Cloud and Infrastructure - Tools, storage and more...

fault tolerance, high availability, cluster, failover, redundancy, ... : let's make thing clearer


software running on one or all of the cluster nodes, working with a pool of resource agents.
Some CRM specialize in dispatching work to cluster nodes (grid computing), whereas others are dedicated to high availability and load balancing.
In short, its role is to make sure everything is going extremely well (health check, "monitor" part), and take according actions ("start/stop/enable/disable" part) when needed.
set of computers that work together so that, in many respects, they can be viewed as a single system
automatic switchover triggered by an incident
high availability
characteristic of a system which aims to ensure an agreed level of operational performance (aka SLA, usually uptime) for a higher than normal period
duplication of critical components of a system with the intention of increasing reliability of the system
resource agent
software running on the cluster nodes to effectively manage and configure devices (IP addresses, ...), start / stop the services (daemons, ...).
These agents are standardized interfaces (i.e. : abstraction layers) for a cluster resource. They are used to :
  • translate a standard set of operations into steps specific to the resource or application
  • and interpret their results as success or failure
Their duty is mostly to :
  • start / enable
  • stop / disable
  • monitor
  • validate configuration
Most resource agents are coded as shell scripts (resource agents implementation, resource-agents).
manual switch from one system to a redundant (aka standby) system upon incident or for maintenance


Name Role Notes
Corosync cluster communication + membership Often associated with Pacemaker, Corosync is the communication layer.
HAProxy TCP + HTTP reverse proxy
  • As for the HTTP reverse proxy functionality only (ignoring factors such as performance or other functionalities such as caching), what HAProxy does could be performed as well by :
  • Even though it has "high availability" in its name, this is a standalone application that needs to be redunded like any other resource.
Heartbeat (1, 2, 3) cluster communication + membership
  • Must be associated to a CRM such as Pacemaker
  • Since up to release 2.1.4 the messaging layer (Heartbeat proper), the Local Resource Manager, "plumbing" infrastructure and STONITH (now known as Cluster Glue), the Resource Agents, and the CRM (now Pacemaker) were all part of a single package named Heartbeat, the name was often applied to the Linux-HA project as a whole.
    This generalization is no longer accurate, the name Heartbeat should thus be used for the messaging layer exclusively. (source)
Pacemaker CRM Often associated with Corosync.

	a cluster-oriented product such as heartbeat will ensure that a shared resource will be present at *at most* one place. This is very important for shared filesystems, disks, etc... It is designed to take a service down on one node and up on another one during a switchover. That way, the shared resource may never be concurrently accessed. This is a very hard task to accomplish and it does it well.
	a network-oriented product such as keepalived will ensure that a shared IP address will be present at *at least* one place. Please note that I'm not talking about a service or resource anymore, it just plays with IP addresses. It will not try to down or up any service, it will just consider a certain number of criteria to decide which node is the most suited to offer the service. But the service must already be up on both nodes. As such, it is very well suited for redundant routers, firewalls and proxies, but not at all for disk arrays nor filesystems.

	==> The difference is very visible in case of a dirty failure such as a split brain. A cluster-based product may very well end up with none of the nodes offering the service, to ensure that the shared resource is never corrupted by concurrent accesses. A network-oriented product may end up with the IP present on both nodes, resulting in the service being available on both of them. This is the reason why you don't want to serve file-systems from shared arrays with ucarp or keepalived.

How to detect a new disk added to a VMware virtual machine ?

for i in /sys/class/scsi_host/host*; do grep -q 'mptspi' "$i/proc_name" && echo '- - -' > "$i/scan"; done; lsblk

How it works

  • In the /sys pseudo filesystem, we have :
    ├── host0
    │    └── proc_name
    ├── host1
    │    └── proc_name
    └── host2
         └── proc_name
  • Each proc_name contains the name of the module used by the SCSI host it belongs to, generally one of :
    • ata_piix
    • mptspi
  • The SCSI host we're trying to identify (i.e. the one holding the virtual disks) is the one using the mptspi module (looks specific to how VMware handles this)
  • Once we've found the SCSI host, we just have to "force rescan" by writing - - - into /sys/class/scsi_host/host2/scan


LinuX Containers is an operating-system-level virtualization method for running multiple isolated Linux systems (i.e. containers) on a control host using a single Linux kernel. Containers offer an environment as close as possible as the one you'd get from a VM but without the overhead that comes with running a separate kernel and simulating all the hardware.

The Linux kernel provides :

LXC combines both to provide an isolated environment for applications.
Early versions of Docker used LXC as the container execution driver, though LXC was made optional in v0.9 and support was dropped in Docker v1.10 (2016-02-04).


apt install lxc
And that's it !

