| Name | ops-scenario JSON |
| Version |
6.0.3
JSON |
| download |
| home_page | |
| Summary | Python library providing a state-transition testing API for Operator Framework charms. |
| upload_time | 2024-03-19 10:40:32 |
| maintainer | |
| docs_url | None |
| author | |
| requires_python | >=3.8 |
| license | Apache-2.0 |
| keywords |
juju
test
|
| VCS |
 |
| bugtrack_url |
|
| requirements |
No requirements were recorded.
|
| Travis-CI |
No Travis.
|
| coveralls test coverage |
No coveralls.
|
# Scenario
[](https://github.com/canonical/ops-scenario/actions/workflows/build_wheels.yaml)
[](https://github.com/canonical/ops-scenario/actions/workflows/quality_checks.yaml?event=pull_request)
[](https://discourse.charmhub.io)
[](https://github.com/PietroPasotti/jhack)
[](https://discourse.charmhub.io/t/rethinking-charm-testing-with-ops-scenario/8649)
[](https://www.python.org/downloads/release/python-380/)
Scenario is a state-transition testing SDK for Operator Framework charms.
Where the Harness enables you to procedurally mock pieces of the state the charm needs to function, Scenario tests allow
you to declaratively define the state all at once, and use it as a sort of context against which you can fire a single
event on the charm and execute its logic.
This puts scenario tests somewhere in between unit and integration tests: some say 'functional', some say 'contract', I prefer 'state-transition'.
Scenario tests nudge you into thinking of a charm as an input->output function. The input is the
union of an `Event` (why am I, charm, being executed), a `State` (am I leader? what is my relation data? what is my
config?...) and the charm's execution `Context` (what relations can I have? what containers can I have?...). The output is another `State`: the state after the charm has had a chance to interact with the
mocked juju model and affect the initial state back.
For example: a charm currently in `unknown` status is executed with a `start` event, and based on whether it has leadership or not (according to its input state), it will decide to set `active` or `blocked` status (which will be reflected in the output state).
Scenario-testing a charm, then, means verifying that:
- the charm does not raise uncaught exceptions while handling the event
- the output state (or the diff with the input state) is as expected.
# Core concepts as a metaphor
I like metaphors, so here we go:
- There is a theatre stage.
- You pick an actor (a Charm) to put on the stage. Not just any actor: an improv one.
- You arrange the stage with content that the actor will have to interact with. This consists of selecting:
- An initial situation (`State`) in which the actor is, e.g. is the actor the main role or an NPC (`is_leader`), or what
other actors are there around it, what is written in those pebble-shaped books on the table?
- Something that has just happened (an `Event`) and to which the actor has to react (e.g. one of the NPCs leaves the
stage (`relation-departed`), or the content of one of the books changes).
- How the actor will react to the event will have an impact on the context: e.g. the actor might knock over a table (a
container), or write something down into one of the books.
# Core concepts not as a metaphor
Scenario tests are about running assertions on atomic state transitions treating the charm being tested like a black
box. An initial state goes in, an event occurs (say, `'start'`) and a new state comes out. Scenario tests are about
validating the transition, that is, consistency-checking the delta between the two states, and verifying the charm
author's expectations.
Comparing scenario tests with `Harness` tests:
- Harness exposes an imperative API: the user is expected to call methods on the Harness driving it to the desired
state, then verify its validity by calling charm methods or inspecting the raw data. In contrast, Scenario is declarative. You fully specify an initial state, an execution context and an event, then you run the charm and inspect the results.
- Harness instantiates the charm once, then allows you to fire multiple events on the charm, which is breeding ground
for subtle bugs. Scenario tests are centered around testing single state transitions, that is, one event at a time.
This ensures that the execution environment is as clean as possible (for a unit test).
- Harness maintains a model of the juju Model, which is a maintenance burden and adds complexity. Scenario mocks at the
level of hook tools and stores all mocking data in a monolithic data structure (the State), which makes it more
lightweight and portable.
# Writing scenario tests
A scenario test consists of three broad steps:
- **Arrange**:
- declare the context
- declare the input state
- select an event to fire
- **Act**:
- run the context (i.e. obtain the output state, given the input state and the event)
- **Assert**:
- verify that the output state (or the delta with the input state) is how you expect it to be
- verify that the charm has seen a certain sequence of statuses, events, and `juju-log` calls
- optionally, you can use a context manager to get a hold of the charm instance and run assertions on internal APIs and the internal state of the charm and operator framework.
The most basic scenario is one in which all is defaulted and barely any data is
available. The charm has no config, no relations, no leadership, and its status is `unknown`.
With that, we can write the simplest possible scenario test:
```python
from scenario import State, Context, Event
from ops.charm import CharmBase
from ops.model import UnknownStatus
class MyCharm(CharmBase):
pass
def test_scenario_base():
ctx = Context(MyCharm, meta={"name": "foo"})
out = ctx.run(Event("start"), State())
assert out.unit_status == UnknownStatus()
```
Now let's start making it more complicated. Our charm sets a special state if it has leadership on 'start':
```python
import pytest
from scenario import State, Context
from ops.charm import CharmBase
from ops.model import ActiveStatus
class MyCharm(CharmBase):
def __init__(self, ...):
self.framework.observe(self.on.start, self._on_start)
def _on_start(self, _):
if self.unit.is_leader():
self.unit.status = ActiveStatus('I rule')
else:
self.unit.status = ActiveStatus('I am ruled')
@pytest.mark.parametrize('leader', (True, False))
def test_status_leader(leader):
ctx = Context(MyCharm,
meta={"name": "foo"})
out = ctx.run('start', State(leader=leader))
assert out.unit_status == ActiveStatus('I rule' if leader else 'I am ruled')
```
By defining the right state we can programmatically define what answers will the charm get to all the questions it can
ask the juju model: am I leader? What are my relations? What is the remote unit I'm talking to? etc...
## Statuses
One of the simplest types of black-box testing available to charmers is to execute the charm and verify that the charm
sets the expected unit/application status. We have seen a simple example above including leadership. But what if the
charm transitions through a sequence of statuses?
```python
from ops.model import MaintenanceStatus, ActiveStatus, WaitingStatus, BlockedStatus
# charm code:
def _on_event(self, _event):
self.unit.status = MaintenanceStatus('determining who the ruler is...')
try:
if self._call_that_takes_a_few_seconds_and_only_passes_on_leadership:
self.unit.status = ActiveStatus('I rule')
else:
self.unit.status = WaitingStatus('checking this is right...')
self._check_that_takes_some_more_time()
self.unit.status = ActiveStatus('I am ruled')
except:
self.unit.status = BlockedStatus('something went wrong')
```
More broadly, often we want to test 'side effects' of executing a charm, such as what events have been emitted, what
statuses it went through, etc... Before we get there, we have to explain what the `Context` represents, and its
relationship with the `State`.
# Context and State
Consider the following tests. Suppose we want to verify that while handling a given toplevel juju event:
- a specific chain of (custom) events was emitted on the charm
- the charm `juju-log`ged these specific strings
- the charm went through this sequence of app/unit statuses (e.g. `maintenance`, then `waiting`, then `active`)
These types of test have a place in Scenario, but that is not State: the contents of the juju log or the status history
are side effects of executing a charm, but are not persisted in a charm-accessible "state" in any meaningful way.
In other words: those data streams are, from the charm's perspective, write-only.
As such, they do not belong in `scenario.State` but in `scenario.Context`: the object representing the charm's execution
context.
## Status history
You can verify that the charm has followed the expected path by checking the unit/app status history like so:
```python
from charm import MyCharm
from ops.model import MaintenanceStatus, ActiveStatus, WaitingStatus, UnknownStatus
from scenario import State, Context
def test_statuses():
ctx = Context(MyCharm,
meta={"name": "foo"})
ctx.run('start', State(leader=False))
assert ctx.unit_status_history == [
UnknownStatus(),
MaintenanceStatus('determining who the ruler is...'),
WaitingStatus('checking this is right...'),
ActiveStatus("I am ruled"),
]
# similarly you can check the app status history:
assert ctx.app_status_history == [
UnknownStatus(),
...
]
```
Note that the current status is not in the **unit status history**.
Also note that, unless you initialize the State with a preexisting status, the first status in the history will always
be `unknown`. That is because, so far as scenario is concerned, each event is "the first event this charm has ever
seen".
If you want to simulate a situation in which the charm already has seen some event, and is in a status other than
Unknown (the default status every charm is born with), you will have to pass the 'initial status' to State.
```python
from ops.model import ActiveStatus
from scenario import State, Status
# ...
ctx.run('start', State(unit_status=ActiveStatus('foo')))
assert ctx.unit_status_history == [
ActiveStatus('foo'), # now the first status is active: 'foo'!
# ...
]
```
## Workload version history
Using a similar api to `*_status_history`, you can assert that the charm has set one or more workload versions during a
hook execution:
```python
from scenario import Context
# ...
ctx: Context
assert ctx.workload_version_history == ['1', '1.2', '1.5']
# ...
```
## Emitted events
If your charm deals with deferred events, custom events, and charm libs that in turn emit their own custom events, it
can be hard to examine the resulting control flow. In these situations it can be useful to verify that, as a result of a
given Juju event triggering (say, 'start'), a specific chain of events is emitted on the charm. The
resulting state, black-box as it is, gives little insight into how exactly it was obtained.
```python
from scenario import Context
from ops.charm import StartEvent
def test_foo():
ctx = Context(...)
ctx.run('start', ...)
assert len(ctx.emitted_events) == 1
assert isinstance(ctx.emitted_events[0], StartEvent)
```
You can configure what events will be captured by passing the following arguments to `Context`:
- `capture_deferred_events`: If you want to include re-emitted deferred events.
- `capture_framework_events`: If you want to include framework events (`pre-commit`, `commit`, and `collect-status`).
For example:
```python
from scenario import Context, Event, State
def test_emitted_full():
ctx = Context(
MyCharm,
capture_deferred_events=True,
capture_framework_events=True,
)
ctx.run("start", State(deferred=[Event("update-status").deferred(MyCharm._foo)]))
assert len(ctx.emitted_events) == 5
assert [e.handle.kind for e in ctx.emitted_events] == [
"update_status",
"start",
"collect_unit_status",
"pre_commit",
"commit",
]
```
### Low-level access: using directly `capture_events`
If you need more control over what events are captured (or you're not into pytest), you can use directly the context
manager that powers the `emitted_events` fixture: `scenario.capture_events`.
This context manager allows you to intercept any events emitted by the framework.
Usage:
```python
from ops.charm import StartEvent, UpdateStatusEvent
from scenario import State, Context, DeferredEvent, capture_events
with capture_events() as emitted:
ctx = Context(...)
state_out = ctx.run(
"update-status",
State(deferred=[DeferredEvent("start", ...)])
)
# deferred events get reemitted first
assert isinstance(emitted[0], StartEvent)
# the main juju event gets emitted next
assert isinstance(emitted[1], UpdateStatusEvent)
# possibly followed by a tail of all custom events that the main juju event triggered in turn
# assert isinstance(emitted[2], MyFooEvent)
# ...
```
You can filter events by type like so:
```python
from ops.charm import StartEvent, RelationEvent
from scenario import capture_events
with capture_events(StartEvent, RelationEvent) as emitted:
# capture all `start` and `*-relation-*` events.
pass
```
Configuration:
- Passing no event types, like: `capture_events()`, is equivalent to `capture_events(EventBase)`.
- By default, **framework events** (`PreCommit`, `Commit`) are not considered for inclusion in the output list even if
they match the instance check. You can toggle that by passing: `capture_events(include_framework=True)`.
