Network spaces

Overview

This document provides a conceptual overview of the role that Juju network spaces play in the deployment and management of Charmed OpenStack.

A network space is a Juju object that abstracts away the OSI Layer 3 network concepts. A space is created on the backing cloud to represent one or more subnets. The purpose of a space is to allow for user-defined network traffic segmentation, which can have various uses (e.g. traffic isolation as a security measure).

Note

See the Juju documentation to fully explore spaces themselves.

Spaces support

For Charmed OpenStack, the recommended Juju backing cloud is MAAS. Spaces can be created and configured within MAAS and the Juju operator can immediately make use of them when deploying or configuring OpenStack applications.

This is a screenshot of the MAAS web UI showing an example spaces configuration:

Spaces in the MAAS web UI


In the above, we see that VLANs are used to associate spaces with subnets. It is also clear that multi-subnet spaces are supported.

Bindings

A space is associated with an application’s traffic type via the latter’s corresponding “endpoint”. This association is known as a “binding”. This specific traffic flow is established once the endpoint is “integrated” with another application’s endpoint (the juju integrate command).

Note

Integration is the process of connecting compatible endpoints (same interface type) between two applications. Endpoints are discerned from a charm’s metadata.yaml file.

Extra bindings

Juju has the notion of an “extra binding”. Like a normal binding, an extra binding requests, via spaces, that a machine be provisioned with network connectivity to specific subnets. The intention is that the charm that supports the binding will then configure its payload service, whose consumer may be external to Juju, to listen on a certain IP address(es). This type of binding is not dependent upon an integration with another application’s endpoint.

Note

Extra bindings are explicitly declared in a charm’s metadata.yaml file.

Default space

A default value for spaces can be set at the model level and at the application level. This value must be known to Juju (see the juju list-spaces command).

A model’s default space is defined with the juju model-config command. If left unset, the system default of ‘alpha’ is used.

juju model-config -m <model> default-space=<space>

Note

The juju model-default command can be used to affect all future models.

A space is defined at the application level at deploy time (juju deploy). This becomes the application’s de facto default space. That is, any scale out operation (juju add-unit) will inherit this value and any further integrations for the application will also use this value.

Note

It has been reported that the default space at the model level does not work for LXD containers. This issue is being tracked in LP #1883164.

Deploying to a space

When deploying an application, the --bind option is used with the juju deploy command to explicitly select a space:

juju deploy --bind <space> <charm>

The default space will be attempted if no space is specified. However the absence of a machine attached to system default space (‘alpha’) will not prevent a machine from being randomly provisioned.

Spaces as a constraint

With the MAAS provider (backing cloud) the deployment of the initial unit can refer to a space as a Juju constraint. This simply ensures that the resulting machine has a network interface on a certain subnet(s). It is not used by the charm to implement any network functionality, as does a normal or extra binding:

juju deploy --constraints spaces=<space> <charm>

Such a constraint can be used in combination with a binding (using different space values naturally):

juju deploy --constraints spaces=<space-1> --bind <space-2> <charm>

LXD containers

There is a subtle distinction between applying space constraints to LXD containers as opposed to bare metal or KVM-based MAAS nodes. For the latter types, a constraint represents a network criterion, whereas for dynamically-provisioned containers, a constraint acts as a request to add network connectivity. The LXD host must also be able to satisfy that request in terms of its own network connectivity.

Subordinate applications

A subordinate application always runs alongside a principal application and is only actually deployed when an integration is formed with its principal. Owing to this behaviour, the chosen bindings of a subordinate application must be a subset of the bindings of the principal application.

Multi-subnet spaces and routing

A binding (or extra binding) should be seen as a request to allocate traffic to certain subnets if possible, and not as a mechanism that alters or verifies routing at the L3 level. In the case of a multi-subnet space, this implies that network connectivity among those subnets must exist externally to Juju. There are two key scenarios to watch out for when a multi-subnet space is involved:

  • multi-unit applications

  • any two integrated applications

In each of the above scenarios, inter-subnet routing must be configured in advance to allow for the application units to reach one another. In other words, all subnets/VLANs within the same space are assumed to have unrestricted traffic. It should also be noted that an integration formed between two applications is not indicative of successful routing.

Working with integrations and bindings

This section will apply many of the learned concepts to an example scenario: that of deploying Vault and connecting it to the cloud database as described in the Vault section of the Deploy Guide.

The metadata.yaml file

Inspecting the metadata.yaml file for charms allows us to better understand how integrations and binding are formed. The pertinent contents of this file for the vault, mysql-router, and mysql-innodb-cluster charms will be shown. The following YAML keys are of interest:

  • extra-bindings

  • provides

  • requires

This results in the following excerpts for each charm:

 1extra-bindings:
 2  access:
 3  external:
 4
 5provides:
 6  nrpe-external-master:
 7    interface: nrpe-external-master
 8    scope: container
 9  secrets:
10    interface: vault-kv
11  certificates:
12    interface: tls-certificates
13
14requires:
15  db:
16    interface: pgsql
17  shared-db:
18    interface: mysql-shared
19  etcd:
20    interface: etcd
21  ha:
22    interface: hacluster
23    scope: container
24  lb-provider:
25    interface: loadbalancer
26    limit: 1

Manual commands

This section shows example commands that can be used to manually achieve the example scenario’s objective.

It is assumed that:

  • there are two spaces available to the MAAS cluster (‘public-space’ and ‘access-space’)

  • a default space is not configured at the model level

  • a cloud database is pre-existing (the mysql-innodb-cluster application)

To deploy the vault application and set its default space to ‘public-space’:

juju deploy --bind public-space --to lxd:3 --channel 1.8/stable vault

To deploy the mysql-router application:

juju deploy --channel 8.0/stable mysql-router vault-mysql-router

To integrate the db-router endpoint of the vault-mysql-router application (lines 10-11 of the mysql-router excerpt) with the db-router endpoint of the mysql-innodb-cluster (lines 4-5 of the mysql-innodb-cluster excerpt) application:

juju integrate vault-mysql-router:db-router mysql-innodb-cluster:db-router

To integrate the shared-db endpoint of the vault application (lines 17-18 of the vault excerpt) with the shared-db endpoint of the vault-mysql-router (lines 2-3 of the mysql-router excerpt) application:

juju integrate vault-mysql-router:shared-db vault:shared-db

The shared-db endpoint traffic will now be confined to space ‘public-space’.

Note

Regardless of the common name of the two endpoints in each of the above integrations (endpoints are generally different), it is technically the common interface type that makes an integration possible. For the first integration it is mysql-router, and for the second it is mysql-shared.

To form a binding between the access endpoint (line 2 of the vault excerpt) and the ‘access-space’ space:

juju bind vault access=access-space

An integration is not necessary for the access endpoint due to it being associated with an extra binding.

The corresponding LXD host (Vault has been deployed to a container) must have connectivity to the ‘access-space’ space in order for this to work.

The bundle method

The entire example deployment can be expressed in a bundle file and then used to fulfil the goal of the example scenario. Let this file be named vault-to-database.yaml:

machines:

  '3':

applications:

  vault-mysql-router:
    charm: mysql-router
    channel: 8.0/stable

  vault:
    charm: vault
    channel: 1.8/stable
    num_units: 1
    bindings:
      "": public-space
      access: access-space
    to:
    - lxd:3

relations:

- - vault-mysql-router:db-router
  - mysql-innodb-cluster:db-router
- - vault:shared-db
  - vault-mysql-router:shared-db

To use the bundle:

juju deploy ./vault-to-database.yaml