OS version for the Controller: Debian 9.8 (stretch)
- Netinstal ISO: https://cdimage.debian.org/debian-cd/current/amd64/iso-cd/debian-9.8.0-amd64-netinst.iso.
Hardware:
- At least two network controllers attacged (in my case both are Ethernets).
The following steps are prepared based on Virtual Box simulation for netinst ISO file (above) and tested on real hardware. As a result you will have a cleanly installed OS.
- Boot the netinstaller from downloaded image.
- Select Install.
- Language: English.
- Country, territory or area: other -> Europe -> Poland.
- Default locale settings: United States.
- Keymap: Polish.
- Primary network interface: enp0s3.
- Hostname: cnc (as this machine will work as an command and control station).
- Domain name: can be ommited (none).
- Root password: you should give some :)
- Normal user: create one if you wish.
- Partition disks: you can do any, but I've basically have chosen Guided - use entire disk.
Now wait for the base system packages to be installed...
- Select an installation mirror (or don't).
- Participate in package popularity contest (or don't).
- Software selection: Uncheck all suggestions. All necessary packages can be installed later via
apt install. - Install the GRUB boot loader on a hard disk: select Yes and /dev/sda.
- Finish!
All commands and scripts assume that you are logged in as root, including remote session.
Install man-pages man reader and ifconfig.
apt install --no-install-recommends --yes man
Install and run OpenSSH Server. Remote terminal is more user friendly, at least for me :)
apt install --no-install-recommends --yes openssh-server net-tools
Install additional helper tools like:
mc- Midnight Commanderhtop- atopalternativetmux- for multiple text terminals in a single windowvim- because you love it :)
apt install --no-install-recommends --yes mc htop tmux vim
At this point we have a machine that has a default network setting.
To see it, run ifconfig -a command. Output should be something like this:
enp0s3: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500
inet6 fe80::a00:27ff:fee9:7f29 prefixlen 64 scopeid 0x20<link>
ether 54:e6:fc:80:75:6e txqueuelen 1000 (Ethernet)
RX packets 0 bytes 0 (0.0 B)
RX errors 0 dropped 0 overruns 0 frame 0
TX packets 8 bytes 648 (648.0 B)
TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0
lo: flags=73<UP,LOOPBACK,RUNNING> mtu 65536
inet 127.0.0.1 netmask 255.0.0.0
inet6 ::1 prefixlen 128 scopeid 0x10<host>
loop txqueuelen 1 (Local Loopback)
RX packets 0 bytes 0 (0.0 B)
RX errors 0 dropped 0 overruns 0 frame 0
TX packets 0 bytes 0 (0.0 B)
TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0
enp0s8: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500
inet6 fe80::a00:27ff:fe82:2160 prefixlen 64 scopeid 0x20<link>
ether 14:dd:a9:79:4c:87 txqueuelen 1000 (Ethernet)
RX packets 173 bytes 16801 (16.4 KiB)
RX errors 0 dropped 0 overruns 0 frame 0
TX packets 145 bytes 29013 (28.3 KiB)
TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0
The lo device is the loopback and should be left alone. However both enps are what we want to play with. Those names are of our two network cards.
And because they will change from configuration to configuration I've decided to normalize them by changing their names.
Bare in mind that new names will be used further in this tutorial along with any config file that is available here.
For this we will use systemd functionality. Since systemd is here and it's not going anywhere soon we should try and use it after all.
Debian installer creates /etc/network/interfaces file by default. Rename it to interfaces.old and replace it with systemd-based approach.
To change name of a network interface we need to know its MAC address (see ifconfig -a output).
Let us assume that 08:00:27:e9:7f:29 identifies an interface that is connected to the internal (inner) network with cluster node machines,
while 08:00:27:82:21:60 is connected to the outer world network (LAN, WAN or whatever - the Internets perhaps?).
If so, create two files: /etc/systemd/network/10-inner.link and /etc/systemd/network/10-outer.link with the following content:
[Match]
MACAddress=xx:xx:xx:xx:xx:xx
[Link]
Name=__name__goes__here__
however replace the xx:xx:xx:xx:xx:xx and __name__goes_here__ parts with MAC address correlated with proper inner/outer name as follows:
- Outer interface file
/etc/systemd/network/10-outer.link:
[Match]
MACAddress=14:dd:a9:79:4c:87
[Link]
Name=outer
- Inner interface file
/etc/systemd/network/10-inner.link:
[Match]
MACAddress=54:e6:fc:80:75:6e
[Link]
Name=inner
Now we should create ISO Layer-3 description for both interfaces with the following files (see /confs/etc-systemd-network/ in this repo).