- By default, **deferred events** are included in the listing if they match the instance check. You can toggle that by
passing: `capture_events(include_deferred=False)`.
## Relations
You can write scenario tests to verify the shape of relation data:
```python
from ops.charm import CharmBase
from scenario import Relation, State, Context
# This charm copies over remote app data to local unit data
class MyCharm(CharmBase):
...
def _on_event(self, e):
rel = e.relation
assert rel.app.name == 'remote'
assert rel.data[self.unit]['abc'] == 'foo'
rel.data[self.unit]['abc'] = rel.data[e.app]['cde']
def test_relation_data():
state_in = State(relations=[
Relation(
endpoint="foo",
interface="bar",
remote_app_name="remote",
local_unit_data={"abc": "foo"},
remote_app_data={"cde": "baz!"},
),
])
ctx = Context(MyCharm,
meta={"name": "foo"})
state_out = ctx.run('start', state_in)
assert state_out.relations[0].local_unit_data == {"abc": "baz!"}
# you can do this to check that there are no other differences:
assert state_out.relations == [
Relation(
endpoint="foo",
interface="bar",
remote_app_name="remote",
local_unit_data={"abc": "baz!"},
remote_app_data={"cde": "baz!"},
),
]
# which is very idiomatic and superbly explicit. Noice.
```
The only mandatory argument to `Relation` (and other relation types, see below) is `endpoint`. The `interface` will be
derived from the charm's `metadata.yaml`. When fully defaulted, a relation is 'empty'. There are no remote units, the
remote application is called `'remote'` and only has a single unit `remote/0`, and nobody has written any data to the
databags yet.
That is typically the state of a relation when the first unit joins it.
When you use `Relation`, you are specifying a regular (conventional) relation. But that is not the only type of
relation. There are also peer relations and subordinate relations. While in the background the data model is the same,
the data access rules and the consistency constraints on them are very different. For example, it does not make sense
for a peer relation to have a different 'remote app' than its 'local app', because it's the same application.
### PeerRelation
To declare a peer relation, you should use `scenario.state.PeerRelation`. The core difference with regular relations is
that peer relations do not have a "remote app" (it's this app, in fact). So unlike `Relation`, a `PeerRelation` does not
have `remote_app_name` or `remote_app_data` arguments. Also, it talks in terms of `peers`:
- `Relation.remote_units_data` maps to `PeerRelation.peers_data`
```python
from scenario.state import PeerRelation
relation = PeerRelation(
endpoint="peers",
peers_data={1: {}, 2: {}, 42: {'foo': 'bar'}},
)
```
be mindful when using `PeerRelation` not to include **"this unit"**'s ID in `peers_data` or `peers_ids`, as that would
be flagged by the Consistency Checker:
```python
from scenario import State, PeerRelation, Context
state_in = State(relations=[
PeerRelation(
endpoint="peers",
peers_data={1: {}, 2: {}, 42: {'foo': 'bar'}},
)])
Context(..., unit_id=1).run("start", state_in) # invalid: this unit's id cannot be the ID of a peer.
```
### SubordinateRelation
To declare a subordinate relation, you should use `scenario.state.SubordinateRelation`. The core difference with regular
relations is that subordinate relations always have exactly one remote unit (there is always exactly one remote unit
that this unit can see).
Because of that, `SubordinateRelation`, compared to `Relation`, always talks in terms of `remote`:
- `Relation.remote_units_data` becomes `SubordinateRelation.remote_unit_data` taking a single `Dict[str:str]`. The remote unit ID can be provided as a separate argument.
- `Relation.remote_unit_ids` becomes `SubordinateRelation.remote_unit_id` (a single ID instead of a list of IDs)
- `Relation.remote_units_data` becomes `SubordinateRelation.remote_unit_data` (a single databag instead of a mapping from unit IDs to databags)
```python
from scenario.state import SubordinateRelation
relation = SubordinateRelation(
endpoint="peers",
remote_unit_data={"foo": "bar"},
remote_app_name="zookeeper",
remote_unit_id=42
)
relation.remote_unit_name # "zookeeper/42"
```
### Triggering Relation Events
If you want to trigger relation events, the easiest way to do so is get a hold of the Relation instance and grab the
event from one of its aptly-named properties:
```python
from scenario import Relation
relation = Relation(endpoint="foo", interface="bar")
changed_event = relation.changed_event
joined_event = relation.joined_event
# ...
```
This is in fact syntactic sugar for:
```python
from scenario import Relation, Event
relation = Relation(endpoint="foo", interface="bar")
changed_event = Event('foo-relation-changed', relation=relation)
```
The reason for this construction is that the event is associated with some relation-specific metadata, that Scenario
needs to set up the process that will run `ops.main` with the right environment variables.
### Working with relation IDs
Every time you instantiate `Relation` (or peer, or subordinate), the new instance will be given a unique `relation_id`.
To inspect the ID the next relation instance will have, you can call `state.next_relation_id`.
```python
from scenario import Relation
from scenario.state import next_relation_id
next_id = next_relation_id(update=False)
rel = Relation('foo')
assert rel.relation_id == next_id
```
This can be handy when using `replace` to create new relations, to avoid relation ID conflicts:
```python
from scenario import Relation
from scenario.state import next_relation_id
rel = Relation('foo')
rel2 = rel.replace(local_app_data={"foo": "bar"}, relation_id=next_relation_id())
assert rel2.relation_id == rel.relation_id + 1
```
If you don't do this, and pass both relations into a `State`, you will trigger a consistency checker error.
### Additional event parameters
All relation events have some additional metadata that does not belong in the Relation object, such as, for a
relation-joined event, the name of the (remote) unit that is joining the relation. That is what determines what
`ops.model.Unit` you get when you get `RelationJoinedEvent().unit` in an event handler.
In order to supply this parameter, you will have to **call** the event object and pass as `remote_unit_id` the id of the
remote unit that the event is about. The reason that this parameter is not supplied to `Relation` but to relation
events, is that the relation already ties 'this app' to some 'remote app' (cfr. the `Relation.remote_app_name` attr),
but not to a specific unit. What remote unit this event is about is not a `State` concern but an `Event` one.
The `remote_unit_id` will default to the first ID found in the relation's `remote_units_data`, but if the test you are
writing is close to that domain, you should probably override it and pass it manually.
```python
from scenario import Relation, Event
relation = Relation(endpoint="foo", interface="bar")
remote_unit_2_is_joining_event = relation.joined_event(remote_unit_id=2)
# which is syntactic sugar for:
remote_unit_2_is_joining_event = Event('foo-relation-changed', relation=relation, relation_remote_unit_id=2)
```
### Networks
Simplifying a bit the Juju "spaces" model, each integration endpoint a charm defines in its metadata is associated with a network. Regardless of whether there is a living relation over that endpoint, that is.
If your charm has a relation `"foo"` (defined in metadata.yaml), then the charm will be able at runtime to do `self.model.get_binding("foo").network`.
The network you'll get by doing so is heavily defaulted (see `state.Network.default`) and good for most use-cases because the charm should typically not be concerned about what IP it gets.
On top of the relation-provided network bindings, a charm can also define some `extra-bindings` in its metadata.yaml and access them at runtime. Note that this is a deprecated feature that should not be relied upon. For completeness, we support it in Scenario.
If you want to, you can override any of these relation or extra-binding associated networks with a custom one by passing it to `State.networks`.
```python
from scenario import State, Network
state = State(networks={
'foo': Network.default(private_address='4.4.4.4')
})
```
Where `foo` can either be the name of an `extra-bindings`-defined binding, or a relation endpoint.
# Containers
When testing a kubernetes charm, you can mock container interactions. When using the null state (`State()`), there will
be no containers. So if the charm were to `self.unit.containers`, it would get back an empty dict.
To give the charm access to some containers, you need to pass them to the input state, like so:
`State(containers=[...])`
An example of a state including some containers:
```python
from scenario.state import Container, State
state = State(containers=[
Container(name="foo", can_connect=True),
Container(name="bar", can_connect=False)
])
```
In this case, `self.unit.get_container('foo').can_connect()` would return `True`, while for 'bar' it would give `False`.
### Container filesystem setup
You can configure a container to have some files in it:
```python
from pathlib import Path
from scenario.state import Container, State, Mount
local_file = Path('/path/to/local/real/file.txt')
container = Container(name="foo", can_connect=True, mounts={'local': Mount('/local/share/config.yaml', local_file)})
state = State(containers=[container])
```
In this case, if the charm were to:
```python
def _on_start(self, _):
foo = self.unit.get_container('foo')
content = foo.pull('/local/share/config.yaml').read()
```
then `content` would be the contents of our locally-supplied `file.txt`. You can use `tempdir` for nicely wrapping
data and passing it to the charm via the container.
`container.push` works similarly, so you can write a test like:
```python
import tempfile
from ops.charm import CharmBase
from scenario import State, Container, Mount, Context
class MyCharm(CharmBase):
def __init__(self, *args):
super().__init__(*args)
self.framework.observe(self.on.foo_pebble_ready, self._on_pebble_ready)
def _on_pebble_ready(self, _):
foo = self.unit.get_container('foo')
foo.push('/local/share/config.yaml', "TEST", make_dirs=True)
def test_pebble_push():
with tempfile.NamedTemporaryFile() as local_file:
container = Container(name='foo',
can_connect=True,
mounts={'local': Mount('/local/share/config.yaml', local_file.name)})
state_in = State(
containers=[container]
)
Context(
MyCharm,
meta={"name": "foo", "containers": {"foo": {}}}
).run(
container.pebble_ready_event(),
state_in,
)
assert local_file.read().decode() == "TEST"
```
`container.pebble_ready_event` is syntactic sugar for: `Event("foo-pebble-ready", container=container)`. The reason we
need to associate the container with the event is that the Framework uses an envvar to determine which container the
pebble-ready event is about (it does not use the event name). Scenario needs that information, similarly, for injecting
that envvar into the charm's runtime.
## Container filesystem post-mortem
If the charm writes files to a container (to a location you didn't Mount as a temporary folder you have access to), you will be able to inspect them using the `get_filesystem` api.
```python
from ops.charm import CharmBase
from scenario import State, Container, Mount, Context
class MyCharm(CharmBase):
def __init__(self, *args):
super().__init__(*args)
self.framework.observe(self.on.foo_pebble_ready, self._on_pebble_ready)
def _on_pebble_ready(self, _):
foo = self.unit.get_container('foo')
foo.push('/local/share/config.yaml', "TEST", make_dirs=True)
def test_pebble_push():
container = Container(name='foo',
can_connect=True)
state_in = State(
containers=[container]
)
ctx = Context(
MyCharm,
meta={"name": "foo", "containers": {"foo": {}}}
)
ctx.run("start", state_in)
# this is the root of the simulated container filesystem. Any mounts will be symlinks in it.
container_root_fs = container.get_filesystem(ctx)
cfg_file = container_root_fs / 'local' / 'share' / 'config.yaml'
assert cfg_file.read_text() == "TEST"
```
## `Container.exec` mocks
`container.exec` is a tad more complicated, but if you get to this low a level of simulation, you probably will have far
worse issues to deal with. You need to specify, for each possible command the charm might run on the container, what the
result of that would be: its return code, what will be written to stdout/stderr.
