Skip to content

Latest commit

 

History

History
416 lines (320 loc) · 12.8 KB

README.md

File metadata and controls

416 lines (320 loc) · 12.8 KB

Catan

Catan is a charm relation interface integration testing sdk. Let's break it down:

  • It's a testing library.
  • It's testing framework-agnostic.
  • It's built on top of ops-scenario.
  • It's mainly meant for testing the integration between different charms as they communicate over relation data.
  • It's about integration testing the charms (and the way they interact with the juju model): NOT the workloads and NOT the substrate they are running on.

Catan is a juju model-level simulation tool.

  • It allows you to simulate a single juju model.
  • It doesn't simulate the workloads, the cloud substrate: only The Juju Model of things:
    • What applications are there
    • What units they have
    • How the apps are integrated
  • It simulates certain dynamic aspects of the model lifecycle too:
    • The admin runs an action
    • The admin deploys a new app, or scales an existing one up/down
    • The admin integrates/disintegrates two endpoints

Why not scenario?

Scenario operates at the level of the single charm unit execution: one charm instance, one juju event at a time. In order to practically verify the interaction between two charms, one has to continually manually play the role of the 'remote' unit and mock the relation data it would present or reply with in a given interaction. This often means hardcoding that data in the tests, with predictable consequences.

Catan, by contrast, operates at the level of multiple related units, one cloud admin action at a time. In scenario, you look at what happens if nginx/0 receives a foo-relation-departed event while in state X. In Catan, given nginx/0 is in state X and mydb/1 is in state Y, you look at what happens if the cloud admin does juju integrate nginx:foo mydb:bar.

Data structures

ModelState

While the primary data structure you play with in Scenario is the State (which in the context of Catan we should probably refer to as "the Unit State"), in Catan the protagonist is the ModelState, which plays a similar role. The ModelState data structure encapsulates:

  • The list of Apps that are present in the model.
  • For each App:
    • The unit IDs and scenario.States of each individual unit of the app.
  • The list of Integrations present in the model.

Usage:

from scenario import State

from catan import ModelState, App, Integration, Catan

app1 = App(...)
app2 = App(...)

foo_bar_integration = Integration.from_endpoints(app1, "foo", app2, "bar")
ms = ModelState(
    unit_states={
        app1: {0: State()},
        app2: {0: State()},
    },
    integrations=[
        foo_bar_integration
    ]
)

c = Catan(ms)
# do things with catan, such as
ms_out: ModelState = c.disintegrate(foo_bar_integration)

# the integration is gone from the model state:
assert not ms_out.integrations

# and catan has queued relation-departed, relation-broken events on all affected units.
assert c._event_queue

# execute all queued events
c.settle()

App

The App data structure encapsulates:

  • App name, such as "nginx"
  • Charm source and metadata (yes, a physical charm's source code)

Usage:

from ops import CharmBase
from scenario import State
from catan import ModelState, App, Catan


class MyCharm(CharmBase):
    ...


app1 = App.from_git("canonical", "traefik-k8s")
app2 = App.from_type(MyCharm, "mycharm")
app3 = App.from_path("/path/to/local/charm/repo", "local-charm")

# use it to declaratively set up a model...
ms = ModelState(
    unit_states={
        app1: {0: State()},
        app2: {0: State()},
        app3: {0: State()},
    }
)

# ... or use it to imperatively do the same
c = Catan()
c.deploy(app1)
c.deploy(app2)
c.deploy(app3)

Integration

The Integration data structure encapsulates:

  • Two Apps and the endpoints by which they are integrated.
from scenario import State
from catan import ModelState, App, Integration, Catan

app1 = App(...)
app2 = App(...)

