Most of the development work is towards creating standard plugins. A plugin usually does three things:
-
It has a
layout
property, consisting of multiple Dash components. -
It has/uses functionality for getting/processing the data it displays/visualizes.
-
It sets some callbacks to add user interactivity triggering actions in the Python backend.
Of the three things mentioned above, it is only the layout
proprety that is
mandatory to provide. A minimal plugin could look like:
from dash import html
from webviz_config import WebvizPluginABC
class ExamplePlugin(WebvizPluginABC):
@property
def layout(self):
return html.Div([
html.H1('This is a static title'),
'And this is just some ordinary text'
])
If the file containing ExamplePlugin
is saved to ./webviz_config/plugins,
and then added to the corresponding __init__ file
you are done. Alternatively you can create your plugins in a separate Python project and setup.py
.
You can then configure the installation by using something like:
setup(
...
entry_points={
"webviz_config_plugins": [
"ExamplePlugin = my_package.my_module:ExamplePlugin"
]
},
...
)
See webviz-subsurface for example of this usage.
After installation, the user can then include the plugin through a configuration file, e.g.
title: Simple Webviz example
pages:
- title: Front page
content:
- ExamplePlugin:
In the generated webviz application, your plugin will as default be given a set of toolbar buttons in a settings drawer. The default buttons to appear are fullscreen and screenshot. Additional buttons appears based on provided information - as plugin author contact information, download data, guided tour information and issue feedback link. See this section for more information regarding downloading plugin data.
If you want to include user interactivity which triggers actions in the Python backend, you can add callbacks. A simple example of this is given below.
from uuid import uuid4
from dash import html, Input, Output
from webviz_config import WebvizPluginABC
class ExamplePlugin(WebvizPluginABC):
def __init__(self, app):
super().__init__()
self.button_id = f'submit-button-{uuid4()}'
self.div_id = f'output-state-{uuid4()}'
self.set_callbacks(app)
@property
def layout(self):
return html.Div([
html.H1('This is a static title'),
html.Button(id=self.button_id, n_clicks=0,
children='Submit'),
html.Div(id=self.div_id)
])
def set_callbacks(self, app):
@app.callback(Output(self.div_id, 'children'),
[Input(self.button_id, 'n_clicks')])
def _update_output(n_clicks):
return f'Button has been pressed {n_clicks} times.'
Callback functions are decorated with @app.callback
. For introduction to
callbacks, see the dash documentation.
There are three fundamental additions to the minimal example without callbacks:
-
You add the argument
app
to your__init__
function. This is a special argument name which will not be originating from the user configuration file, but rather automatically given to the plugin by the core functionality ofwebviz-config
. -
You add a class function
set_callbacks
which contains the different callbacks to add. This function is called from the__init__
function, such that the callbacks are set when the plugin instance is created. -
Since the components are reusable (i.e. a user can use the plugin multiple times within the same application), the plugin IDs mentioned in the
@app.callback(...)
decorator needs to be unique. One simple way of ensuring this is to create unique IDs in the__init__
function using uuid.uuid4(), as demonstrated in the example above.
There is a data download button provided by
the WebvizPluginABC
class. However, it will only appear if the corresponding
callback is set. A typical compressed data download callback will look like
@app.callback(self.plugin_data_output,
self.plugin_data_requested)
def _user_download_data(data_requested):
return (
WebvizPluginABC.plugin_compressed_data(
filename="webviz-data.zip",
content=[{"filename": "some_file.txt", "content": "Some download data"}],
)
if data_requested
else None
)
A typical CSV data download from e.g. a pandas.DataFrame
will look like:
@app.callback(self.plugin_data_output,
self.plugin_data_requested)
def _user_download_data(data_requested):
return (
{
"filename": "some-file.csv",
"content": base64.b64encode(
some_pandas_dataframe.to_csv().encode()
).decode("ascii"),
"mime_type": "text/csv",
}
if data_requested
else None
)
By letting the plugin define the callback, the plugin author is able to utilize the whole callback machinery, including e.g. state of the individual components in the plugin. This way the data downloaded can e.g. depend on the visual state or user selection.