```python
from ops.charm import CharmBase
from scenario import State, Container, ExecOutput, Context
LS_LL = """
.rw-rw-r-- 228 ubuntu ubuntu 18 jan 12:05 -- charmcraft.yaml
.rw-rw-r-- 497 ubuntu ubuntu 18 jan 12:05 -- config.yaml
.rw-rw-r-- 900 ubuntu ubuntu 18 jan 12:05 -- CONTRIBUTING.md
drwxrwxr-x - ubuntu ubuntu 18 jan 12:06 -- lib
"""
class MyCharm(CharmBase):
def _on_start(self, _):
foo = self.unit.get_container('foo')
proc = foo.exec(['ls', '-ll'])
stdout, _ = proc.wait_output()
assert stdout == LS_LL
def test_pebble_exec():
container = Container(
name='foo',
exec_mock={
('ls', '-ll'): # this is the command we're mocking
ExecOutput(return_code=0, # this data structure contains all we need to mock the call.
stdout=LS_LL)
}
)
state_in = State(
containers=[container]
)
state_out = Context(
MyCharm,
meta={"name": "foo", "containers": {"foo": {}}},
).run(
container.pebble_ready_event,
state_in,
)
```
# Storage
If your charm defines `storage` in its metadata, you can use `scenario.state.Storage` to instruct Scenario to make (mocked) filesystem storage available to the charm at runtime.
Using the same `get_filesystem` API as `Container`, you can access the tempdir used by Scenario to mock the filesystem root before and after the scenario runs.
```python
from scenario import Storage, Context, State
# some charm with a 'foo' filesystem-type storage defined in metadata.yaml
ctx = Context(MyCharm)
storage = Storage("foo")
# setup storage with some content
(storage.get_filesystem(ctx) / "myfile.txt").write_text("helloworld")
with ctx.manager("update-status", State(storage=[storage])) as mgr:
foo = mgr.charm.model.storages["foo"][0]
loc = foo.location
path = loc / "myfile.txt"
assert path.exists()
assert path.read_text() == "helloworld"
myfile = loc / "path.py"
myfile.write_text("helloworlds")
# post-mortem: inspect fs contents.
assert (
storage.get_filesystem(ctx) / "path.py"
).read_text() == "helloworlds"
```
Note that State only wants to know about **attached** storages. A storage which is not attached to the charm can simply be omitted from State and the charm will be none the wiser.
## Storage-add
If a charm requests adding more storage instances while handling some event, you can inspect that from the `Context.requested_storage` API.
```python
# in MyCharm._on_foo:
# the charm requests two new "foo" storage instances to be provisioned
self.model.storages.request("foo", 2)
```
From test code, you can inspect that:
```python
from scenario import Context, State
ctx = Context(MyCharm)
ctx.run('some-event-that-will-cause_on_foo-to-be-called', State())
# the charm has requested two 'foo' storages to be provisioned
assert ctx.requested_storages['foo'] == 2
```
Requesting storages has no other consequence in Scenario. In real life, this request will trigger Juju to provision the storage and execute the charm again with `foo-storage-attached`.
So a natural follow-up Scenario test suite for this case would be:
```python
from scenario import Context, State, Storage
ctx = Context(MyCharm)
foo_0 = Storage('foo')
# the charm is notified that one of the storages it has requested is ready
ctx.run(foo_0.attached_event, State(storage=[foo_0]))
foo_1 = Storage('foo')
# the charm is notified that the other storage is also ready
ctx.run(foo_1.attached_event, State(storage=[foo_0, foo_1]))
```
# Ports
Since `ops 2.6.0`, charms can invoke the `open-port`, `close-port`, and `opened-ports` hook tools to manage the ports opened on the host vm/container. Using the `State.opened_ports` api, you can:
- simulate a charm run with a port opened by some previous execution
```python
from scenario import State, Port, Context
ctx = Context(MyCharm)
ctx.run("start", State(opened_ports=[Port("tcp", 42)]))
```
- assert that a charm has called `open-port` or `close-port`:
```python
from scenario import State, Port, Context
ctx = Context(MyCharm)
state1 = ctx.run("start", State())
assert state1.opened_ports == [Port("tcp", 42)]
state2 = ctx.run("stop", state1)
assert state2.opened_ports == []
```
# Secrets
Scenario has secrets. Here's how you use them.
```python
from scenario import State, Secret
state = State(
secrets=[
Secret(
id='foo',
contents={0: {'key': 'public'}}
)
]
)
```
The only mandatory arguments to Secret are its secret ID (which should be unique) and its 'contents': that is, a mapping
from revision numbers (integers) to a `str:str` dict representing the payload of the revision.
There are three cases:
- the secret is owned by this app but not this unit, in which case this charm can only manage it if we are the leader
- the secret is owned by this unit, in which case this charm can always manage it (leader or not)
- (default) the secret is not owned by this app nor unit, which means we can't manage it but only view it
Thus by default, the secret is not owned by **this charm**, but, implicitly, by some unknown 'other charm', and that other charm has granted us view rights.
The presence of the secret in `State.secrets` entails that we have access to it, either as owners or as grantees. Therefore, if we're not owners, we must be grantees. Absence of a Secret from the known secrets list means we are not entitled to obtaining it in any way. The charm, indeed, shouldn't even know it exists.
[note]
If this charm does not own the secret, but also it was not granted view rights by the (remote) owner, you model this in Scenario by _not adding it to State.secrets_! The presence of a `Secret` in `State.secrets` means, in other words, that the charm has view rights (otherwise, why would we put it there?). If the charm owns the secret, or is leader, it will _also_ have manage rights on top of view ones.
[/note]
To specify a secret owned by this unit (or app):
```python
from scenario import State, Secret
state = State(
secrets=[
Secret(
id='foo',
contents={0: {'key': 'private'}},
owner='unit', # or 'app'
remote_grants={0: {"remote"}}
# the secret owner has granted access to the "remote" app over some relation with ID 0
)
]
)
```
To specify a secret owned by some other application and give this unit (or app) access to it:
```python
from scenario import State, Secret
state = State(
secrets=[
Secret(
id='foo',
contents={0: {'key': 'public'}},
# owner=None, which is the default
revision=0, # the revision that this unit (or app) is currently tracking
)
]
)
```
# Actions
An action is a special sort of event, even though `ops` handles them almost identically.
In most cases, you'll want to inspect the 'results' of an action, or whether it has failed or
logged something while executing. Many actions don't have a direct effect on the output state.
For this reason, the output state is less prominent in the return type of `Context.run_action`.
How to test actions with scenario:
## Actions without parameters
```python
from scenario import Context, State, ActionOutput
from charm import MyCharm
def test_backup_action():
ctx = Context(MyCharm)
# If you didn't declare do_backup in the charm's `actions.yaml`,
# the `ConsistencyChecker` will slap you on the wrist and refuse to proceed.
out: ActionOutput = ctx.run_action("do_backup_action", State())
# you can assert action results, logs, failure using the ActionOutput interface
assert out.logs == ['baz', 'qux']
if out.success:
# if the action did not fail, we can read the results:
assert out.results == {'foo': 'bar'}
else:
# if the action fails, we can read a failure message
assert out.failure == 'boo-hoo'
```
## Parametrized Actions
If the action takes parameters, you'll need to instantiate an `Action`.
```python
from scenario import Action, Context, State, ActionOutput
from charm import MyCharm
def test_backup_action():
# define an action
action = Action('do_backup', params={'a': 'b'})
ctx = Context(MyCharm)
# if the parameters (or their type) don't match what declared in actions.yaml,
# the `ConsistencyChecker` will slap you on the other wrist.
out: ActionOutput = ctx.run_action(action, State())
# ...
```
# Deferred events
Scenario allows you to accurately simulate the Operator Framework's event queue. The event queue is responsible for
keeping track of the deferred events. On the input side, you can verify that if the charm triggers with this and that
event in its queue (they would be there because they had been deferred in the previous run), then the output state is
valid.
```python
from scenario import State, deferred, Context
class MyCharm(...):
...
def _on_update_status(self, e):
e.defer()
def _on_start(self, e):
e.defer()
def test_start_on_deferred_update_status(MyCharm):
"""Test charm execution if a 'start' is dispatched when in the previous run an update-status had been deferred."""
state_in = State(
deferred=[
deferred('update_status',
handler=MyCharm._on_update_status)
]
)
state_out = Context(MyCharm).run('start', state_in)
assert len(state_out.deferred) == 1
assert state_out.deferred[0].name == 'start'
```
You can also generate the 'deferred' data structure (called a DeferredEvent) from the corresponding Event (and the
handler):
```python
from scenario import Event, Relation
class MyCharm(...):
...
deferred_start = Event('start').deferred(MyCharm._on_start)
deferred_install = Event('install').deferred(MyCharm._on_start)
```
## relation events:
```python
foo_relation = Relation('foo')
deferred_relation_changed_evt = foo_relation.changed_event.deferred(handler=MyCharm._on_foo_relation_changed)
```
On the output side, you can verify that an event that you expect to have been deferred during this trigger, has indeed
been deferred.
```python
from scenario import State, Context
class MyCharm(...):
...
def _on_start(self, e):
e.defer()
def test_defer(MyCharm):
out = Context(MyCharm).run('start', State())
assert len(out.deferred) == 1
assert out.deferred[0].name == 'start'
```
## Deferring relation events
If you want to test relation event deferrals, some extra care needs to be taken. RelationEvents hold references to the
Relation instance they are about. So do they in Scenario. You can use the deferred helper to generate the data
structure:
```python
from scenario import State, Relation, deferred
class MyCharm(...):
...
def _on_foo_relation_changed(self, e):
e.defer()
def test_start_on_deferred_update_status(MyCharm):
foo_relation = Relation('foo')
State(
relations=[foo_relation],
deferred=[
deferred('foo_relation_changed',
handler=MyCharm._on_foo_relation_changed,
relation=foo_relation)
]
)
```
but you can also use a shortcut from the relation event itself, as mentioned above:
```python
from scenario import Relation
class MyCharm(...):
...
foo_relation = Relation('foo')
foo_relation.changed_event.deferred(handler=MyCharm._on_foo_relation_changed)
```
### Fine-tuning
The deferred helper Scenario provides will not support out of the box all custom event subclasses, or events emitted by
charm libraries or objects other than the main charm class.
For general-purpose usage, you will need to instantiate DeferredEvent directly.
```python
from scenario import DeferredEvent
my_deferred_event = DeferredEvent(
handle_path='MyCharm/MyCharmLib/on/database_ready[1]',
owner='MyCharmLib', # the object observing the event. Could also be MyCharm.
observer='_on_database_ready'
)
```
# StoredState
Scenario can simulate StoredState. You can define it on the input side as:
```python
from ops.charm import CharmBase
from ops.framework import StoredState as Ops_StoredState, Framework
from scenario import State, StoredState
class MyCharmType(CharmBase):
my_stored_state = Ops_StoredState()
def __init__(self, framework: Framework):
super().__init__(framework)
assert self.my_stored_state.foo == 'bar' # this will pass!
state = State(stored_state=[
StoredState(
owner_path="MyCharmType",
name="my_stored_state",
content={
'foo': 'bar',
'baz': {42: 42},
})
])
```
And the charm's runtime will see `self.stored_State.foo` and `.baz` as expected. Also, you can run assertions on it on
the output side the same as any other bit of state.
# Resources
If your charm requires access to resources, you can make them available to it through `State.resources`.
From the perspective of a 'real' deployed charm, if your charm _has_ resources defined in `metadata.yaml`, they _must_ be made available to the charm. That is a Juju-enforced constraint: you can't deploy a charm without attaching all resources it needs to it.
However, when testing, this constraint is unnecessarily strict (and it would also mean the great majority of all existing tests would break) since a charm will only notice that a resource is not available when it explicitly asks for it, which not many charms do.
