Multiserver environment setup

I've been reading through the 3.1 user manual on how to set up the example environment (web, email, db, ns1, ns2) - all very clear, but I have a couple questions since the manual is 6 years out of date - clearly some of the RPM-specific install commands have changed.

1. using the 5-server setup, would it be best to use the portions of the Perfect Server installation guide for my OS that relate to each server use, and get the softwares installed then make the configuration changes the user manual suggests?

2. and if I were to use a 3-server setup instead and merge some features (web/email, db/ns1, ns2), would i (using the web/email server as an example) do the web AND email software installs and OS configs BEFORE doing the ISPConfig-specifc instructions? seems logical, and less likely to change something if I go through the ISPConfig stuff twice targeting different settings.

3. And a clarification on the Mirror setup - the ISPConfig-related configuration data (accounts, email address, domains, and DNS data) will all be set up on the mirror machine - it's just the user-specific data (specifically web, email and databases) won't be mirrored. the /var/www and /var/vmail substructures will all be there (cur/new/tmp vmail folders for each user, for example) , but with no user data in them. and the databases will all exist, but be empty. is that correct?

4. And in the mirror setup, for services like DNS that are not user-specific, i should be able to use the mirror (2nd) server as a live, second DNS server, correct? because changes on the first will immediately be updated to the second in near real time, correct?

5. i see in the user manual instructions for mirroring you include the Unison and Dovecot instructions for mirroring the data as well - thanks! are those still up-to-date and accurate, or is there a new preferred mechanism?

6. I didn't see mention of how to promote a slave to a master within ISPConfig if the master fails. I assume there's a mechanism for it, maybe I missed it?

7. when promoting the slave to master, besides the IP redirect, should the server name be changed to the old master name (ns2 -> ns1 if ns1 was master and dies) so that IP/names match again? and then a new slave can be built using the old slave name and fall into place pretty seamlessly?

Sorry for all the questions, but trying to decide how best to set things up. I've been running 2 separate servers for a while, and that's worked, but I like the idea of mirrored servers so i can hot-swap if one crashes (which is usually my reason for replacing a server). redirecting a static external IP to a different internal IP is a quick fix, and the mirroring would be a good backup scenario too.

 

thanks!
are the instructions similar for CentOS 8?

 

personally, i would use a separate server for email.
it's by far the part that takes up most time, customers complaining can't connect, or not receiving an email that they're "sure" has been sent to them. having to find out what server their mailbox is on first is just extra time wasted, plus it means everyone uses the same mail connection settings, so that simplifies walking technophobes through connecting their brand new phone they have no idea how to use.

also, the idea of a central db server(s) seems great. can upgrade it as and when required without taking sites offline etc. that's what i thought anyway, and how i first setup a multi-server ispconfig system. But even with fast networks, the way cms's like wordpress build their pages with sequential db requests instead of sending them all off at once, even that minimal extra latency can translate into sites running much slower than having the database on the same server as apache/nginx and using localhost to connect to the database.

 

But you and other users of this linux distro can help by forking and making the auto installer to suit CentOS and other similar distos. It will be 'yum'my once cooked.

 

I am curios, how much problems do you expirience with SELinux on CentOS regarding ISPConfig users? I assume there are quiet a few things that may be buggy due to the default SELinux implemenation, or is it just some very sepcial cases`?