The attributes self.plugin_data_output
and self.plugin_data_requested
are Dash Output
and Input
instances respectively, and are provided by
the base class WebvizPluginABC
(i.e. include them as shown here).
The function WebvizPluginABC.plugin_compressed_data
is a utility function
which takes a file name and a list of dictionaries, containing file names and corresponding data,
and compresses them to a zip archive which is then downloaded by the user.
Since the plugins are reusable and generic, they usually take in some user provided arguments. A minimal example could look like:
from dash import html
from webviz_config import WebvizPluginABC
class ExamplePlugin(WebvizPluginABC):
def __init__(self, title: str, number: int=42):
super().__init__()
self.title = title
self.number = number
@property
def layout(self):
return html.Div([
html.H1(self.title),
f'This is your number: {self.number}'
])
The user given configuration file can now look like this:
title: Simple Webviz example
pages:
- title: Front page
content:
- ExamplePlugin:
title: My special title
The core functionality of webviz-config
will provide user friendly
error messages in the following scenarios:
-
A required argument is not provided (in example above, that would be if
title
is not stated in the configuration file). -
An unknown argument is given (that would be something else than
title
andnumber
in this case). -
Mismatch between type of user provided value, and the (optionally) provided type hint in the
__init__
function.
The __init__
function above could be replaced with an equivalent argument
specification without type hints, e.g.
def __init__(self, title, number=42):
However, the benefit of adding type hints is that webviz-config
will
provide an user friendly error message immediately when parsing the
configuration file, if there is a type mismatch. E.g., if the user
provided the configuration file
title: Simple Webviz example
pages:
- title: Front page
content:
- ExamplePlugin:
title: My special title
number: Some text instead of number
this error message will be given:
The value provided for argument `number` given to plugin `ExamplePlugin` is of type `str`. Expected type `int`
An additional benefit is that if the plugin author says an argument should
be of type pathlib.Path
, the configuration parser will make sure that
the user provided path (which is a string to begin with in configuration file,
potentially non-absolute and relative to the configuration file itself) is
given to the plugin as an absolute path, and of type pathlib.Path
.
The plugins get data input through ordinary Python functions.
Since these functions can be costly to call, we utilize the
flask-caching package.
By decorating the costly functions with @CACHE.memoize()
the result is cached, such that if the same function is called more than once,
within the timeout, the cached result will be used instead of starting
a new calculation.
Functionality used by multiple plugins should be put in a common module. The applications common cache instance can be imported using
from webviz_config.common_cache import CACHE
There are use cases where the generated webviz instance ideally is portable and self-contained. At the same time, there are use cases where the data input ideally comes from "live sources" (simulations still running, production database...).
Asking each plugin to take care of both these scenarios to the full extent
involves a lot of duplicate work. The core of webviz-config
therefore
facilitates this transition in the following way.
Assume you have a function:
import pandas as pd
def get_some_data(some, arguments) -> pd.DataFrame:
...
It takes in some possibly user given arguments, reads e.g. data somewhere from
the file system, and then returns a pd.DataFrame
. If we want
webviz-config
to facilitate the transition to a portable webviz instance,
the plugin author needs only do three things:
-
Import the decorator:
from webviz_config.webviz_store import webvizstore
-
Decorate the function as this:
@webvizstore def get_some_data(some, arguments) -> pd.DataFrame:
-
In the plugin class, define a class method
add_webvizstore
which returns a list of tuples. The first item in each tuple is a reference to a function, the second item is a list of argument combinations (formatted as dictionaries with'argument_name': value
pairs). Arguments with a default value in the function signature does not need to be provided in the argument dictionaries.The author of the plugin should provide a list of all the argument combinations that are imaginable during runtime of the application. Since it is a class method, the author has access to all the user provided arguments.