So, the only consistency-level check we enforce in Scenario when it comes to resource is that if a resource is provided in State, it needs to have been declared in metadata.
```python
from scenario import State, Context
ctx = Context(MyCharm, meta={'name': 'juliette', "resources": {"foo": {"type": "oci-image"}}})
with ctx.manager("start", State(resources={'foo': '/path/to/resource.tar'})) as mgr:
# if the charm, at runtime, were to call self.model.resources.fetch("foo"), it would get '/path/to/resource.tar' back.
path = mgr.charm.model.resources.fetch('foo')
assert path == '/path/to/resource.tar'
```
# Emitting custom events
While the main use case of Scenario is to emit juju events, i.e. the built-in `start`, `install`, `*-relation-changed`,
etc..., it can be sometimes handy to directly trigger custom events defined on arbitrary Objects in your hierarchy.
Suppose your charm uses a charm library providing an `ingress_provided` event.
The 'proper' way to emit it is to run the event that causes that custom event to be emitted by the library, whatever
that may be, for example a `foo-relation-changed`.
However, that may mean that you have to set up all sorts of State and mocks so that the right preconditions are met and
the event is emitted at all.
If for whatever reason you don't want to do that and you attempt to run that event directly you will get an error:
```python
from scenario import Context, State
Context(...).run("ingress_provided", State()) # raises scenario.ops_main_mock.NoObserverError
```
This happens because the framework, by default, searches for an event source named `ingress_provided` in `charm.on`, but
since the event is defined on another Object, it will fail to find it.
You can prefix the event name with the path leading to its owner to tell Scenario where to find the event source:
```python
from scenario import Context, State
Context(...).run("my_charm_lib.on.foo", State())
```
This will instruct Scenario to emit `my_charm.my_charm_lib.on.foo`.
(always omit the 'root', i.e. the charm framework key, from the path)
# Live charm introspection
Scenario is a black-box, state-transition testing framework. It makes it trivial to assert that a status went from A to
B, but not to assert that, in the context of this charm execution, with this state, a certain charm-internal method was called and returned a
given piece of data, or would return this and that _if_ it had been called.
Scenario offers a cheekily-named context manager for this use case specifically:
```python
from ops import CharmBase, StoredState
from charms.bar.lib_name.v1.charm_lib import CharmLib
from scenario import Context, State
class MyCharm(CharmBase):
META = {"name": "mycharm"}
_stored = StoredState()
def __init__(self, framework):
super().__init__(framework)
self._stored.set_default(a="a")
self.my_charm_lib = CharmLib()
framework.observe(self.on.start, self._on_start)
def _on_start(self, event):
self._stored.a = "b"
def test_live_charm_introspection(mycharm):
ctx = Context(mycharm, meta=mycharm.META)
# If you want to do this with actions, you can use `Context.action_manager` instead.
with ctx.manager("start", State()) as manager:
# this is your charm instance, after ops has set it up
charm: MyCharm = manager.charm
# we can check attributes on nested Objects or the charm itself
assert charm.my_charm_lib.foo == "foo"
# such as stored state
assert charm._stored.a == "a"
# this will tell ops.main to proceed with normal execution and emit the "start" event on the charm
state_out = manager.run()
# after that is done, we are handed back control, and we can again do some introspection
assert charm.my_charm_lib.foo == "bar"
# and check that the charm's internal state is as we expect
assert charm._stored.a == "b"
# state_out is, as in regular scenario tests, a State object you can assert on:
assert state_out.unit_status == ...
```
Note that you can't call `manager.run()` multiple times: the manager is a context that ensures that `ops.main` 'pauses' right
before emitting the event to hand you some introspection hooks, but for the rest this is a regular scenario test: you
can't emit multiple events in a single charm execution.
# The virtual charm root
Before executing the charm, Scenario copies the charm's `/src`, any libs, the metadata, config, and actions `yaml`s to a temporary directory. The
charm will see that tempdir as its 'root'. This allows us to keep things simple when dealing with metadata that can be
either inferred from the charm type being passed to `Context` or be passed to it as an argument, thereby overriding
the inferred one. This also allows you to test charms defined on the fly, as in:
```python
from ops.charm import CharmBase
from scenario import State, Context
class MyCharmType(CharmBase):
pass
ctx = Context(charm_type=MyCharmType,
meta={'name': 'my-charm-name'})
ctx.run('start', State())
```
A consequence of this fact is that you have no direct control over the tempdir that we are creating to put the metadata
you are passing to `.run()` (because `ops` expects it to be a file...). That is, unless you pass your own:
```python
from ops.charm import CharmBase
from scenario import State, Context
import tempfile
class MyCharmType(CharmBase):
pass
td = tempfile.TemporaryDirectory()
state = Context(
charm_type=MyCharmType,
meta={'name': 'my-charm-name'},
charm_root=td.name
).run('start', State())
```
Do this, and you will be able to set up said directory as you like before the charm is run, as well as verify its
contents after the charm has run. Do keep in mind that any metadata files you create in it will be overwritten by Scenario, and therefore
ignored, if you pass any metadata keys to `Context`. Omit `meta` in the call
above, and Scenario will instead attempt to read `metadata.yaml` from the
temporary directory.
# Immutability
All of the data structures in `state`, e.g. `State, Relation, Container`, etc... are immutable (implemented as frozen
dataclasses).
This means that all components of the state that goes into a `context.run()` call are not mutated by the call, and the
state that you obtain in return is a different instance, and all parts of it have been (deep)copied.
This ensures that you can do delta-based comparison of states without worrying about them being mutated by scenario.
If you want to modify any of these data structures, you will need to either reinstantiate it from scratch, or use
the `replace` api.
```python
from scenario import Relation
relation = Relation('foo', remote_app_data={"1": "2"})
# make a copy of relation, but with remote_app_data set to {"3", "4"}
relation2 = relation.replace(remote_app_data={"3", "4"})
```
# Consistency checks
A Scenario, that is, the combination of an event, a state, and a charm, is consistent if it's plausible in JujuLand. For
example, Juju can't emit a `foo-relation-changed` event on your charm unless your charm has declared a `foo` relation
endpoint in its `metadata.yaml`. If that happens, that's a juju bug. Scenario however assumes that Juju is bug-free,
therefore, so far as we're concerned, that can't happen, and therefore we help you verify that the scenarios you create
are consistent and raise an exception if that isn't so.
That happens automatically behind the scenes whenever you trigger an event;
`scenario.consistency_checker.check_consistency` is called and verifies that the scenario makes sense.
## Caveats:
- False positives: not all checks are implemented yet; more will come.
- False negatives: it is possible that a scenario you know to be consistent is seen as inconsistent. That is probably a
bug in the consistency checker itself, please report it.
- Inherent limitations: if you have a custom event whose name conflicts with a builtin one, the consistency constraints
of the builtin one will apply. For example: if you decide to name your custom event `bar-pebble-ready`, but you are
working on a machine charm or don't have either way a `bar` container in your `metadata.yaml`, Scenario will flag that
as inconsistent.
## Bypassing the checker
If you have a clear false negative, are explicitly testing 'edge', inconsistent situations, or for whatever reason the
checker is in your way, you can set the `SCENARIO_SKIP_CONSISTENCY_CHECKS` envvar and skip it altogether. Hopefully you
don't need that.
# Jhack integrations
Up until `v5.6.0`, `scenario` shipped with a cli tool called `snapshot`, used to interact with a live charm's state.
The functionality [has been moved over to `jhack`](https://github.com/PietroPasotti/jhack/pull/111),
to allow us to keep working on it independently, and to streamline
the profile of `scenario` itself as it becomes more broadly adopted and ready for widespread usage.
Raw data
{
"_id": null,
"home_page": "",
"name": "ops-scenario",
"maintainer": "",
"docs_url": null,
"requires_python": ">=3.8",
"maintainer_email": "",
"keywords": "juju,test",
"author": "",
"author_email": "Pietro Pasotti <pietro.pasotti@canonical.com>",
"download_url": "https://files.pythonhosted.org/packages/4a/a7/b3774d2354dffb506ad3e56016c43b9b6f8c712786113c34cea1e7700089/ops-scenario-6.0.3.tar.gz",
"platform": null,
"description": "# Scenario\n\n[](https://github.com/canonical/ops-scenario/actions/workflows/build_wheels.yaml)\n[](https://github.com/canonical/ops-scenario/actions/workflows/quality_checks.yaml?event=pull_request)\n[](https://discourse.charmhub.io)\n[](https://github.com/PietroPasotti/jhack)\n[](https://discourse.charmhub.io/t/rethinking-charm-testing-with-ops-scenario/8649)\n[](https://www.python.org/downloads/release/python-380/)\n\nScenario is a state-transition testing SDK for Operator Framework charms.\n\nWhere the Harness enables you to procedurally mock pieces of the state the charm needs to function, Scenario tests allow\nyou to declaratively define the state all at once, and use it as a sort of context against which you can fire a single\nevent on the charm and execute its logic.\n\nThis puts scenario tests somewhere in between unit and integration tests: some say 'functional', some say 'contract', I prefer 'state-transition'.\n\nScenario tests nudge you into thinking of a charm as an input->output function. The input is the\nunion of an `Event` (why am I, charm, being executed), a `State` (am I leader? what is my relation data? what is my\nconfig?...) and the charm's execution `Context` (what relations can I have? what containers can I have?...). The output is another `State`: the state after the charm has had a chance to interact with the\nmocked juju model and affect the initial state back.\n\n\n\nFor example: a charm currently in `unknown` status is executed with a `start` event, and based on whether it has leadership or not (according to its input state), it will decide to set `active` or `blocked` status (which will be reflected in the output state).\n\nScenario-testing a charm, then, means verifying that:\n\n- the charm does not raise uncaught exceptions while handling the event\n- the output state (or the diff with the input state) is as expected.\n\n# Core concepts as a metaphor\n\nI like metaphors, so here we go:\n\n- There is a theatre stage.\n- You pick an actor (a Charm) to put on the stage. Not just any actor: an improv one.\n- You arrange the stage with content that the actor will have to interact with. This consists of selecting:\n - An initial situation (`State`) in which the actor is, e.g. is the actor the main role or an NPC (`is_leader`), or what\n other actors are there around it, what is written in those pebble-shaped books on the table?\n - Something that has just happened (an `Event`) and to which the actor has to react (e.g. one of the NPCs leaves the\n stage (`relation-departed`), or the content of one of the books changes).\n- How the actor will react to the event will have an impact on the context: e.g. the actor might knock over a table (a\n container), or write something down into one of the books.