Interface layers

Interface layers are responsible for the communication that transpires over a relation between two applications. This type of layer encapsulates a single "interface protocol" and is generally written and maintained by the author of the primary charm that provides that interface. However, it does cover both sides (provides and requires) of the relation and turns the two-way key-value store that are Juju relations under-the-hood into a full-fledged API for interacting with charms supporting that interface.

It is important to note that interface layers do not actually implement either side of the relation. Instead, they are solely responsible for the communication that goes on over the relation, relying on charms on either end to decide what to do with the results of that communication.

Interface layers currently must be written in Python and extend the Endpoint class, though they can then be used by any language using the built-in CLI API.


Historically, the term "relation" has been used ambiguously when discussing Juju, charms, and applications deployed with Juju using charms. To make things more clear and consistent, we use the following terms:

  • A charm is the set of code that encapsulates the operational knowledge that manages the life-cycle of an application when deployed.

  • An application is a deployed instance of a charm. It is hosted on a machine, VM, or container, it has a copy of the charm code, and it is managed by a Juju controller.

  • The interface is the protocol used by two charms to communicate over a relation between them. This protocol is what interface layers codify.

  • An endpoint is one charm's connection point at which a relation can be made. An endpoint is defined in a charm's metadata.yaml and specify a name, a role (one of requires, provides, or peers), and an interface that relations established at the endpoint will use.

  • A relation is an established connection between endpoints of two applications. Relations have a relation ID that is tracked by the Juju controller and relation data associated with each unit of each application involved in the relation. Relations can only be established between endpoints that specify the same interface, and a given endpoint can only have one relation to a specific application. However, a given application can be related to an application any number of times on different endpoints, as long as they specify the same interface, and even to itself if it has two endpoints of the same interface (or if the endpoint is of the type peers).

Here's how these terms map to a charm's metadata.yaml and layer.yaml file.

# metadata.yaml
name: mattermost  # Charm name
# ...
requires:  # Endpoint role
    postgres:  # Endpoint name
        interface: pgsql  # Interface name
provides:  # Endpoint role
    website:  # Endpoint name
        interface: http  # Interface name
# ...
# layer.yaml
 - 'layer:basic'
   # ...
 - 'interface:pgsql'  # This interface layer codifies the `pgsql` protocol. The
                      # `requires` role of this interface layer will by used by
                      # the `postgres` endpoint.

 - 'interface:http'  # This interface layer codifies the `http` protocol. The
                     # `website` endpoint will use the `provides` role of this
                     # interface.
# ...

Design considerations

When writing an interface, there is a small amount of pre-planning into what that interface should look like in terms of the communication between applications/unit(s) participating in the relationship.

Common questions to answer:

  • What units need to participate in the conversation?
  • What data is being sent on the wire?
  • Are we sending data that is essentially static to any application connecting over this interface?
  • How should this data be made available to the requirer?
  • What flags should this interface set on the provider?
  • What flags should this interface set on the requirer?

For the last couple of considerations, it's important to note that the Endpoint base class will automatically manage a few flags that are common to all interfaces:

  • endpoint.{endpoint_name}.joined Set whenever any remote unit joins
  • endpoint.{endpoint_name}.changed Set whenever any relation data changes
  • endpoint.{endpoint_name}.changed.{field} Set for each field that changes
  • endpoint.{endpoint_name}.departed Set whenever any remote unit leaves

These are mainly intended to be used by the interface layer itself, but can be documented as part of the official API that the interface layer exposes. Additionally, the joined flag will be automatically cleared when there are no remote units remaining on any relation, but neither of the other flags will be automatically cleared.

Creating an interface layer

First off, you require a local charm repository in which to work. This involves creating three directories -- layers, interfaces, and charms -- and setting some environment variables.

The layers directory contains the source code of the layered charm covered in our examples. The interfaces directory is where you'd place any interface-layers you may wish to write, and the charms directory holds the assembled, ready to deploy charm.



Note: Exporting the environment variables in this way only sets the variables for the current terminal. If you wish to make these changes persist, add the same export statements to a resource file that are evaluated when you create a new console such as ~/.bashrc depending on your shell.

The export of INTERFACE_PATH is an environment variable which tells the charm build process where to scan for local interfaces not found in the layer registry.

