the last question, so, when say, server 1 goes down, >> server 2 mount server1's home partition and start getting mails (and >> serving IMAP requests) as if it were server1, right? I'll reserve my >> other comments until I get more information on this. > > Yes, if server1 goes down the cluster will mount up server1 interface and > home partition on server2. Server2 will temporarily act as server1 and > server2 untill server1 is back on line. Ok. > > > > >> > 1) I've never used Courier IMAP, but reading on the internet i've found >> > that it use gethostbyname() to resolve and this may cause an infinite loop >> > for a host with 2 interface but one hostname if one server goes down. >> > So, i'm afraid that i will be able to installa courier-imap on each server >> > and this will cause a big problem to the scalability of the system. >> > >> > 2) Has anyone ever tryed something like this and its ablet to redirect me >> > to some good documentation? >> >> I don't use Courier (several reasons, off-topic here), I use Dovecot >> (I can even use SASL from Dovecot to authenticate Postifx's smtp, >> really handy along with reject_authenticated_sender_login_mismatch and >> smtpd_sender_login_maps on ldap). > > Nice, i gave a look to this Dovecot and its looks pretty nice! i'vent tested > it yet but it seems it can be a valid alternative. Im trying to move from > Qmail to postfix because Qmail is patch-project(dead), while i think Postfix > is a valid alternative that will make life a lot easyer. postfix is extremely flexible. Believe me: you can replace qmail with *no* changes to your ldap entries. Your project looks interesting, Ildefonso Camargo