\n\n# Core concepts not as a metaphor\n\nScenario tests are about running assertions on atomic state transitions treating the charm being tested like a black\nbox. An initial state goes in, an event occurs (say, `'start'`) and a new state comes out. Scenario tests are about\nvalidating the transition, that is, consistency-checking the delta between the two states, and verifying the charm\nauthor's expectations.\n\nComparing scenario tests with `Harness` tests:\n\n- Harness exposes an imperative API: the user is expected to call methods on the Harness driving it to the desired\n state, then verify its validity by calling charm methods or inspecting the raw data. In contrast, Scenario is declarative. You fully specify an initial state, an execution context and an event, then you run the charm and inspect the results.\n- Harness instantiates the charm once, then allows you to fire multiple events on the charm, which is breeding ground\n for subtle bugs. Scenario tests are centered around testing single state transitions, that is, one event at a time.\n This ensures that the execution environment is as clean as possible (for a unit test).\n- Harness maintains a model of the juju Model, which is a maintenance burden and adds complexity. Scenario mocks at the\n level of hook tools and stores all mocking data in a monolithic data structure (the State), which makes it more\n lightweight and portable.\n\n# Writing scenario tests\n\nA scenario test consists of three broad steps:\n\n- **Arrange**:\n - declare the context \n - declare the input state\n - select an event to fire\n- **Act**:\n - run the context (i.e. obtain the output state, given the input state and the event)\n- **Assert**:\n - verify that the output state (or the delta with the input state) is how you expect it to be\n - verify that the charm has seen a certain sequence of statuses, events, and `juju-log` calls\n - optionally, you can use a context manager to get a hold of the charm instance and run assertions on internal APIs and the internal state of the charm and operator framework.\n\nThe most basic scenario is one in which all is defaulted and barely any data is\navailable. The charm has no config, no relations, no leadership, and its status is `unknown`.\n\nWith that, we can write the simplest possible scenario test:\n\n```python\nfrom scenario import State, Context, Event\nfrom ops.charm import CharmBase\nfrom ops.model import UnknownStatus\n\n\nclass MyCharm(CharmBase):\n pass\n\n\ndef test_scenario_base():\n ctx = Context(MyCharm, meta={\"name\": \"foo\"})\n out = ctx.run(Event(\"start\"), State())\n assert out.unit_status == UnknownStatus()\n```\n\nNow let's start making it more complicated. Our charm sets a special state if it has leadership on 'start':\n\n```python\nimport pytest\nfrom scenario import State, Context\nfrom ops.charm import CharmBase\nfrom ops.model import ActiveStatus\n\n\nclass MyCharm(CharmBase):\n def __init__(self, ...):\n self.framework.observe(self.on.start, self._on_start)\n\n def _on_start(self, _):\n if self.unit.is_leader():\n self.unit.status = ActiveStatus('I rule')\n else:\n self.unit.status = ActiveStatus('I am ruled')\n\n\n@pytest.mark.parametrize('leader', (True, False))\ndef test_status_leader(leader):\n ctx = Context(MyCharm,\n meta={\"name\": \"foo\"})\n out = ctx.run('start', State(leader=leader))\n assert out.unit_status == ActiveStatus('I rule' if leader else 'I am ruled')\n```\n\nBy defining the right state we can programmatically define what answers will the charm get to all the questions it can\nask the juju model: am I leader? What are my relations? What is the remote unit I'm talking to? etc...\n\n## Statuses\n\nOne of the simplest types of black-box testing available to charmers is to execute the charm and verify that the charm\nsets the expected unit/application status. We have seen a simple example above including leadership. But what if the\ncharm transitions through a sequence of statuses?\n\n```python\nfrom ops.model import MaintenanceStatus, ActiveStatus, WaitingStatus, BlockedStatus\n\n\n# charm code:\ndef _on_event(self, _event):\n self.unit.status = MaintenanceStatus('determining who the ruler is...')\n try:\n if self._call_that_takes_a_few_seconds_and_only_passes_on_leadership:\n self.unit.status = ActiveStatus('I rule')\n else:\n self.unit.status = WaitingStatus('checking this is right...')\n self._check_that_takes_some_more_time()\n self.unit.status = ActiveStatus('I am ruled')\n except:\n self.unit.status = BlockedStatus('something went wrong')\n```\n\nMore broadly, often we want to test 'side effects' of executing a charm, such as what events have been emitted, what\nstatuses it went through, etc... Before we get there, we have to explain what the `Context` represents, and its\nrelationship with the `State`.\n\n# Context and State\n\nConsider the following tests. Suppose we want to verify that while handling a given toplevel juju event:\n\n- a specific chain of (custom) events was emitted on the charm\n- the charm `juju-log`ged these specific strings\n- the charm went through this sequence of app/unit statuses (e.g. `maintenance`, then `waiting`, then `active`)\n\nThese types of test have a place in Scenario, but that is not State: the contents of the juju log or the status history\nare side effects of executing a charm, but are not persisted in a charm-accessible \"state\" in any meaningful way.\nIn other words: those data streams are, from the charm's perspective, write-only.\n\nAs such, they do not belong in `scenario.State` but in `scenario.Context`: the object representing the charm's execution\ncontext.\n\n## Status history\n\nYou can verify that the charm has followed the expected path by checking the unit/app status history like so:\n\n```python\nfrom charm import MyCharm\nfrom ops.model import MaintenanceStatus, ActiveStatus, WaitingStatus, UnknownStatus\nfrom scenario import State, Context\n\n\ndef test_statuses():\n ctx = Context(MyCharm,\n meta={\"name\": \"foo\"})\n ctx.run('start', State(leader=False))\n assert ctx.unit_status_history == [\n UnknownStatus(),\n MaintenanceStatus('determining who the ruler is...'),\n WaitingStatus('checking this is right...'),\n ActiveStatus(\"I am ruled\"),\n ]\n \n # similarly you can check the app status history:\n assert ctx.app_status_history == [\n UnknownStatus(),\n ...\n ]\n```\n\nNote that the current status is not in the **unit status history**.\n\nAlso note that, unless you initialize the State with a preexisting status, the first status in the history will always\nbe `unknown`. That is because, so far as scenario is concerned, each event is \"the first event this charm has ever\nseen\".\n\nIf you want to simulate a situation in which the charm already has seen some event, and is in a status other than\nUnknown (the default status every charm is born with), you will have to pass the 'initial status' to State.\n\n```python\nfrom ops.model import ActiveStatus\nfrom scenario import State, Status\n\n# ...\nctx.run('start', State(unit_status=ActiveStatus('foo')))\nassert ctx.unit_status_history == [\n ActiveStatus('foo'), # now the first status is active: 'foo'!\n # ...\n]\n\n```\n\n## Workload version history\n\nUsing a similar api to `*_status_history`, you can assert that the charm has set one or more workload versions during a\nhook execution:\n\n```python\nfrom scenario import Context\n\n# ...\nctx: Context\nassert ctx.workload_version_history == ['1', '1.2', '1.5']\n# ...\n```\n\n## Emitted events\n\nIf your charm deals with deferred events, custom events, and charm libs that in turn emit their own custom events, it\ncan be hard to examine the resulting control flow. In these situations it can be useful to verify that, as a result of a\ngiven Juju event triggering (say, 'start'), a specific chain of events is emitted on the charm. The\nresulting state, black-box as it is, gives little insight into how exactly it was obtained.\n\n```python\nfrom scenario import Context\nfrom ops.charm import StartEvent\n\n\ndef test_foo():\n ctx = Context(...)\n ctx.run('start', ...)\n\n assert len(ctx.emitted_events) == 1\n assert isinstance(ctx.emitted_events[0], StartEvent)\n```\n\nYou can configure what events will be captured by passing the following arguments to `Context`:\n- `capture_deferred_events`: If you want to include re-emitted deferred events.\n- `capture_framework_events`: If you want to include framework events (`pre-commit`, `commit`, and `collect-status`). \n\nFor example:\n```python\nfrom scenario import Context, Event, State\n\ndef test_emitted_full():\n ctx = Context(\n MyCharm,\n capture_deferred_events=True,\n capture_framework_events=True,\n )\n ctx.run(\"start\", State(deferred=[Event(\"update-status\").deferred(MyCharm._foo)]))\n\n assert len(ctx.emitted_events) == 5\n assert [e.handle.kind for e in ctx.emitted_events] == [\n \"update_status\",\n \"start\",\n \"collect_unit_status\",\n \"pre_commit\",\n \"commit\",\n ]\n```\n\n\n### Low-level access: using directly `capture_events`\n\nIf you need more control over what events are captured (or you're not into pytest), you can use directly the context\nmanager that powers the `emitted_events` fixture: `scenario.capture_events`.\nThis context manager allows you to intercept any events emitted by the framework.\n\nUsage:\n\n```python\nfrom ops.charm import StartEvent, UpdateStatusEvent\nfrom scenario import State, Context, DeferredEvent, capture_events\n\nwith capture_events() as emitted:\n ctx = Context(...)\n state_out = ctx.run(\n \"update-status\",\n State(deferred=[DeferredEvent(\"start\", ...)])\n )\n\n# deferred events get reemitted first\nassert isinstance(emitted[0], StartEvent)\n# the main juju event gets emitted next\nassert isinstance(emitted[1], UpdateStatusEvent)\n# possibly followed by a tail of all custom events that the main juju event triggered in turn\n# assert isinstance(emitted[2], MyFooEvent)\n# ...\n```\n\nYou can filter events by type like so:\n\n```python\nfrom ops.charm import StartEvent, RelationEvent\nfrom scenario import capture_events\n\nwith capture_events(StartEvent, RelationEvent) as emitted:\n # capture all `start` and `*-relation-*` events.\n pass\n```\n\nConfiguration:\n\n- Passing no event types, like: `capture_events()`, is equivalent to `capture_events(EventBase)`.\n- By default, **framework events** (`PreCommit`, `Commit`) are not considered for inclusion in the output list even if\n they match the instance check. You can toggle that by passing: `capture_events(include_framework=True)`.\n- By default, **deferred events** are included in the listing if they match the instance check. You can toggle that by\n passing: `capture_events(include_deferred=False)`.\n\n## Relations\n\nYou can write scenario tests to verify the shape of relation data:\n\n```python\nfrom ops.charm import CharmBase\n\nfrom scenario import Relation, State, Context\n\n\n# This charm copies over remote app data to local unit data\nclass MyCharm(CharmBase):\n ...\n\n def _on_event(self, e):\n rel = e.relation\n assert rel.app.name == 'remote'\n assert rel.data[self.unit]['abc'] == 'foo'\n rel.data[self.unit]['abc'] = rel.data[e.app]['cde']\n\n\ndef test_relation_data():\n state_in = State(relations=[\n Relation(\n endpoint=\"foo\",\n interface=\"bar\",\n remote_app_name=\"remote\",\n local_unit_data={\"abc\": \"foo\"},\n remote_app_data={\"cde\": \"baz!\"},\n ),\n ])\n ctx = Context(MyCharm,\n meta={\"name\": \"foo\"})\n\n state_out = ctx.run('start', state_in)\n\n assert state_out.relations[0].local_unit_data == {\"abc\": \"baz!\"}\n # you can do this to check that there are no other differences:\n assert state_out.relations == [\n Relation(\n endpoint=\"foo\",\n interface=\"bar\",\n remote_app_name=\"remote\",\n local_unit_data={\"abc\": \"baz!\"},\n remote_app_data={\"cde\": \"baz!\"},\n ),\n ]\n\n# which is very idiomatic and superbly explicit. Noice.