A full example could look like e.g.:
import pandas as pd
from webviz_config.webviz_store import webvizstore
from webviz_config.common_cache import CACHE
from webviz_config import WebvizPluginABC
class ExamplePortable(WebvizPluginABC):
def __init__(self, some_number: int):
super().__init__()
self.some_number = some_number
def add_webvizstore(self):
return [(input_data_function, [{'some_number': self.some_number}])]
@property
def layout(self):
return str(input_data_function(self.some_number))
@CACHE.memoize()
@webvizstore
def input_data_function(some_number) -> pd.DataFrame:
print("This time I'm actually doing the calculation...")
return pd.DataFrame(data={'col1': [some_number, some_number*2],
'col2': [some_number*3, some_number*4]})
Creating a portable webviz instance is done from the command line using
webviz build some_config.yaml --portable /some/outputfolder
The core of webviz-config
will do the following:
- If the user has not asked for a portable version (i.e. not given the
--portable
command line argument, the decorator@webvizstore
will not change the attached function. - If the user asks for a portable version, it will
- Before writing the actual dash code, it will run all decorated functions
with the given argument combinations. The resulting dataframes are stored
in a folder
./resources/webviz_storage
as parquet files. - It writes the webviz-dash code (as usual), but this the decorated
functions will return the dataframe from the stored
.parquet
files, instead of running the actual function code.
- Before writing the actual dash code, it will run all decorated functions
with the given argument combinations. The resulting dataframes are stored
in a folder
Currently only functions returning pd.DataFrame
, io.BytesIO
or pathlib.Path
are
supported. A NotImplementedError
will be raised if the function being
decorated with @webvizstore
does not have a type hint being supported.
If the return type is pathlib.Path
, the file corresponding to the path will
be copied over to the portable webviz
instance, and the decorated function
will return a pathlib.Path
instance pointing to the stored file. For most
use cases, the decorated function get_resource
in
webviz_config.webviz_store
can be imported and used.
📑 The argument hashing method used for
@webvizstore
is using the same principles as in theflask-caching
package. The input arguments to a function decorated with@webvizstore
could be mutable objects like e.g. instances of a class, but make sure that the object has a__repr__
function associated with it such that instances representing different input also have different string output from__repr__
.
🚀 If you nest decorations, e.g. use both
@webvizstore
and@CACHE.memoize
, following the same order as in the example above usually gives best performance.
If you create multiple plugins that have some settings in common, you can
"subscribe" to keys in the dictionary shared_settings
, defined by the user
in the configuration file. E.g. assume that the user enters something like this at
top level in the configuration file:
shared_settings:
some_key:
some_settings1: ...
some_settings2: ...
You can then early in the application loading process get to run a function checking, and potentially transforming, the settings. The latter is useful e.g. if you want to convert some string, representing a path relative to the configuration file, to an absolute path.
The way you subscribe to a shared setting is that you in your Python package add something like
import webviz_config
@webviz_config.SHARED_SETTINGS_SUBSCRIPTIONS.subscribe("some_key")
def subscribe(some_key, config_folder, portable):
# The input argument some_key given to this function is equal to
# shared_settings["some_key"] provided by the user from the configuration file.
# Do some checks on the settings?
# Do some transformation on the settings?
return some_key # The returned value here is put back into shared_settings["some_key"]
Stating the input arguments named config_folder
and/or portable
in the function
signature is not necessary, however if you do you will get a
pathlib.Path
instance representing the absolute path to the configuration file that was used, and/or
a boolean value stating if the Webviz application running is a portable one.
The (optionally transformed) shared_settings
can be retrieved in a plugin by adding
a specially named webviz_settings
argument to the plugin's __init__
function. The
webviz_settings
argument works similar to the app
argument in that it is a special
argument name that will not be originating from the configuration file, but will be
automatically given to the plugin by the core functionality of webviz-config.
