Virtual cluster

On the host, you must set these sysctl settings:

net.bridge.bridge-nf-call-arptables = 0
net.bridge.bridge-nf-call-ip6tables = 0
net.bridge.bridge-nf-call-iptables = 0

You must define 3 network interfaces on each host of your cluster.

• One interface connects to a virtual network in NAT mode
• Two interfaces connect to two virtual networks with a MTU set to 9000 (it's to simulate an ethernet cable between two machines)

Inventories

The inventory must define these hosts to run:

• cluster_machines: Set of hosts in the cluster
• hypervisors: Set of hosts to launch virtual machines
• standalone_machine: To define only the cluster is composed with one host (replace cluster_machines)

The inventory must define these variables:

• ansible_connection: Protocol to use to connection to machine
• ansible_python_interpreter: Path to the python interpreter binary

Prerequisite

• ansible_ssh_common_args: Arguments to add for the SSH connection
• ansible_user: Login to use for the connection to machine

Playbooks

Prerequisite

When the host is installed, the ansible/playbooks/cluster_setup_prerequisdebian.yaml need to launch to finish the installationThis step is useful only for the installed hosts by a debian installer build with build_debian_iso.

The inventory must define these variables to run the step:playbook:

• admin_user: Default user with admin privileges
• admin_passwd: Password hash (optional)
• admin_ssh_keys: (optional)

• apply_network_config: Boolean to apply the network configuration

• admin_ip_addr: IP address for SNMP

• cpumachinesnort: Range of allowed CPUs for no RT machines

• cpumachines: Range of allowed CPUs for machines (RT and no RT)

• cpumachinesrt: Range of allowed CPUs for RT machines

• cpuovs: Range of allowed CPUs for OpenVSwitch

• cpusystem: Range of allowed CPUs for the system

• cpuuser: Range of allowed CPUs for the user

• irqmask: Set the IRQBALANCE_BANNED_CPUS environment variable, see irqbalance manual
• livemigration_user:
• logstash_server_ip: IP address for logstash-seapath alias in /etc/hosts

• main_disk: Main disk device to observe his temperature

• workqueuemask: The negation of the irqmask (= ~irqmask)

Network

The inventory must define these variables to run the step:

• br_rstp_priority: TODO Multiple of 4096

• cluster_ip_addr: IP address for team0 interface


• gateway_addr: IP address of a gateway, it doesn't have to work

• ip_addr: IP address to communicate with the host

• network_interface: Network interface to communicate with the host

• ntp_primary_server: Address of a NTP server, it's the first server to requests

• ntp_secondary_server: Address of a NTP server, it's the secondary server to requests

• syslog_server_ip: Address of a SYSLOG server

• team0_0: Network interface to connect to team0 bridge

• team0_1: Other network interface to connect to team0 bridge

...

In this part, the playbook define the scheduling and the prioritization (see the section).

Hardening

The ansible/playbooks/cluster_setup_hardening_debian.yaml playbook enables system hardening and the ansible/playbooks/cluster_setup_unhardening_debian.yaml playbook disables it.

The hardened elements are:

• the kernel with the parameters of the command line (see below section), the sysfs and modules;
• the GRUB;
• the systemd services;
• adding of bash profiles;
• SSH server;
• adding of sudo rules;
• the shadow password suite configuration;
• the secure tty;
• the audit daemon.

Kernel

The project uses a real-time kernel, the Linux kernel with the PREEMPT_RT patch. So, he needs to have some parameters as:

• cpufreq.default_governor=performance: Use the performance governor by default (more details here).
• hugepagesz=1G: Uses 1 giga-bytes for HugeTLB pages (more details here).
• intel_pstate=disable: Disables the intel_pstate as the default scaling driver for supported processors (more details here).
• isolcpus=nohz,domain,managed_irq: nohz to disable the tick when a single task runs; domain to isolate from the general SMP balancing and scheduling algorithms; managed_irq to isolate from being targeted by managed. See the Scheduling and priorization section.
• no_debug_object: Disables object debugging.
• nosoftlockup: Disable the soft-lockup detector (more details here).
• processors.max_cstate=1 and intel_idle.max_cstate=1: Discards of all the idle states deeper than idle state 1, for the acpi_idle and intel_idle drivers, respectively (more details here).
• rcu_nocbs: See the Scheduling and priorization section.
• rcu_nocb_poll: Make the kthreads poll for callbacks.
• rcutree.kthread_prio=10: Set the SCHED_FIFO priority of the RCU per-CPU kthreads.
• skew_tick=1: Helps to smooth jitter on systems with latency-sensitive applications running.
• tsc=reliable: Disables clocksource verification at runtime, as well as the stability checks done at bootup.

In the hardening system, the kernel has these parameters:

• init_on_alloc=1: Fill newly allocated pages and heap objects with zeroes.
• init_on_free=1: Fill freed pages and heap objects with zeroes.
• slab_nomerge: Disable merging of slabs with similar size.
• pti=on: Enable the control Page Table Isolation of user and kernel address spaces.
• slub_debug=ZF: Enable red zoning (Z) and zanity checks (F) on for all slabs (more details here).
• randomize_kstack_offset=on: Enable kernel stack offset randomization.
• slab_common.usercopy_fallback=N:
• iommu=pt: Get best performance using the SR-IOV (TODO).
• security=yama: Use the yama security module to enable at boot.
• mce=0: Disables the time in us to wait for other CPUs on machine checks.
• rng_core.default_quality=500: Set the value of the entropy for the system.
• lsm=apparmor,lockdown,capability,landlock,yama,bpf: Set the order of LSM initialization.

More details on the kernel's parameters here.

Warnings

At the end of this step, make sure that:

• Each host in the cluster pings other (simple and fragmented package)
• Hosts use a DNS server
• Hosts are synchronized with NTP server (it's necessary for the shared storage)

Shared storage (via ceph)

The inventory may define these hosts to run the step completely:

• clients: Set of hosts that must be ceph-client. These hosts will access to the storage cluster

• mons: Set of hosts to that must be ceph-mon. These hosts will maintain a map of the state of the cluster
  

• osds: Set of hosts to that must be ceph-osd. These hosts will interact with the logical disk to stock the data

More details in the documentation here.

The inventory must define these variables to run the step:

• ceph_cluster_network: Address block to access to cluster network
• ceph_public_network: Address block to access to public network (ie to the world)

• ceph_osd_disk: Device to stock datas (only for ceph-osd hosts)

• osd_pool_default_min_size: Minimal number of available OSD to ensure cluster success (best: ceil(osd_pool_default_size / 2.0))

• osd_pool_default_size: Number of OSD in the cluster

...

Ceph provides ansible rules to configure the software, you can read the documentation here.

Warnings

If this step is failed, you must restart the step at the previous step. Use snapshot LVM to do this.

...