\n```\n\nThe only mandatory argument to `Relation` (and other relation types, see below) is `endpoint`. The `interface` will be\nderived from the charm's `metadata.yaml`. When fully defaulted, a relation is 'empty'. There are no remote units, the\nremote application is called `'remote'` and only has a single unit `remote/0`, and nobody has written any data to the\ndatabags yet.\n\nThat is typically the state of a relation when the first unit joins it.\n\nWhen you use `Relation`, you are specifying a regular (conventional) relation. But that is not the only type of\nrelation. There are also peer relations and subordinate relations. While in the background the data model is the same,\nthe data access rules and the consistency constraints on them are very different. For example, it does not make sense\nfor a peer relation to have a different 'remote app' than its 'local app', because it's the same application.\n\n### PeerRelation\n\nTo declare a peer relation, you should use `scenario.state.PeerRelation`. The core difference with regular relations is\nthat peer relations do not have a \"remote app\" (it's this app, in fact). So unlike `Relation`, a `PeerRelation` does not\nhave `remote_app_name` or `remote_app_data` arguments. Also, it talks in terms of `peers`:\n\n- `Relation.remote_units_data` maps to `PeerRelation.peers_data`\n\n```python\nfrom scenario.state import PeerRelation\n\nrelation = PeerRelation(\n endpoint=\"peers\",\n peers_data={1: {}, 2: {}, 42: {'foo': 'bar'}},\n)\n```\n\nbe mindful when using `PeerRelation` not to include **\"this unit\"**'s ID in `peers_data` or `peers_ids`, as that would\nbe flagged by the Consistency Checker:\n\n```python\nfrom scenario import State, PeerRelation, Context\n\nstate_in = State(relations=[\n PeerRelation(\n endpoint=\"peers\",\n peers_data={1: {}, 2: {}, 42: {'foo': 'bar'}},\n )])\n\nContext(..., unit_id=1).run(\"start\", state_in) # invalid: this unit's id cannot be the ID of a peer.\n\n\n```\n\n### SubordinateRelation\n\nTo declare a subordinate relation, you should use `scenario.state.SubordinateRelation`. The core difference with regular\nrelations is that subordinate relations always have exactly one remote unit (there is always exactly one remote unit\nthat this unit can see). \nBecause of that, `SubordinateRelation`, compared to `Relation`, always talks in terms of `remote`:\n\n- `Relation.remote_units_data` becomes `SubordinateRelation.remote_unit_data` taking a single `Dict[str:str]`. The remote unit ID can be provided as a separate argument. \n- `Relation.remote_unit_ids` becomes `SubordinateRelation.remote_unit_id` (a single ID instead of a list of IDs)\n- `Relation.remote_units_data` becomes `SubordinateRelation.remote_unit_data` (a single databag instead of a mapping from unit IDs to databags)\n\n```python\nfrom scenario.state import SubordinateRelation\n\nrelation = SubordinateRelation(\n endpoint=\"peers\",\n remote_unit_data={\"foo\": \"bar\"},\n remote_app_name=\"zookeeper\",\n remote_unit_id=42\n)\nrelation.remote_unit_name # \"zookeeper/42\"\n```\n\n### Triggering Relation Events\n\nIf you want to trigger relation events, the easiest way to do so is get a hold of the Relation instance and grab the\nevent from one of its aptly-named properties:\n\n```python\nfrom scenario import Relation\n\nrelation = Relation(endpoint=\"foo\", interface=\"bar\")\nchanged_event = relation.changed_event\njoined_event = relation.joined_event\n# ...\n```\n\nThis is in fact syntactic sugar for:\n\n```python\nfrom scenario import Relation, Event\n\nrelation = Relation(endpoint=\"foo\", interface=\"bar\")\nchanged_event = Event('foo-relation-changed', relation=relation)\n```\n\nThe reason for this construction is that the event is associated with some relation-specific metadata, that Scenario\nneeds to set up the process that will run `ops.main` with the right environment variables.\n\n### Working with relation IDs\n\nEvery time you instantiate `Relation` (or peer, or subordinate), the new instance will be given a unique `relation_id`.\nTo inspect the ID the next relation instance will have, you can call `state.next_relation_id`.\n\n```python\nfrom scenario import Relation\nfrom scenario.state import next_relation_id\n\nnext_id = next_relation_id(update=False)\nrel = Relation('foo')\nassert rel.relation_id == next_id\n``` \n\nThis can be handy when using `replace` to create new relations, to avoid relation ID conflicts:\n\n```python\nfrom scenario import Relation\nfrom scenario.state import next_relation_id\n\nrel = Relation('foo')\nrel2 = rel.replace(local_app_data={\"foo\": \"bar\"}, relation_id=next_relation_id())\nassert rel2.relation_id == rel.relation_id + 1 \n``` \n\nIf you don't do this, and pass both relations into a `State`, you will trigger a consistency checker error.\n\n### Additional event parameters\n\nAll relation events have some additional metadata that does not belong in the Relation object, such as, for a\nrelation-joined event, the name of the (remote) unit that is joining the relation. That is what determines what\n`ops.model.Unit` you get when you get `RelationJoinedEvent().unit` in an event handler.\n\nIn order to supply this parameter, you will have to **call** the event object and pass as `remote_unit_id` the id of the\nremote unit that the event is about. The reason that this parameter is not supplied to `Relation` but to relation\nevents, is that the relation already ties 'this app' to some 'remote app' (cfr. the `Relation.remote_app_name` attr),\nbut not to a specific unit. What remote unit this event is about is not a `State` concern but an `Event` one.\n\nThe `remote_unit_id` will default to the first ID found in the relation's `remote_units_data`, but if the test you are\nwriting is close to that domain, you should probably override it and pass it manually.\n\n```python\nfrom scenario import Relation, Event\n\nrelation = Relation(endpoint=\"foo\", interface=\"bar\")\nremote_unit_2_is_joining_event = relation.joined_event(remote_unit_id=2)\n\n# which is syntactic sugar for:\nremote_unit_2_is_joining_event = Event('foo-relation-changed', relation=relation, relation_remote_unit_id=2)\n```\n\n### Networks\n\nSimplifying a bit the Juju \"spaces\" model, each integration endpoint a charm defines in its metadata is associated with a network. Regardless of whether there is a living relation over that endpoint, that is. \n\nIf your charm has a relation `\"foo\"` (defined in metadata.yaml), then the charm will be able at runtime to do `self.model.get_binding(\"foo\").network`.\nThe network you'll get by doing so is heavily defaulted (see `state.Network.default`) and good for most use-cases because the charm should typically not be concerned about what IP it gets. \n\nOn top of the relation-provided network bindings, a charm can also define some `extra-bindings` in its metadata.yaml and access them at runtime. Note that this is a deprecated feature that should not be relied upon. For completeness, we support it in Scenario.\n\nIf you want to, you can override any of these relation or extra-binding associated networks with a custom one by passing it to `State.networks`.\n\n```python\nfrom scenario import State, Network\nstate = State(networks={\n 'foo': Network.default(private_address='4.4.4.4')\n})\n```\n\nWhere `foo` can either be the name of an `extra-bindings`-defined binding, or a relation endpoint.\n\n# Containers\n\nWhen testing a kubernetes charm, you can mock container interactions. When using the null state (`State()`), there will\nbe no containers. So if the charm were to `self.unit.containers`, it would get back an empty dict.\n\nTo give the charm access to some containers, you need to pass them to the input state, like so:\n`State(containers=[...])`\n\nAn example of a state including some containers:\n\n```python\nfrom scenario.state import Container, State\n\nstate = State(containers=[\n Container(name=\"foo\", can_connect=True),\n Container(name=\"bar\", can_connect=False)\n])\n```\n\nIn this case, `self.unit.get_container('foo').can_connect()` would return `True`, while for 'bar' it would give `False`.\n\n### Container filesystem setup\nYou can configure a container to have some files in it:\n\n```python\nfrom pathlib import Path\n\nfrom scenario.state import Container, State, Mount\n\nlocal_file = Path('/path/to/local/real/file.txt')\n\ncontainer = Container(name=\"foo\", can_connect=True, mounts={'local': Mount('/local/share/config.yaml', local_file)})\nstate = State(containers=[container])\n```\n\nIn this case, if the charm were to:\n\n```python\ndef _on_start(self, _):\n foo = self.unit.get_container('foo')\n content = foo.pull('/local/share/config.yaml').read()\n```\n\nthen `content` would be the contents of our locally-supplied `file.txt`. You can use `tempdir` for nicely wrapping\ndata and passing it to the charm via the container.\n\n`container.push` works similarly, so you can write a test like:\n\n```python\nimport tempfile\nfrom ops.charm import CharmBase\nfrom scenario import State, Container, Mount, Context\n\n\nclass MyCharm(CharmBase):\n def __init__(self, *args):\n super().__init__(*args)\n self.framework.observe(self.on.foo_pebble_ready, self._on_pebble_ready)\n\n def _on_pebble_ready(self, _):\n foo = self.unit.get_container('foo')\n foo.push('/local/share/config.yaml', \"TEST\", make_dirs=True)\n\n\ndef test_pebble_push():\n with tempfile.NamedTemporaryFile() as local_file:\n container = Container(name='foo',\n can_connect=True,\n mounts={'local': Mount('/local/share/config.yaml', local_file.name)})\n state_in = State(\n containers=[container]\n )\n Context(\n MyCharm,\n meta={\"name\": \"foo\", \"containers\": {\"foo\": {}}}\n ).run(\n container.pebble_ready_event(),\n state_in,\n )\n assert local_file.read().decode() == \"TEST\"\n```\n\n`container.pebble_ready_event` is syntactic sugar for: `Event(\"foo-pebble-ready\", container=container)`. The reason we\nneed to associate the container with the event is that the Framework uses an envvar to determine which container the\npebble-ready event is about (it does not use the event name). Scenario needs that information, similarly, for injecting\nthat envvar into the charm's runtime.\n\n## Container filesystem post-mortem\nIf the charm writes files to a container (to a location you didn't Mount as a temporary folder you have access to), you will be able to inspect them using the `get_filesystem` api.\n\n```python\nfrom ops.charm import CharmBase\nfrom scenario import State, Container, Mount, Context\n\n\nclass MyCharm(CharmBase):\n def __init__(self, *args):\n super().__init__(*args)\n self.framework.observe(self.on.foo_pebble_ready, self._on_pebble_ready)\n\n def _on_pebble_ready(self, _):\n foo = self.unit.get_container('foo')\n foo.push('/local/share/config.yaml', \"TEST\", make_dirs=True)\n\n\ndef test_pebble_push():\n container = Container(name='foo',\n can_connect=True)\n state_in = State(\n containers=[container]\n )\n ctx = Context(\n MyCharm,\n meta={\"name\": \"foo\", \"containers\": {\"foo\": {}}}\n )\n \n ctx.run(\"start\", state_in)\n\n # this is the root of the simulated container filesystem. Any mounts will be symlinks in it.\n container_root_fs = container.get_filesystem(ctx)\n cfg_file = container_root_fs / 'local' / 'share' / 'config.yaml'\n assert cfg_file.read_text() == \"TEST\"\n```\n\n## `Container.exec` mocks\n\n`container.exec` is a tad more complicated, but if you get to this low a level of simulation, you probably will have far\nworse issues to deal with. You need to specify, for each possible command the charm might run on the container, what the\nresult of that would be: its return code, what will be written to stdout/stderr.