# use it to declaratively set up a model...
ms = ModelState(
    unit_states={
        app1: {0: State()},
        app2: {0: State()},
    },
    integrations=[
        Integration.from_endpoints(app1, "foo", app2, "bar")
    ]
)

# ...or use it to imperatively do the same
c = Catan()
c.deploy(app1)
c.deploy(app2)
c.integrate(app1, "foo", app2, "bar")

Using Catan

Core concepts

A Catan test will generally consist of these three broad steps:

  • Arrange:
    • Set up the ModelState by declaring what Apps there are, how they are related, how many units of each there are and what State each unit is in.
  • Act:
    • Declare a change of the ModelState, for example, by:
      • Adding/removing an integration
      • Adding/removing an app
      • Scaling up/down an app
      • Running an action on one of the units
      • Triggering manually an event on one or more units
      • (TODO): causing a secret to expire
    • Let Catan.settle() which means flush the event queue and keep doing that until it's empty (emitting an event might put more events in the queue!)
  • Assert:
    • Verify that the ModelState you obtain back from Catan is what you expect it to be, for example:
      • Check that a unit is in a specific state
      • Check that a given event was fired on a unit
      • Check that a new secret revision was published by its owner, or viewed by its observer
      • Check that the relations involved in an Integration contain the data you expect

Example

from unittest.mock import patch
from scenario import State
from catan import Catan, ModelState, App


def test_integrate():
    # GIVEN: 
    # - the tempo and traefik applications
    tempo = App.from_path(
        "/path/to/tempo-k8s-operator/",
        patches=[
            patch("charm.KubernetesServicePatch")
        ])
    traefik = App.from_git(
        "canonical",
        "traefik-k8s-operator",
        patches=[
            patch("charm.KubernetesServicePatch")
        ])
    
    c = Catan(ModelState({
        tempo: {
            0: State(leader=True),
            1: State(),
        },
        traefik: {0: State(leader=True)}
    }))
    # WHEN: 
    # - we simulate doing `juju relate tempo:tracing traefik:tracing`
    c.integrate(tempo, "tracing", traefik, "tracing")
    
    # we tell Catan to flush the event queue and keep running until it's empty
    # output is the model state in its final form
    ms: ModelState = c.settle()
    
    # we can inspect what has been emitted by catan
    assert c._emitted_repr == [
        # this is the initial event sequence, programmed by juju
        'tempo/0 :: tracing_relation_created',
        'tempo/1 :: tracing_relation_created',
        'traefik/0 :: tracing_relation_created',
        'tempo/0 :: tracing_relation_joined',
        'tempo/1 :: tracing_relation_joined',
        'traefik/0 :: tracing_relation_joined',
        'tempo/0 :: tracing_relation_changed',
        'tempo/1 :: tracing_relation_changed',
        'traefik/0 :: tracing_relation_changed',
        
        # tempo notices traefik has made databag changes
        'tempo/0 :: tracing_relation_changed',
        'tempo/1 :: tracing_relation_changed',

        # traefik notices tempo has made databag changes
        'traefik/0 :: tracing_relation_changed',
        'traefik/0 :: tracing_relation_changed'
        
        # it could go on longer for multi-step interface protocols
    ]
    traefik_tracing_out = ms.unit_states[traefik][0].get_relations('tracing')[0]
    assert traefik_tracing_out.remote_app_data

The event queue

Catan is all about managing an event queue and keeping the several scenario States in sync with one another every time a charm executes.

Much of the Catan API is about helping you to populate the event queue in a way that makes sense, while keeping the ModelState consistent with the history you're trying to tell.

Getting started

You can instantiate Catan with an empty ModelState. This means that there are no apps and no integrations.

import catan
c = catan.Catan()

Next you can mutate the model state.

