New Cluster some basic questions

B

bofh

Well-Known Member
Nov 7, 2017
148
18
58
45
Hello,

I want todo the following setup. Outbound only, (basically sending host for applications and printer), 3 nodes on 3 differfent locations via site2site wireguards.
configuration should be identical on each.

as far i understood, each can work independent (which is the point to have one on each site) and only config will fail in case of a node fail or network error.

now on the clientside (send smtp setting for my clients) i would like to have also the option to be kinda redundant.
so multiple A entries for that cluster like mailcluster.example.local gets 3x CNAME entry to mailnode1.example.local, mailnode2.example.local and mailnode3

is there any better way to not have total roundrobin (id rather have a fallback instead)
anything else iam missing or should that work this way ?
and is latency for the cluster not a problem like with corosync ?
 
Last edited:
sb-jw said:
At least in postfix you can define several relay hosts [0], which are used one after the other. It's just not clear from the documentation whether these are checked beforehand or not.

[0] https://www.postfix.org/postconf.5.html#relayhost
Click to expand...
yea i know but the idea is to get failover for the internal gateway. so if one machine has to reboot still another one answers.
main sender is a software that will simply discard that mail without notice if the defined server times out (stupid i know) . and we have about 1-3k mails per day.
but i can onyl set it up like any other regular smtp client. postfix relay hosts would only help for redundancy for sending outside. that i wont need, thats a 3rd party mail provider promising they have a cluster running, but even if not, once postfix has it it wont get lost and just sent later - doesnt matter

its the dreaded last mile, stupid mail clients. oh scanner are also such an offender. they also simply trow an error and discard if they cant reach that smtp asap
 
Last edited:
Then you will probably have to connect something like a HAProxy. With GlusterFS and keepalived you can quickly build a good and stable solution. It's just annoying that you have to connect an additional service.

Theoretically, you could of course also set up corresponding HA IPs on the PMG with keepalived. Personally, I wouldn't do that for reasons of upgradeability.

Or you have to build a script on the sending system that pings the PMG beforehand and asks whether it is there.
 
You must log in or register to reply here.
Top Bottom