Those files will start outer network interface in DHCP mode and inner as 10.10.0.1.
- Outer network file
/etc/systemd/network/outer.network:
[Match]
Name=outer
[Network]
DHCP=v4
- Inner network file
/etc/systemd/network/inner.network:
[Match]
Name=inner
[Network]
Address=10.10.0.1/16
#Gateway=192.168.0.104
#DNS=8.8.8.8
Now update your initrd configuration by running update-initramfs -u and reboot your machine. New settings will be used during the next boot.
If not, try to enable networkd service by running start to start and enable to start it at boot time:
systemctl start systemd-networkd
systemctl enable systemd-networkd
After that check your new settings (ifconfig -a) and you should see something like this:
inner: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500
inet 10.10.0.1 netmask 255.255.0.0 broadcast 10.10.255.255
inet6 fe80::a00:27ff:fee9:7f29 prefixlen 64 scopeid 0x20<link>
ether 54:e6:fc:80:75:6e txqueuelen 1000 (Ethernet)
RX packets 0 bytes 0 (0.0 B)
RX errors 0 dropped 0 overruns 0 frame 0
TX packets 8 bytes 648 (648.0 B)
TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0
outer: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500
inet 192.168.1.108 netmask 255.255.255.0 broadcast 192.168.1.255
inet6 fe80::a00:27ff:fe82:2160 prefixlen 64 scopeid 0x20<link>
ether 14:dd:a9:79:4c:87 txqueuelen 1000 (Ethernet)
RX packets 318 bytes 28663 (27.9 KiB)
RX errors 0 dropped 0 overruns 0 frame 0
TX packets 264 bytes 44977 (43.9 KiB)
TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0
If so, connect a SSH client (eg. putty) to the IP address shown (here it's 192.168.1.108) and now you can work remotely.
Remote session doesn't allow you to connect as root (and that's good).
You need to create your private user, log into it and switch to root by issuing simple:
su
Firstly you have to install git tool and https support (SSL):
apt install --no-install-recommends --yes git
apt install --no-install-recommends --yes ca-certificates openssl
and clone this repository into /srv/dc directory.
cd /srv
git clone https://github.com/tomekjaworski/DanteCluster dc
For this task you can follow the Official Docker Docs here or run 01_install_docker.sh script based on it.
./01_install_docker.sh
At this point network and Docker is up and running. We can start TFTP and DHCP services for the inner network. No machine will boot up since there is no kernel provided yet (no NFS filesystem is set) but at least we will know it this basic setup is ok.
To do this tart both services via ./make.sh and ./runme.sh from both /srv/dc/service_dhcp and /srv/dc/service_tftp. The runme scripts can be configured to run in foreground. If so, use tmux as message sink or change those scripts to run docker detached.
While any cluster machine is powered on you should see DHCPLEASE being requested and followed by pxelinux.0 retrieval.
Of course this machine should be configured to bootup via PXE. If so you should see somethig more or less like this:
From Syslinux Wiki.
In order to provide / directory (and its content) to the nodes the NFS server is used. To install and initialize it, use the following commands:
apt install nfs-kernel-server
Now, to initialize the client tracker (nfsdcltrack) type:
mkdir /var/lib/nfs/nfsdcltrack
nfsdcltrack init
After that the file /var/lib/nfs/nfsdcltrack/main.sqlite should exist.
Now add the following line to /etc/exports and restart the NFS server with systemctl restart nfs-server.service:
/srv/dc/nfsroot 10.10.0.1/255.255.0.0(ro,sync,no_root_squash,fsid=0)
It will open a remote mount point named /srv/dc/nfsroot available at 10.10.0.1 (this machine) for the whole subset /16.
You can now go and try to test mount this resource by typing:
mkdir /test_mount
moun -t nfs 10.10.0.1:/srv/dc/nfsroot /test_mount
You should be now able to see your /srv/dc/nfsroot content under /test_mount.
This step is based on debootstrap Debian package and modified scripts, provided by Markus Rosenstihl for his own high-performance cluster.
More information you can find on his Bitbucket webpage and Wiki.
Before we can start, the debootstrap package should be installed:
apt install --no-install-recommends --yes debootstrap
Now run main script:
./bootstrap.sh -d /srv/dc/nfsroot
This will take couple of minutes - some coffee is suggested :) After successful instalation you will have a Debian distro in /srv/dc/nfsroot.
WORK IN PROGRESS