LinK
Link is not Keepalived
LinK is a networking agent that will let multiple hosts share a virtual IP. It
chooses which host must bind this IP and inform other members of the
network of the host owning this IP.
The IP owner election is performed using etcd lease system and other hosts on
this network is informed of the current IP owner using gratuitous ARP
requests (see How do we bind IPs?).
To ease the cluster administration, LinK comes with it's
own CLI.
Demo
Project goals
- KISS: our goal is to follow the UNIX philosophy: "Do one thing and do it
well". This component is only responsible of the IP attribution part. It
will not manage load balancing or other higher level stuff.
- If an IP is registered on the cluster there must always be at least one
server that binds the IP
Architecture
** No central manager** Each agent only have knowledge of their local
configuration. They do not know nor care if other IP exists or if other hosts
have the same IP configured. The synchronization is done by creating locks in
etcd.
** Fault resilience** If for any reason something went wrong (lost connection
with etcd) LinK will always try to have at least one host this means that
if one agent fails to contact the etcd cluster it will take the IP.
Installation
In order to be able to run LinK, you must have a working etcd cluster.
Installation and configuration instructions are available on the etcd
website.
LinK uses etcd v3 API. So you'll need etcd version 3.0.0 or higher.
The easiest way to get LinK up and running is to use pre-build binary available
on the release pages.
State machine
Each LinK agent can be in any of these three states:
ACTIVATED
: This machine owns the virtual IP
STANDBY
: This machine does not own the virtual IP but is available for election
FAILING
: Health checks for this host failed, this machine is not available for election
At any point five types of events can happen:
fault
: There was some error when coordinating with other nodes.
elected
: This machine was elected to own the virtual IP.
demoted
: This machine just lost ownership of the virtual IP.
health_check_fail
: The health checks configured with this IP failed.
health_check_success
: The health checks configured with this IP succeeded.
This is what the state machine looks like:
Configuration
LinK configuration is entirely done by setting environment variables.
INTERFACE
: Name of the interface where LinK should add and remove IPs.
USER
: Username used for basic auth
PASSWORD
: Password used for basic auth
PORT
(default: 1313): Port where the LinK HTTP interface will be available
ETCD_HOSTS
: The different endpoints of etcd members
ETCD_TLS_CERT
: Path to the TLS X.509 certificate
ETCD_TLS_KEY
: Path to the private key authenticating the certificate
ETCD_CACERT
: Path to the CA cert signing the etcd member certificates
Endpoints
GET /ips
: List all currently configured IPs
POST /ips
: Add an IP
GET /ips/:id
: Get a single IP
DELETE /ips/:id
: Remove an IP
POST /ips/:id/lock
: Try to get the lock on this IP
How do we bind the IPs?
To add an interface LinK adds the IP to the configured interface and send an
unsolicited ARP request on the network (see Gratuitous
ARP).
This is the equivalent of:
ip addr add MY_IP dev MY_INTERFACE
arping -B -S MY_IP -I MY_INTERFACE
To unbind an IP we will just remove it from the interface.
This is the equivalent of:
ip addr del MY_IP dev MY_INTERFACE
Dev environment
To make it work in dev you might want to make some dummy interfaces:
modprobe dummy
ip link add eth10 type dummy
ip link set eth10 up
ip link add eth11 type dummy
ip link set eth11 up
ip link add eth12 type dummy
ip link set eth12 up
The script start.sh
can be executed as root to automatically do that.