Deploying apps

import catan
# you can inspect the return object to view what's in the `ModelState` at this point.
ms: catan.ModelState = c.deploy(catan.App.from_path("/path/to/charm", name="foo"), [0,1], ...)

this is going to add two units of "foo": foo/0 and foo/1 to the ModelState, and queue a full setup sequence for both:

  • *-storage-attached (todo)
  • install
  • leader-elected on the leader unit, leader-settings-changed on the followers
  • config-changed
  • start

Adding units

import catan, scenario
app = catan.App.from_path("/path/to/charm", name="foo")
# this app has scale zero
ms = catan.ModelState(unit_states={app: {}})
catan.Catan(ms).add_unit(app, 3, state=scenario.State(leader=True))  # adds foo/3

this is going to add "foo/3" to the input ModelState, and queue a full setup sequence for that unit:

Removing units

import catan, scenario
app = catan.App.from_path("/path/to/charm", name="foo")
# this app has scale zero
ms = catan.ModelState(unit_states={app: {
  1: scenario.State(leader=False), # foo/1
  2: scenario.State(leader=True), # foo/2
}})
catan.Catan(ms).remove_unit(app, 1)  # kills foo/1

this is going to remove "foo/1" from the input ModelState, and queue a full teardown sequence for that unit:

  • (todo) storage-detached for all storages
  • stop
  • remove

as well as a leader-elected on foo/1, since foo/2 was the leader!

If the app had relations, we'd also see

  • (todo) relation-departed for peer relations
  • relation-departed + relation-broken for regular relations

and all remote units would also see a relation-departed for foo/1.

Removing apps

import catan, scenario
app = catan.App.from_path("/path/to/charm", name="foo")
# this app has scale zero
ms = catan.ModelState(unit_states={app: {
  1: scenario.State(leader=False), # foo/1
  2: scenario.State(leader=True), # foo/2
}})
catan.Catan(ms).remove_app(app)

this is going to remove "foo/1" and "foo/2" from the input ModelState, and queue a full teardown sequence for both units.

If the app had relations, we'd also see the expected departed/broken hooks.

Adding integrations

You can define your input ModelState to already have an integration:

from catan import Catan, ModelState, Integration, Binding
from scenario import State
c = Catan(
    ModelState(
        {
            app1: {0: State(leader=True)},
            app2: {0: State(leader=True)},
        },
      integrations=[
        Integration(
          Binding(app1, "tracing"),
          Binding(app2, "tracing"),
          )
      ]
    )
)

or you can add one and queue the corresponding events:

from catan import Catan, ModelState
from scenario import State
c = Catan(
    ModelState(
        {
            app1: {0: State(leader=True)},
            app2: {0: State(leader=True)},
        }
    )
)

# juju relate app1:tracing app2:tracing
c.integrate(app1, "tracing", app2, "tracing")

This would queue:

  • on all app1 units:
    • tracing-relation-created
    • for all app2 units:
      • tracing-relation-joined
    • tracing-relation-changed
  • on all app2 units:
    • tracing-relation-created
    • for all app1 units:
      • tracing-relation-joined
    • tracing-relation-changed

Typically, on relation-changed events, a charm can write data to their side of the relation. Catan will notice this and queue additional relation-changed events on the remote units.

So usually you'll see a back-and-forth of relation-changed events until the charms settle and stop reacting to one another's writes, depending on the protocol.

Running actions

Running actions is done via the run_action API.

from catan import Catan, App
from scenario import Action

c = Catan()
app1 = App.from_git("canonical", "tempo-k8s")
c.deploy(app1)

c.run_action("do-something", app1)  # on all units
c.run_action("do-something", app1, 0)  # on app1/0
c.run_action(Action("do-something", params={"foo": "bar"}), app1, 0)  # with parameters

Randomization

After you've populated the event queue, you can call Catan.shuffle() to randomize it in a way that still makes juju-sense. For example, a start event should not precede an install event. Catan.shuffle() ensures that the event sequences can interleave with other sequences, while their internal relative ordering remains intact.

Events that are not part of a sequence can be shuffled around anywhere in the queue.

If you are manually queuing events, and you want to declare them as a sequence, you can use the Catan.fixed_sequence API:

import catan
c = catan.Catan()
with c.fixed_sequence():
  c.queue("update-status", app, 0)
  c.queue("stop", app, 0)

This will make sure that, if you do c.shuffle(), the relative position of stop relative to update-status will remain unchanged in the queue.