Create a new container

lxc-create --template download --name myContainer
echo 'USE_LXC_BRIDGE="true"' >> /etc/default/lxc-net

systemctl status lxc
● lxc.service - LXC Container Initialization and Autoboot Code
   Loaded: loaded (/lib/systemd/system/lxc.service; enabled; vendor preset: enabled)
   Active: inactive (dead)
     Docs: man:lxc-autostart

systemctl start lxc

	Everything should be stated as "enable" in green color. If not, try to reboot the system.
lxc-create --name ubuntu -t download
Setting up the GPG keyring
gpg --no-default-keyring --keyring trustedkeys.gpg --fingerprint
lxc-create --name ubuntu -t download -B best

Setting up the GPG keyring
ERROR: Unable to fetch GPG key from keyserver.

lxc-create --name ubuntu -t download -B best --logpriority=DEBUG
# The DNS name of the LXD server to source the base container cache from

# The keyservers to use when validating GPG keys for the downloaded cache
lxc_image_cache_primary_keyserver: hkp://
lxc_image_cache_secondary_keyserver: hkp://

wget hkp://

gpg --keyserver hkp:// --recv-keys E084DAB9

nc -vz 80 11371
nc -vz 80 11371

nmap -sT -p 80,443
nmap -sT -p 80,11371

As root :
lxc-create --template download --name myContainer -- -d debian -r strech -a amd64

Manage containers

list containers :
lxc-ls -f
destroy a container :
lxc-destroy --name myContainer

Start / stop a container


How to register a Debian host into Active Directory ?

  1. apt-get install adcli libnss-sss libpam-sss realmd samba-common-bin sssd sssd-tools
  2. systemctl enable sssd
  3. realm discover myDomain.local
    	type: kerberos
    	realm-name: MYDOMAIN.LOCAL
    	domain-name: myDomain.local
    	configured: no
    	server-software: active-directory
    	client-software: sssd
    	required-package: sssd-tools
    	required-package: sssd
    	required-package: libnss-sss
    	required-package: libpam-sss
    	required-package: adcli
    	required-package: samba-common-bin
  4. realm join --user=domainAdmin myDomain.local
  5. systemctl start sssd
  6. check :
    getent passwd user@myDomain.local
    user@myDomain.local:*:919801223:919800513:John SMITH:/home/myDomain.local/user:/bin/bash
    If you get an answer like that, it's ok
  7. Enable creation of a home directory for domain users (with the specified skeleton) :
    echo 'session required skel=/etc/skel/ umask=0022' | sudo tee -a /etc/pam.d/common-session
  8. Let domain administrators become local admins (i.e. sudoers) (complete the domain name below ;-) :
    1. aptitude install libsss-sudo
    2. echo '%domain\ admins@myDomain.local ALL=(ALL) ALL' | sudo tee -a /etc/sudoers.d/domain_admins
  9. To login as user instead of user@myDomain.local, set in /etc/sssd/sssd.conf :
    use_fully_qualified_names = False
  10. You'll also need in /etc/sssd/sssd.conf :
    simple_allow_groups = list,of,AD,groups,allowed,to,login
  11. systemctl restart sssd
  12. update DNS records accordingly

How to mount Google Drive on Debian GNU/Linux ?

Setup :

As root :
  1. apt-get install opam ocaml make fuse camlp4-extra build-essential pkg-config
  2. groupadd fuse
  3. adduser bob fuse
  4. chown root.fuse /dev/fuse; chmod 660 /dev/fuse
As Bob :
  1. just in case, backup ~/.profile
  2. opam init
  3. opam update
  4. opam install depext
  5. opam depext google-drive-ocamlfuse
  6. opam install google-drive-ocamlfuse
  7. . $HOME/.opam/opam-init/
  8. google-drive-ocamlfuse
    This will open a browser window on the Google Drive page, asking for credentials + allowing google-drive-ocamlfuse to access files.

Mount :

mountPoint='/home/bob/googleDrive'; mkdir "$mountPoint"; google-drive-ocamlfuse "$mountPoint"

On the first time, this will open a browser window to check permissions.

Check : mount | grep -q "$mountPoint" && echo 'GOOGLE DRIVE IS MOUNTED' || echo 'GOOGLE DRIVE NOT MOUNTED'

Unmount :

  • sudo umount "$mountPoint"
  • kill -1 $(pidof google-drive-ocamlfuse)

Load balancing

Galera cluster load balancing

Cluster deployment variants

An Introduction to HAProxy and Load Balancing Concepts

Load balancing types (source) :

Layer 4 load balancing :
Relying on the transport layer, this forwards a request to a backend server based on the destination IP+port of the request. All backend servers are expected to be able to return identical responses to any given request.
HA Proxy layer 4 load balancing
Layer 7 load balancing :
Relying on the application layer, this forwards a request to a backend server based on the contents of the request itself (e.g. the requested URL). In this mode, backend servers need not being clones of each other since they may serve different contents.
HA Proxy layer 4 load balancing

Algorithms (source) :

select servers in turns (default in HAProxy)
select the server with the least number of connections (recommended for longer sessions)
select which server to use based on a hash of the source IP address. This ensures that a user will connect to the same server.

High availability of the load balancer itself (source, details) :

The load balancer itself mustn't be the SPOF of the infrastructure. This is why it has to become highly available via redundancy :

  • a floating IP is assigned to the 1st load balancer, which performs normally
  • a 2nd (passive) load balancer is added to the infrastructure
  • health check is performed on both load balancers
  • should the active load balancer fail, the floating IP is transferred to the spare load balancer, which becomes active
High Availability setup animated

This can be achieved with tools such as :