CentOS - Nov 2019 - Post-installation setup script for CentOS 7 servers

Le 17/11/2019 ? 14:15, Jonathan Billings a ?crit?:
> I?m curious why you list these as ?cruft? packages?
> 
> chrony
> firewalld
> iperf
> NetworkManager-libnm
* chrony: I'm using ntpd and ntpdate

* firewalld: https://github.com/kikinovak/firewall

* iperf: replaced by iperf3

* NetworkManager: great on laptops, useless on servers
> 
> Also, I?m sure it?s helpful for you, but setting all the default CentOS
repos to ones in France as a baseurl means that anyone using it would get that
too, so anyone thinking of forking the repo, don?t replace the CentOS repo
files. The mirrorlist url should give you geographically close mirrors.
You're right. The main reason I did this is because some of my servers
are behind filtering proxies, and more often than not, CentOS mirrors
are blocked.

I have to think about a practical solution for this.

Thanks for the feedback !

Niki

-- 
Microlinux - Solutions informatiques durables
7, place de l'?glise - 30730 Montpezat
Site : https://www.microlinux.fr
Mail : info at microlinux.fr
T?l. : 04 66 63 10 32
Mob. : 06 51 80 12 12

On Nov 17, 2019, at 11:58 AM, Nicolas Kovacs <info at microlinux.fr>
wrote
> * chrony: I'm using ntpd and ntpdate
You should never be using ntpdate anymore (which is why the ntp project is
deprecating it, http://support.ntp.org/bin/view/Dev/DeprecatingNtpdate
<http://support.ntp.org/bin/view/Dev/DeprecatingNtpdate> ).  I really only
ever suggest ntpd unless you?re running an NTP server that provides NTP service
to your network, and needs to supported time source hardware.  NTPd actually
worse for laptops and other devices with intermittent/congested connections, and
VMs that experience time jumps during migrations.  Chrony also tends to use less
RAM and power than NTPd due to how it does time management and generally smaller
footprint.
> * firewalld: https://github.com/kikinovak/firewall
<https://github.com/kikinovak/firewall>
Kinda looks like you?ve reinvented the wheel here, breaking down firewall rules
into separate files and managed by a single service.  Plus, firewalld supports
ipsets along side iptables rules in C7, and uses nftables by default in C8,
keeping you with the fastest way of setting up rules.  But I get it, not
everyone cares for firewalld.  On c6, I managed the iptables file with a
template in configuration management, breaking up the individual config files
into separate, role-based chunks.

Also, the ?fail2ban? service has firewalld support, which uses ipsets for its
blocks, improving overall performance.
> * NetworkManager: great on laptops, useless on servers
Untrue.  NM is great for servers.  I think I?ve told this story a dozen times on
this list, but nearly all our servers use NM.  We experienced a power outage in
our datacenter due to some clumsy UPS maintenance people, and when power was
restored to the floor, the servers booted faster than the networking equipment. 
Everything using the old ?network? service booted up, detected no network, and
gave up and completed the boot, with no network at all.  Had to visit the
datacenter to reboot them.  All the NM systems had the network start fail, and
continued with the boot, and as soon as the interface comes online, NM brings up
the network and triggers all network-dependent services to come online.

NM supports event-based dispatching (in /etc/NetworkManager/dispatcher.d/) so
you can run custom scripts when the network state changes.  NM in CentOS7 is a
lot better than you had in C6, the default settings don?t restart the interface
if you change the ifcfg-* files (a stupid problem in C6) and supports a lot more
features like bridges and bonds that were missing in earlier versions.  You can
interact with it via dbus (if that?s your thing) and the nmcli tool is really
handy for CLI-based settings.

The only time I?ve seen a need to use the old network service was when I
discovered that you can?t set custom routes on the loopback interface with NM,
since it doesn?t manage the loopback interface.



--
Jonathan Billings

Le 17/11/2019 ? 18:56, Jonathan Billings a ?crit?:
> You should never be using ntpdate anymore (which is why the ntp project is
deprecating it, http://support.ntp.org/bin/view/Dev/DeprecatingNtpdate
<http://support.ntp.org/bin/view/Dev/DeprecatingNtpdate> ).  I really only
ever suggest ntpd unless you?re running an NTP server that provides NTP service
to your network, and needs to supported time source hardware.  NTPd actually
worse for laptops and other devices with intermittent/congested connections, and
VMs that experience time jumps during migrations.  Chrony also tends to use less
RAM and power than NTPd due to how it does time management and generally smaller
footprint.
I know ntpdate has officially been deprecated for ages. This being said,
it works reliable when you have some serious lagging on the server.
> 
>> * firewalld: https://github.com/kikinovak/firewall
<https://github.com/kikinovak/firewall>
> Kinda looks like you?ve reinvented the wheel here, breaking down firewall
rules into separate files and managed by a single service.  Plus, firewalld
supports ipsets along side iptables rules in C7, and uses nftables by default in
C8, keeping you with the fastest way of setting up rules.  But I get it, not
everyone cares for firewalld.  On c6, I managed the iptables file with a
template in configuration management, breaking up the individual config files
into separate, role-based chunks.
This is probably my Slackware background, but for many years,
firewalling was essentially a shell script with iptables rules. From
this point of view, firewalld has reinvented the wheel, so I simply
stick with what works and what I'm familiar with. And since Linux obeys
the Great Rule of Herding Cats, along comes nftables. BTW, the file
snippets are just templates meant to be copy/pasted with Vim using split
mode. :o)

> 
>> * NetworkManager: great on laptops, useless on servers
> Untrue.  NM is great for servers.  I think I?ve told this story a dozen
times on this list, but nearly all our servers use NM.  We experienced a power
outage in our datacenter due to some clumsy UPS maintenance people, and when
power was restored to the floor, the servers booted faster than the networking
equipment.  Everything using the old ?network? service booted up, detected no
network, and gave up and completed the boot, with no network at all.  Had to
visit the datacenter to reboot them.  All the NM systems had the network start
fail, and continued with the boot, and as soon as the interface comes online, NM
brings up the network and triggers all network-dependent services to come
online.
Again, this is probably the ex-Slacker in me throwing all the junk out
and just keeping what's really needed.

Cheers & thanks for your detailed explanations. You've tickled my
curiosity, so as soon as I finish writing my current Linux book (around
X-mas I guess) I'll have a deeper look at all that stuff.

Niki


-- 
Microlinux - Solutions informatiques durables
7, place de l'?glise - 30730 Montpezat
Site : https://www.microlinux.fr
Mail : info at microlinux.fr
T?l. : 04 66 63 10 32
Mob. : 06 51 80 12 12