With our interface repository created, we can now create our new interface.

Next, create the directory to warehouse your interface.

mkdir -p $INTERFACE_PATH/http

And declare the interface's metadata in interface.yaml.

name: http
summary: Basic HTTP interface
version: 1

The HTTP interface has two roles: the provides side is for an http-accessible webservice and the requires is for an application that uses the webservice. In this example, the provides side is a website and the requires side is a reverseproxy.

Creating the provides role

We're now ready to implement the provides interface in

from charmhelpers.core import hookenv
from charms.reactive import set_flag, clear_flag
from charms.reactive import Endpoint

class HttpProvides(Endpoint):
    def publish_info(self, port, hostname=None):
        Publish the port and hostname of the website over the relationship so
        it is accessible to the remote units at the other side of the

        If no hostname is provided, the unit's private-address is used.
        for relation in self.relations:
            relation.to_publish['hostname'] = hostname or hookenv.unit_get('private_address')
            # Publishing data with `to_publish` is the only way to communicate
            # with remote units. Flags are local-only, they are not shared with
            # remote units!
            relation.to_publish['port'] = port



Now we can use this provides endpoint interface in our charm. The first step is to define the relation using this interface in metadata.yaml.

provides:  # This side of the relationship implements the `provides` role
  website:  # We call our endpoint `website`. All flags of this endpoint will
            # be prefixed with ``.
    interface: http  # The `website` endpoint uses the `provides`
                     # side of the `http` interface

Now we can create handlers that react to the flags set by the endpoint. These handlers are part of a charm that deploys a webservice. The following handler publishes the information about the webservice after the webservice started and a relation is established with a charm that wants to use the webservice.

from charms.reactive import context

# `website` is the endpoint name as defined in `metadata.yaml`
def publish_website_info():
    # Retrieve the Endpoint object. This object will be an instance of the
    # HttpProvides class defined above.
    website = endpoint_from_flag('')

Creating the requires role

We're now ready to implement the requires role of this interface in

from charms.reactive import when_any, when_not
from charms.reactive import set_flag, clear_flag
from charms.reactive import Endpoint

class HttpRequires(Endpoint):
    # {endpoint_name} will be filled in by the reactive framework. This is the
    # name of the endpoint as defined in `metadata.yaml`
    def new_website(self):
        # Detect changes to the hostname or port field on any remote unit
        # and translate that into the new-website flag. Then, clear the
        # changed field flags so that we can detect further changes.

    def broken(self):

    def websites(self):
        Get the list of websites that remote units have published over all the
        relationships connected to this endpoint.

        Returns a list of dicts, where each dict contains the hostname (address)
        and the port (as a string) that the website is listening on, as well as
        the relation ID and remote unit name that provided the site.

        For example::
                    'hostname': '',
                    'port': '80',
                    'relation_id': 'reverseproxy:1',
                    'remote_unit_name': 'myblog/0',
        websites = []
        # Multiple relations can connect to the same endpoint. All relations
        # of the same endpoint will be handled by the same Endpoint
        # class.
        # Loop over all units of all relations connected to this enpoint,
        # read the port and hostname they published, add them to a dict and
        # return that information.
        for relation in self.relations:
            for unit in relation.units:
                hostname = unit.received['hostname']
                port = unit.received['port']
                if not (hostname and port):
                    'hostname': hostname,
                    'port': port,
                    'relation_id': relation.relation_id,
                    'remote_unit_name': unit.unit_name,
        return websites


Now we can use this requires endpoint interface in our charm. The first step is to define the relation using this interface in metadata.yaml.

    interface: http

Now we can create handlers that react to the flags set by the endpoint. This handler is part of a charm that deploys a reverseproxy. The handler logs the published information of all the connected webservices.

from charms.reactive import when
from charms.reactive import clear_flag
from charms.reactive import context

# `reverseproxy` is the endpoint name as defined in `metadata.yaml`
def update_reverse_proxy_config():
    # Retrieve the Endpoint object. This object will be an instance of the
    # HttpRequires class defined above.
    reverseproxy = endpoint_from_flag('')
    # Note that `reverseproxy.websites()` returns _all_ websites, not just
    # the website that triggered the `` flag.
    for website in reverseproxy.websites():
        hookenv.log('Website: {}:{}'.format(
        # ...