Shared settings can then be accessed through webviz_settings
. E.g., in the case
above, the wanted settings are found as webviz_settings.shared_settings["some_key"]
as shown in the example below:
from webviz_config import WebvizPluginABC, WebvizSettings
class ExamplePlugin(WebvizPluginABC):
def __init__(self, app, webviz_settings: WebvizSettings, title: str, number: int=42):
super().__init__()
self.title = title
self.number = number
self.some_key = webviz_settings.shared_settings["some_key"]
self.set_callbacks(app)
It is possible to use OAuth 2.0 Authorization Code flow to secure a webviz
application.
In order to do so, add oauth2
attribute in a custom plugin with a boolean True
value.
The following is an example.
class OurCustomPlugin(WebvizPluginABC):
def __init__(self):
super().__init__()
self.use_oauth2 = True
@property
def oauth2(self):
return self.use_oauth2
Information related to the application for OAuth 2.0 has to be provided in environment
variables. These environment variables are WEBVIZ_TENANT_ID
, WEBVIZ_CLIENT_ID
,
WEBVIZ_CLIENT_SECRET
, WEBVIZ_SCOPE
.
The values can be found in the Azure AD configuration page. Short explanation of these environment variables:
WEBVIZ_TENANT_ID
: The organization's Azure tenant ID (Equinor has exactly one tenant ID).WEBVIZ_CLIENT_ID
: ID of the Webviz Azure AD app.WEBVIZ_CLIENT_SECRET
: Webviz Azure AD app's client secret.WEBVIZ_SCOPE
: The API permission for this Webviz Azure AD app.
If you are serving behind a proxy, you might need to configure trust for X-FORWARD headers. Internally, this is done by using a ProxyFix class, as described in the Flask docs. To enable the use of the ProxyFix class, set one or all of the following variables to an integer describing the number of trusted forwards:
WEBVIZ_X_FORWARDED_FOR
: Corresponds to x_for of the ProxyFix classWEBVIZ_X_FORWARDED_PROTO
: Corresponds to x_proto of the ProxyFix classWEBVIZ_X_FORWARDED_HOST
: Corresponds to x_host of the ProxyFix classWEBVIZ_X_FORWARDED_PORT
: Corresponds to x_port of the ProxyFix classWEBVIZ_X_FORWARDED_PREFIX
: Corresponds to x_prefix of the ProxyFix class
Any omitted argument will be set to 0.
Webviz will store the users oauth and refresh tokens in the flask session. To ensure that this session is treated consistently across server replicas, you should set the environment variable WEBVIZ_SESSION_SECRET_KEY
to the same value in all replicas.
To get the access token of the authenticated user in the flask session, use flask.session.get("access_token")
.
To get the expiration date of the token, use flask.session.get("expiration_date")
.
To run tests it is necessary to first install the selenium chrome driver. Then install the Python development requirements:
pip install .[tests]
You can then run the tests using
pytest tests --forked
Linting can be checked by:
black --check webviz_config tests
webviz-config
can automatically build documentation for all installed plugins. E.g.
the end user can get an overview of all installed plugins, and their arguments, by
running
webviz docs
in the terminal. Behind the scenes, webviz-config
will then create a
docsify
setup and open it localhost
.
The setup can also be deployed to e.g. GitHub Pages directly. To store the documentation output run
webviz docs --portable ./built_docs --skip-open
The --skip-open
argument is useful in a CI/CD setting, to prevent webviz-config
from automatically trying to open the created documentation in the browser.
Auto-built webviz
documentation will:
- Find all installed plugins.
- Group them according to top package name.
- Show a
YAML
snippet with the plugin argument options.- Arguments with default values will be presented with the default value, and be marked as optional.
- If an argument has a type annotation, that will be included in the documentation.
In addition, if the plugin class has a docstring, the content in the docstring will
be used as a short introduction to the plugin. If the docstring has several parts,
when split by a line containing only ---
, they will be used as follows:
- First part is the introduction to the plugin.
- Second part is a more detailed explanation of the plugin arguments.