\n\n```python\nfrom ops.charm import CharmBase\n\nfrom scenario import State, Container, ExecOutput, Context\n\nLS_LL = \"\"\"\n.rw-rw-r-- 228 ubuntu ubuntu 18 jan 12:05 -- charmcraft.yaml\n.rw-rw-r-- 497 ubuntu ubuntu 18 jan 12:05 -- config.yaml\n.rw-rw-r-- 900 ubuntu ubuntu 18 jan 12:05 -- CONTRIBUTING.md\ndrwxrwxr-x - ubuntu ubuntu 18 jan 12:06 -- lib\n\"\"\"\n\n\nclass MyCharm(CharmBase):\n def _on_start(self, _):\n foo = self.unit.get_container('foo')\n proc = foo.exec(['ls', '-ll'])\n stdout, _ = proc.wait_output()\n assert stdout == LS_LL\n\n\ndef test_pebble_exec():\n container = Container(\n name='foo',\n exec_mock={\n ('ls', '-ll'): # this is the command we're mocking\n ExecOutput(return_code=0, # this data structure contains all we need to mock the call.\n stdout=LS_LL)\n }\n )\n state_in = State(\n containers=[container]\n )\n state_out = Context(\n MyCharm,\n meta={\"name\": \"foo\", \"containers\": {\"foo\": {}}},\n ).run(\n container.pebble_ready_event,\n state_in,\n )\n```\n\n# Storage\n\nIf your charm defines `storage` in its metadata, you can use `scenario.state.Storage` to instruct Scenario to make (mocked) filesystem storage available to the charm at runtime.\n\nUsing the same `get_filesystem` API as `Container`, you can access the tempdir used by Scenario to mock the filesystem root before and after the scenario runs.\n\n```python\nfrom scenario import Storage, Context, State\n# some charm with a 'foo' filesystem-type storage defined in metadata.yaml \nctx = Context(MyCharm)\nstorage = Storage(\"foo\")\n# setup storage with some content\n(storage.get_filesystem(ctx) / \"myfile.txt\").write_text(\"helloworld\")\n\nwith ctx.manager(\"update-status\", State(storage=[storage])) as mgr:\n foo = mgr.charm.model.storages[\"foo\"][0]\n loc = foo.location\n path = loc / \"myfile.txt\"\n assert path.exists()\n assert path.read_text() == \"helloworld\"\n\n myfile = loc / \"path.py\"\n myfile.write_text(\"helloworlds\")\n\n# post-mortem: inspect fs contents.\nassert (\n storage.get_filesystem(ctx) / \"path.py\"\n).read_text() == \"helloworlds\"\n```\n\nNote that State only wants to know about **attached** storages. A storage which is not attached to the charm can simply be omitted from State and the charm will be none the wiser.\n\n## Storage-add\n\nIf a charm requests adding more storage instances while handling some event, you can inspect that from the `Context.requested_storage` API.\n\n```python\n# in MyCharm._on_foo:\n# the charm requests two new \"foo\" storage instances to be provisioned \nself.model.storages.request(\"foo\", 2)\n```\n\nFrom test code, you can inspect that:\n\n```python\nfrom scenario import Context, State\n\nctx = Context(MyCharm)\nctx.run('some-event-that-will-cause_on_foo-to-be-called', State())\n\n# the charm has requested two 'foo' storages to be provisioned\nassert ctx.requested_storages['foo'] == 2\n```\n\nRequesting storages has no other consequence in Scenario. In real life, this request will trigger Juju to provision the storage and execute the charm again with `foo-storage-attached`.\nSo a natural follow-up Scenario test suite for this case would be:\n\n```python\nfrom scenario import Context, State, Storage\n\nctx = Context(MyCharm)\nfoo_0 = Storage('foo')\n# the charm is notified that one of the storages it has requested is ready\nctx.run(foo_0.attached_event, State(storage=[foo_0]))\n\nfoo_1 = Storage('foo')\n# the charm is notified that the other storage is also ready\nctx.run(foo_1.attached_event, State(storage=[foo_0, foo_1]))\n```\n\n\n# Ports\n\nSince `ops 2.6.0`, charms can invoke the `open-port`, `close-port`, and `opened-ports` hook tools to manage the ports opened on the host vm/container. Using the `State.opened_ports` api, you can: \n\n- simulate a charm run with a port opened by some previous execution\n```python\nfrom scenario import State, Port, Context\n\nctx = Context(MyCharm)\nctx.run(\"start\", State(opened_ports=[Port(\"tcp\", 42)]))\n```\n- assert that a charm has called `open-port` or `close-port`:\n```python\nfrom scenario import State, Port, Context\n\nctx = Context(MyCharm)\nstate1 = ctx.run(\"start\", State())\nassert state1.opened_ports == [Port(\"tcp\", 42)]\n\nstate2 = ctx.run(\"stop\", state1)\nassert state2.opened_ports == []\n```\n\n\n# Secrets\n\nScenario has secrets. Here's how you use them.\n\n```python\nfrom scenario import State, Secret\n\nstate = State(\n secrets=[\n Secret(\n id='foo',\n contents={0: {'key': 'public'}}\n )\n ]\n)\n```\n\nThe only mandatory arguments to Secret are its secret ID (which should be unique) and its 'contents': that is, a mapping\nfrom revision numbers (integers) to a `str:str` dict representing the payload of the revision.\n\nThere are three cases:\n- the secret is owned by this app but not this unit, in which case this charm can only manage it if we are the leader\n- the secret is owned by this unit, in which case this charm can always manage it (leader or not)\n- (default) the secret is not owned by this app nor unit, which means we can't manage it but only view it\n\nThus by default, the secret is not owned by **this charm**, but, implicitly, by some unknown 'other charm', and that other charm has granted us view rights.\n\n\nThe presence of the secret in `State.secrets` entails that we have access to it, either as owners or as grantees. Therefore, if we're not owners, we must be grantees. Absence of a Secret from the known secrets list means we are not entitled to obtaining it in any way. The charm, indeed, shouldn't even know it exists.\n\n[note]\nIf this charm does not own the secret, but also it was not granted view rights by the (remote) owner, you model this in Scenario by _not adding it to State.secrets_! The presence of a `Secret` in `State.secrets` means, in other words, that the charm has view rights (otherwise, why would we put it there?). If the charm owns the secret, or is leader, it will _also_ have manage rights on top of view ones.\n[/note]\n\nTo specify a secret owned by this unit (or app):\n\n```python\nfrom scenario import State, Secret\n\nstate = State(\n secrets=[\n Secret(\n id='foo',\n contents={0: {'key': 'private'}},\n owner='unit', # or 'app'\n remote_grants={0: {\"remote\"}}\n # the secret owner has granted access to the \"remote\" app over some relation with ID 0\n )\n ]\n)\n```\n\nTo specify a secret owned by some other application and give this unit (or app) access to it:\n\n```python\nfrom scenario import State, Secret\n\nstate = State(\n secrets=[\n Secret(\n id='foo',\n contents={0: {'key': 'public'}},\n # owner=None, which is the default\n revision=0, # the revision that this unit (or app) is currently tracking\n )\n ]\n)\n```\n\n# Actions\n\nAn action is a special sort of event, even though `ops` handles them almost identically.\nIn most cases, you'll want to inspect the 'results' of an action, or whether it has failed or\nlogged something while executing. Many actions don't have a direct effect on the output state.\nFor this reason, the output state is less prominent in the return type of `Context.run_action`.\n\nHow to test actions with scenario:\n\n## Actions without parameters\n\n```python\nfrom scenario import Context, State, ActionOutput\nfrom charm import MyCharm\n\n\ndef test_backup_action():\n ctx = Context(MyCharm)\n\n # If you didn't declare do_backup in the charm's `actions.yaml`, \n # the `ConsistencyChecker` will slap you on the wrist and refuse to proceed.\n out: ActionOutput = ctx.run_action(\"do_backup_action\", State())\n\n # you can assert action results, logs, failure using the ActionOutput interface\n assert out.logs == ['baz', 'qux']\n \n if out.success:\n # if the action did not fail, we can read the results:\n assert out.results == {'foo': 'bar'}\n\n else:\n # if the action fails, we can read a failure message\n assert out.failure == 'boo-hoo'\n```\n\n## Parametrized Actions\n\nIf the action takes parameters, you'll need to instantiate an `Action`.\n\n```python\nfrom scenario import Action, Context, State, ActionOutput\nfrom charm import MyCharm\n\n\ndef test_backup_action():\n # define an action\n action = Action('do_backup', params={'a': 'b'})\n ctx = Context(MyCharm)\n\n # if the parameters (or their type) don't match what declared in actions.yaml, \n # the `ConsistencyChecker` will slap you on the other wrist. \n out: ActionOutput = ctx.run_action(action, State())\n\n # ...\n```\n\n# Deferred events\n\nScenario allows you to accurately simulate the Operator Framework's event queue. The event queue is responsible for\nkeeping track of the deferred events. On the input side, you can verify that if the charm triggers with this and that\nevent in its queue (they would be there because they had been deferred in the previous run), then the output state is\nvalid.\n\n```python\nfrom scenario import State, deferred, Context\n\n\nclass MyCharm(...):\n ...\n\n def _on_update_status(self, e):\n e.defer()\n\n def _on_start(self, e):\n e.defer()\n\n\ndef test_start_on_deferred_update_status(MyCharm):\n \"\"\"Test charm execution if a 'start' is dispatched when in the previous run an update-status had been deferred.\"\"\"\n state_in = State(\n deferred=[\n deferred('update_status',\n handler=MyCharm._on_update_status)\n ]\n )\n state_out = Context(MyCharm).run('start', state_in)\n assert len(state_out.deferred) == 1\n assert state_out.deferred[0].name == 'start'\n```\n\nYou can also generate the 'deferred' data structure (called a DeferredEvent) from the corresponding Event (and the\nhandler):\n\n```python\nfrom scenario import Event, Relation\n\n\nclass MyCharm(...):\n ...\n\n\ndeferred_start = Event('start').deferred(MyCharm._on_start)\ndeferred_install = Event('install').deferred(MyCharm._on_start)\n```\n\n## relation events:\n\n```python\nfoo_relation = Relation('foo')\ndeferred_relation_changed_evt = foo_relation.changed_event.deferred(handler=MyCharm._on_foo_relation_changed)\n```\n\nOn the output side, you can verify that an event that you expect to have been deferred during this trigger, has indeed\nbeen deferred.\n\n```python\nfrom scenario import State, Context\n\n\nclass MyCharm(...):\n ...\n\n def _on_start(self, e):\n e.defer()\n\n\ndef test_defer(MyCharm):\n out = Context(MyCharm).run('start', State())\n assert len(out.deferred) == 1\n assert out.deferred[0].name == 'start'\n```\n\n## Deferring relation events\n\nIf you want to test relation event deferrals, some extra care needs to be taken. RelationEvents hold references to the\nRelation instance they are about. So do they in Scenario. You can use the deferred helper to generate the data\nstructure:\n\n```python\nfrom scenario import State, Relation, deferred\n\n\nclass MyCharm(...):\n ...\n\n def _on_foo_relation_changed(self, e):\n e.defer()\n\n\ndef test_start_on_deferred_update_status(MyCharm):\n foo_relation = Relation('foo')\n State(\n relations=[foo_relation],\n deferred=[\n deferred('foo_relation_changed',\n handler=MyCharm._on_foo_relation_changed,\n relation=foo_relation)\n ]\n )\n```\n\nbut you can also use a shortcut from the relation event itself, as mentioned above:\n\n```python\n\nfrom scenario import Relation\n\n\nclass MyCharm(...):\n ...\n\n\nfoo_relation = Relation('foo')\nfoo_relation.changed_event.deferred(handler=MyCharm._on_foo_relation_changed)\n```\n\n### Fine-tuning\n\nThe deferred helper Scenario provides will not support out of the box all custom event subclasses, or events emitted by\ncharm libraries or objects other than the main charm class.\n\nFor general-purpose usage, you will need to instantiate DeferredEvent directly.\n\n```python\nfrom scenario import DeferredEvent\n\nmy_deferred_event = DeferredEvent(\n handle_path='MyCharm/MyCharmLib/on/database_ready[1]',\n owner='MyCharmLib', # the object observing the event. Could also be MyCharm.\n observer='_on_database_ready'\n)\n```\n\n# StoredState\n\nScenario can simulate StoredState. You can define it on the input side as:\n\n```python\nfrom ops.charm import CharmBase\nfrom ops.framework import StoredState as Ops_StoredState, Framework\nfrom scenario import State, StoredState\n\n\nclass MyCharmType(CharmBase):\n my_stored_state = Ops_StoredState()\n\n def __init__(self, framework: Framework):\n super().__init__(framework)\n assert self.my_stored_state.foo == 'bar' # this will pass!