Service ip resources

Introduction

A service must expose its middleware to other services and to end users through ip address. It is possible to expose through the service instance's node ip addresses, or through one or more ip addresses handled by the service.

Failover services should prefer exposing through service-private ip addresses, so that:

  • relocating the service running instance on another node in the same network retains the same ip addresses.
  • the service ip addresses can be used to avoid starting multiple instances simultaneously

Even flex services can benefit from private ip addresses:

  • middlewares of different services running on the same node can bind the same port without conflict

This chapter presents the different ip management capabilities offered by the OpenSVC agent.

Failover

In this use-case, the middlewares are installed in the node either through operating system packages or as service-private binaries. They bind the service ip address.

_images/agent.service.ip.failover.png

Example non-containerized ip resource definition:

[ip#0]
ipname@nodes = ipname1
ipname@drpnodes = ipname2
ipdev = eth0

Flex with instance ip

In this use-case, the middlewares are installed in the node either through operating system packages or as service-private binaries. They bind each service instance ip addresses.

_images/agent.service.ip.flex.png

Example non-containerized ip resource definition:

[ip#0]
ipname@node1 = ipname1
ipname@node2 = ipname2
ipname@node3 = ipname3
ipdev = eth0

Failover with application installed in a container

In this use-case, the middlewares are installed in the container through operating system packages. They bind the container ip address.

_images/agent.service.ip.failover.container.png

Example configuration:

[container#0]
name = vm1

Failover with application installed in a container with service encapsulation

In this use-case, the middlewares are installed in the container through operating system packages or as service-private binaries. They bind the encapsulated service ip address. The service containers are set to start on each node just to highlight the fact that each container can have its own ip addresses. This can also be useful if you want sync resources at the encapsulated layer, as standby instance containers have to be running to receive the replication data.

_images/agent.service.ip.failover.encap.png

Example configuration:

[container#0]
name = vm1
always_on = nodes drpnodes

[ip#0]
ipname = ipname1
ipdev = eth0
tags = encap

Failover with application installed as micro-containers

In this use-case, the middlewares are installed as docker micro-containers. The micro-containers share a common namespace through the --net run parameter. The service plumbs a public ip in this namespace. Exposed middlewares bind this to ip. Non exposed middlewares can still bind the loopback ip 127.0.0.1 to service the other middlewares.

_images/agent.service.ip.failover.docker.png

Example configuration:

[DEFAULT]
nodes = n1 n2
drpnodes = n3
docker_data_dir = /srv/svc1/docker_data_dir

[ip#0]
type = docker
ipdev = eth0
ipname = 10.0.3.3
netmask = 255.255.255.0
gateway = 10.0.3.1
container_rid = container#0

[container#0]
type = docker
run_image = ubuntu:14.10
run_args = --net=none
           -v /etc/localtime:/etc/localtime:ro
run_command = /bin/bash

[container#1]
type = docker
run_image = opensvc/nginx:build5
run_args = -v /etc/localtime:/etc/localtime:ro
           --net=container:svc1.container.0