- Third part is information regarding plugin data input. E.g assumptions, prerequisites and/or required/assumed data format.
Since docsify
is used behind the scenes, you can create information boxes, warning boxes
and use GitHub emojis in the plugin docstring.
See docsify
documentation for details.
KaTeX is also used behind the scenes, meaning that you can add
math (TeX syntax) to your docstrings and get it rendered in the auto-built
documentation. Remember that \
is an escape character in Python, i.e. either
escape it (\\
) or use raw strings:
class HistoryMatch(WebvizPluginABC):
r"""This is a docstring with some inline math $\alpha$ and some block math:
$$\alpha = \frac{\beta}{\gamma}$$
"""
Example of auto-built documentation for webviz-config
can be seen
here on github.
It is strongly recommended to store your plugin project code base in a git
solution,
e.g. in a new GitHub repository. If possible, it is also
highly recommended to let it be an open source repository. This has several advantages:
- Others can reuse your work - and help each other achieve better code quality, add more features and share maintenance
- You will from the beginning make sure you keep a good separation between visualization code and data loading/processing code.
- It becomes easier for your plugin users to use it in e.g. Docker and in cloud hosting.
webviz-config
will make sure that portable builds, which uses one (or more) plugins
from your plugin project, includes installation instructions for the project also in
the Docker build instructions. In order for webviz-config
to do this, you should add
project_urls
metadata
in the project's setup.py
. An example:
project_urls={
"Download": "https://pypi.org/project/webviz-config",
"Source": "https://github.com/equinor/webviz-config",
},
The following logic applies when the user creates a portable application:
- If
Download
is specified and the plugin project version installed is available on PyPI, the Dockerfile willpip install
the same version from PyPI. - Otherwise the Dockerfile will install from the
Source
url. From thesetuptools_scm
provided version, the correct commit/version/tag will be installed.
Note that if you are a developer and working on a fork, you might want to temporarily
override the Source
url to your fork. You can do this by specifying an environment
variable SOURCE_URL_NAME_PLUGIN_PROJECT=https://urlyourfork
. If you also want to
explicitly state git pointer/reference (thereby not use the one derived from
setuptools_scm
version) you can set the environment variable
GIT_POINTER_NAME_PLUGIN_PROJECT
.
For private repositories, a GitHub SSH deploy key will need to be provided to the Docker
build process (see instructions in README
created with the portable application).
Plugins can be marked as deprecated by using the @deprecated_plugin(deprecation_info)
decorator.
from webviz_config.deprecation_decorators import deprecated_plugin
@deprecated_plugin("An optional explanation of why the plugin has been deprecated.")
class MyPlugin(WebvizPluginABC):
...
Plugin arguments can be marked as deprecated by using the @deprecated_plugin_arguments(check={})
decorator in front of the __init__
function.
Arguments can either be marked as deprecated in any case (see MyPluginExample1
) or their values can be checked within a function (see MyPluginExample2
) which returns a tuple containing a short string shown to the end user in the app and a long string shown in the plugin's documentation.
from typing import Optional, Tuple
from webviz_config.deprecation_decorators import deprecated_plugin_arguments
class MyPluginExample1(WebvizPluginABC):
...
@deprecated_plugin_arguments(
{
"arg3": (
"Short message shown to the end user both in the app and documentation.",
(
"This can be a long message, which is shown only in the documentation, explaining "
"e.g. why it is deprecated and which plugin should be used instead."
)
)
}
)
def __init__(self, arg1: str, arg2: int, arg3: Optional[int] = None):
...
def check_deprecation_example2(arg1: int, arg3: int) -> Optional[Tuple[str, str]]:
if arg3 == arg1:
return ("This message is shown to the end user in the app.", "This message is shown in the documentation of the plugin.")
return None
class MyPluginExample2(WebvizPluginABC):
...
@deprecated_plugin_arguments(check_deprecation_example2)
def __init__(self, arg1: int, arg2: int, arg3: int):
...