\n\n\nstate = State(stored_state=[\n StoredState(\n owner_path=\"MyCharmType\",\n name=\"my_stored_state\",\n content={\n 'foo': 'bar',\n 'baz': {42: 42},\n })\n])\n```\n\nAnd the charm's runtime will see `self.stored_State.foo` and `.baz` as expected. Also, you can run assertions on it on\nthe output side the same as any other bit of state.\n\n# Resources\n\nIf your charm requires access to resources, you can make them available to it through `State.resources`.\nFrom the perspective of a 'real' deployed charm, if your charm _has_ resources defined in `metadata.yaml`, they _must_ be made available to the charm. That is a Juju-enforced constraint: you can't deploy a charm without attaching all resources it needs to it.\nHowever, when testing, this constraint is unnecessarily strict (and it would also mean the great majority of all existing tests would break) since a charm will only notice that a resource is not available when it explicitly asks for it, which not many charms do.\n\nSo, the only consistency-level check we enforce in Scenario when it comes to resource is that if a resource is provided in State, it needs to have been declared in metadata.\n\n```python\nfrom scenario import State, Context\nctx = Context(MyCharm, meta={'name': 'juliette', \"resources\": {\"foo\": {\"type\": \"oci-image\"}}})\nwith ctx.manager(\"start\", State(resources={'foo': '/path/to/resource.tar'})) as mgr:\n # if the charm, at runtime, were to call self.model.resources.fetch(\"foo\"), it would get '/path/to/resource.tar' back.\n path = mgr.charm.model.resources.fetch('foo')\n assert path == '/path/to/resource.tar' \n```\n\n# Emitting custom events\n\nWhile the main use case of Scenario is to emit juju events, i.e. the built-in `start`, `install`, `*-relation-changed`,\netc..., it can be sometimes handy to directly trigger custom events defined on arbitrary Objects in your hierarchy.\n\nSuppose your charm uses a charm library providing an `ingress_provided` event.\nThe 'proper' way to emit it is to run the event that causes that custom event to be emitted by the library, whatever\nthat may be, for example a `foo-relation-changed`.\n\nHowever, that may mean that you have to set up all sorts of State and mocks so that the right preconditions are met and\nthe event is emitted at all.\n\nIf for whatever reason you don't want to do that and you attempt to run that event directly you will get an error:\n\n```python\nfrom scenario import Context, State\n\nContext(...).run(\"ingress_provided\", State()) # raises scenario.ops_main_mock.NoObserverError\n```\n\nThis happens because the framework, by default, searches for an event source named `ingress_provided` in `charm.on`, but\nsince the event is defined on another Object, it will fail to find it.\n\nYou can prefix the event name with the path leading to its owner to tell Scenario where to find the event source:\n\n```python\nfrom scenario import Context, State\n\nContext(...).run(\"my_charm_lib.on.foo\", State())\n```\n\nThis will instruct Scenario to emit `my_charm.my_charm_lib.on.foo`.\n\n(always omit the 'root', i.e. the charm framework key, from the path)\n\n# Live charm introspection\n\nScenario is a black-box, state-transition testing framework. It makes it trivial to assert that a status went from A to\nB, but not to assert that, in the context of this charm execution, with this state, a certain charm-internal method was called and returned a\ngiven piece of data, or would return this and that _if_ it had been called.\n\nScenario offers a cheekily-named context manager for this use case specifically:\n\n```python\nfrom ops import CharmBase, StoredState\n\nfrom charms.bar.lib_name.v1.charm_lib import CharmLib\nfrom scenario import Context, State\n\n\nclass MyCharm(CharmBase):\n META = {\"name\": \"mycharm\"}\n _stored = StoredState()\n \n def __init__(self, framework):\n super().__init__(framework)\n self._stored.set_default(a=\"a\")\n self.my_charm_lib = CharmLib()\n framework.observe(self.on.start, self._on_start)\n\n def _on_start(self, event):\n self._stored.a = \"b\"\n\n\ndef test_live_charm_introspection(mycharm):\n ctx = Context(mycharm, meta=mycharm.META)\n # If you want to do this with actions, you can use `Context.action_manager` instead.\n with ctx.manager(\"start\", State()) as manager:\n # this is your charm instance, after ops has set it up\n charm: MyCharm = manager.charm\n \n # we can check attributes on nested Objects or the charm itself \n assert charm.my_charm_lib.foo == \"foo\"\n # such as stored state\n assert charm._stored.a == \"a\"\n\n # this will tell ops.main to proceed with normal execution and emit the \"start\" event on the charm\n state_out = manager.run()\n \n # after that is done, we are handed back control, and we can again do some introspection\n assert charm.my_charm_lib.foo == \"bar\"\n # and check that the charm's internal state is as we expect\n assert charm._stored.a == \"b\"\n\n # state_out is, as in regular scenario tests, a State object you can assert on:\n assert state_out.unit_status == ...\n```\n\nNote that you can't call `manager.run()` multiple times: the manager is a context that ensures that `ops.main` 'pauses' right\nbefore emitting the event to hand you some introspection hooks, but for the rest this is a regular scenario test: you\ncan't emit multiple events in a single charm execution.\n\n# The virtual charm root\n\nBefore executing the charm, Scenario copies the charm's `/src`, any libs, the metadata, config, and actions `yaml`s to a temporary directory. The\ncharm will see that tempdir as its 'root'. This allows us to keep things simple when dealing with metadata that can be\neither inferred from the charm type being passed to `Context` or be passed to it as an argument, thereby overriding\nthe inferred one. This also allows you to test charms defined on the fly, as in:\n\n```python\nfrom ops.charm import CharmBase\nfrom scenario import State, Context\n\n\nclass MyCharmType(CharmBase):\n pass\n\n\nctx = Context(charm_type=MyCharmType,\n meta={'name': 'my-charm-name'})\nctx.run('start', State())\n```\n\nA consequence of this fact is that you have no direct control over the tempdir that we are creating to put the metadata\nyou are passing to `.run()` (because `ops` expects it to be a file...). That is, unless you pass your own:\n\n```python\nfrom ops.charm import CharmBase\nfrom scenario import State, Context\nimport tempfile\n\n\nclass MyCharmType(CharmBase):\n pass\n\n\ntd = tempfile.TemporaryDirectory()\nstate = Context(\n charm_type=MyCharmType,\n meta={'name': 'my-charm-name'},\n charm_root=td.name\n).run('start', State())\n```\n\nDo this, and you will be able to set up said directory as you like before the charm is run, as well as verify its\ncontents after the charm has run. Do keep in mind that any metadata files you create in it will be overwritten by Scenario, and therefore\nignored, if you pass any metadata keys to `Context`. Omit `meta` in the call\nabove, and Scenario will instead attempt to read `metadata.yaml` from the\ntemporary directory.\n\n\n# Immutability\n\nAll of the data structures in `state`, e.g. `State, Relation, Container`, etc... are immutable (implemented as frozen\ndataclasses).\n\nThis means that all components of the state that goes into a `context.run()` call are not mutated by the call, and the\nstate that you obtain in return is a different instance, and all parts of it have been (deep)copied.\nThis ensures that you can do delta-based comparison of states without worrying about them being mutated by scenario.\n\nIf you want to modify any of these data structures, you will need to either reinstantiate it from scratch, or use\nthe `replace` api.\n\n```python\nfrom scenario import Relation\n\nrelation = Relation('foo', remote_app_data={\"1\": \"2\"})\n# make a copy of relation, but with remote_app_data set to {\"3\", \"4\"} \nrelation2 = relation.replace(remote_app_data={\"3\", \"4\"})\n```\n\n# Consistency checks\n\nA Scenario, that is, the combination of an event, a state, and a charm, is consistent if it's plausible in JujuLand. For\nexample, Juju can't emit a `foo-relation-changed` event on your charm unless your charm has declared a `foo` relation\nendpoint in its `metadata.yaml`. If that happens, that's a juju bug. Scenario however assumes that Juju is bug-free,\ntherefore, so far as we're concerned, that can't happen, and therefore we help you verify that the scenarios you create\nare consistent and raise an exception if that isn't so.\n\nThat happens automatically behind the scenes whenever you trigger an event;\n`scenario.consistency_checker.check_consistency` is called and verifies that the scenario makes sense.\n\n## Caveats:\n\n- False positives: not all checks are implemented yet; more will come.\n- False negatives: it is possible that a scenario you know to be consistent is seen as inconsistent. That is probably a\n bug in the consistency checker itself, please report it.\n- Inherent limitations: if you have a custom event whose name conflicts with a builtin one, the consistency constraints\n of the builtin one will apply. For example: if you decide to name your custom event `bar-pebble-ready`, but you are\n working on a machine charm or don't have either way a `bar` container in your `metadata.yaml`, Scenario will flag that\n as inconsistent.\n\n## Bypassing the checker\n\nIf you have a clear false negative, are explicitly testing 'edge', inconsistent situations, or for whatever reason the\nchecker is in your way, you can set the `SCENARIO_SKIP_CONSISTENCY_CHECKS` envvar and skip it altogether. Hopefully you\ndon't need that.\n\n# Jhack integrations\n\nUp until `v5.6.0`, `scenario` shipped with a cli tool called `snapshot`, used to interact with a live charm's state.\nThe functionality [has been moved over to `jhack`](https://github.com/PietroPasotti/jhack/pull/111), \nto allow us to keep working on it independently, and to streamline \nthe profile of `scenario` itself as it becomes more broadly adopted and ready for widespread usage.\n\n",
"bugtrack_url": null,
"license": "Apache-2.0",
"summary": "Python library providing a state-transition testing API for Operator Framework charms.",
"version": "6.0.3",
"project_urls": {
"Bug Tracker": "https://github.com/canonical/ops-scenario/issues",
"Homepage": "https://github.com/canonical/ops-scenario"
},
"split_keywords": [
"juju",
"test"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "1628f059772616e3aa43549351fc9acfa01ae842d2558d652f3ad1d435200087",
"md5": "c7f66e69abf6add561b85eeec2dd2141",
"sha256": "94215895aea55410a4b5eda222ab22284e6921f61c9f3cc6b77eae09e9dc5087"
},
"downloads": -1,
"filename": "ops_scenario-6.0.3-py3-none-any.whl",
"has_sig": false,
"md5_digest": "c7f66e69abf6add561b85eeec2dd2141",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.8",
"size": 63926,
"upload_time": "2024-03-19T10:40:30",
"upload_time_iso_8601": "2024-03-19T10:40:30.101478Z",
"url": "https://files.pythonhosted.org/packages/16/28/f059772616e3aa43549351fc9acfa01ae842d2558d652f3ad1d435200087/ops_scenario-6.0.3-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "4aa7b3774d2354dffb506ad3e56016c43b9b6f8c712786113c34cea1e7700089",
"md5": "0e88fde527942a800782ec9c53ad2876",
"sha256": "b729b7bb4677ff271bc5696e99a54b56c2be04119f3daaeaedbefa904625db1b"
},
"downloads": -1,
"filename": "ops-scenario-6.0.3.tar.gz",
"has_sig": false,
"md5_digest": "0e88fde527942a800782ec9c53ad2876",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.8",
"size": 162088,
"upload_time": "2024-03-19T10:40:32",
"upload_time_iso_8601": "2024-03-19T10:40:32.006170Z",
"url": "https://files.pythonhosted.org/packages/4a/a7/b3774d2354dffb506ad3e56016c43b9b6f8c712786113c34cea1e7700089/ops-scenario-6.0.3.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-03-19 10:40:32",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "canonical",
"github_project": "ops-scenario",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"tox": true,
"lcname": "ops-scenario"
}
