sartography
/
spiff-arena
mirror of https://github.com/sartography/spiff-arena.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Squashed 'SpiffWorkflow/' changes from 63db3e459..9d597627b 

9d597627b Merge pull request #250 from sartography/feature/documentation-updates-for-1.2
896447557 Merge pull request #249 from sartography/feature/improved-logging
10b710498 remove comments
3e8a29d6f Merge branch 'main' into feature/improved-logging
ca8a88d9e first draft of updated docs
f0bb63294 remove one unnecessary log statement during BpmnWorkflow init and add some debug info about the other two
acce225e0 do not log task & workflow changes when deserializing

git-subtree-dir: SpiffWorkflow
git-subtree-split: 9d597627b46d236c684ca5d62cae16cfed6e1dec
This commit is contained in:
Jon Herron 2022-10-13 20:45:00 -04:00
parent 0892db6fa7
commit c04049105f
10 changed files with 398 additions and 277 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

396
doc/bpmn/advanced.rst
View File

@ -1,11 +1,8 @@
A More In-Depth Look at Some of SpiffWorkflow's Features A More In-Depth Look at Some of SpiffWorkflow's Features
======================================================== ========================================================
Displaying Workflow State
-------------------------
Filtering Tasks Filtering Tasks
^^^^^^^^^^^^^^^ ---------------
In our earlier example, all we did was check the lane a task was in and display In our earlier example, all we did was check the lane a task was in and display
it along with the task name and state. it along with the task name and state.
@ -36,203 +33,49 @@ correspond to which states).
.. code:: python .. code:: python
from SpiffWorkflow.task import Task from SpiffWorkflow.task import TaskState
To get a list of completed tasks To get a list of completed tasks
.. code:: python .. code:: python
tasks = workflow.get_tasks(Task.COMPLETED) tasks = workflow.get_tasks(TaskState.COMPLETED)
The tasks themselves are not particularly intuitive to work with. So SpiffWorkflow The tasks themselves are not particularly intuitive to work with. So SpiffWorkflow
provides some facilities for obtaining a more user-friendly version of upcoming tasks. provides some facilities for obtaining a more user-friendly version of upcoming tasks.
Nav(igation) List Logging
^^^^^^^^^^^^^^^^^ -------
In order to get the navigation list, we can call the workflow.get_nav_list() function. This Spiff provides several loggers:
will return a list of dictionaries with information about each task and decision point in the - the :code:`spiff` logger, which emits messages when a workflow is initialized and when tasks change state
workflow. Each item in this list returns some information about the tasks that are in the workflow, - the :code:`spiff.metrics` logger, which emits messages containing the elapsed duration of tasks
and how it relates to the other tasks. - the :code:`spiff.data` logger, which emits message when task or workflow data is updated.
To give you an idea of what is in the list I'll include a segment from the documentation:: Log level :code:`INFO` will provide reasonably detailed information about state changes.
id - TaskSpec or Sequence flow id As usual, log level :code:`DEBUG` will probably provide more logs than you really want
task_id - The uuid of the actual task instance, if it exists. to see, but the logs will contain the task and task internal data.
name - The name of the task spec (or sequence)
description - Text description
backtracks - Boolean, if this backtracks back up the list or not
level - Depth in the tree - probably not needed
indent - A hint for indentation
child_count - The number of children that should be associated with
this item.
lane - This is the swimlane for the task if indicated.
state - Text based state (may be half baked in the case that we have
more than one state for a task spec - but I don't think those
are being reported in the list, so it may not matter)
Any task with a blank or None as the description are excluded from the list (i.e. gateways)
Data can be included at any level less than :code:`INFO`. In our exmple application,
Because the output from this list may be used in a variety of contexts, the implementation is left to the user. we define a custom log level
MultiInstance Notes
-------------------
**loopCardinality** - This variable can be a text representation of a
number - for example '2' or it can be the name of a variable in
task.data that resolves to a text representation of a number.
It can also be a collection such as a list or a dictionary. In the
case that it is a list, the loop cardinality is equal to the length of
the list and in the case of a dictionary, it is equal to the list of
the keys of the dictionary.
If loopCardinality is left blank and the Collection is defined, or if
loopCardinality and Collection are the same collection, then the
MultiInstance will loop over the collection and update each element of
that collection with the new information. In this case, it is assumed
that the incoming collection is a dictionary, currently behavior for
working with a list in this manner is not defined and will raise an error.
**Collection** This is the name of the collection that is created from
the data generated when the task is run. Examples of this would be
form data that is generated from a UserTask or data that is generated
from a script that is run. Currently the collection is built up to be
a dictionary with a numeric key that corresponds to the place in the
loopCardinality. For example, if we set the loopCardinality to be a
list such as ['a','b','c] the resulting collection would be {1:'result
from a',2:'result from b',3:'result from c'} - and this would be true
even if it is a parallel MultiInstance where it was filled out in a
different order.
**Element Variable** This is the variable name for the current
iteration of the MultiInstance. In the case of the loopCardinality
being just a number, this would be 1,2,3, . . . If the
loopCardinality variable is mapped to a collection it would be either
the list value from that position, or it would be the value from the
dictionary where the keys are in sorted order. It is the content of the
element variable that should be updated in the task.data. This content
will then be added to the collection each time the task is completed.
Example:
In a sequential MultiInstance, loop cardinality is ['a','b','c'] and elementVariable is 'myvar'
then in the case of a sequential multiinstance the first call would
have 'myvar':'a' in the first run of the task and 'myvar':'b' in the
second.
Example:
In a Parallel MultiInstance, Loop cardinality is a variable that contains
{'a':'A','b':'B','c':'C'} and elementVariable is 'myvar' - when the multiinstance is ready, there
will be 3 tasks. If we choose the second task, the task.data will
contain 'myvar':'B'.
Custom Script Engines
---------------------
You may need to modify the default script engine, whether because you need to make additional
functionality available to it, or because you might want to restrict its capabilities for
security reasons.
.. warning::
The default script engine does little to no sanitization and uses :code:`eval`
and :code:`exec`! If you have security concerns, you should definitely investigate
replacing the default with your own implementation.
The default script engine imports the following objects:
- :code:`timedelta`
- :code:`datetime`
- :code:`dateparser`
- :code:`pytz`
You could add other functions or classes from the standard python modules or any code you've
implemented yourself.
In our example models so far, we've been using DMN tables to obtain product information. DMN
tables have a **lot** of uses so we wanted to feature them prominently, but in a simple way.
If a customer was selecting a product, we would surely have information about how the product
could be customized in a database somewhere. We would not hard code product information in
our diagram (although it is much easier to modify the BPMN diagram than to change the code
itself!). Our shipping costs would not be static, but would depend on the size of the order and
where it was being shipped -- maybe we'd query an API provided by our shipper.
SpiffWorkflow is obviously **not** going to know how to make a call to **your** database or
make API calls to **your** vendors. However, you can implement the calls yourself and make them
available as a method that can be used within a script task.
We are not going to actually include a database or API and write code for connecting to and querying
it, but we can model our database with a simple dictionary lookup since we only have 7 products
and just return the same static info for shipping for the purposes of the tutorial.
.. code:: python .. code:: python
from collections import namedtuple logging.addLevelName(15, 'DATA_LOG')
from SpiffWorkflow.bpmn.PythonScriptEngine import PythonScriptEngine so that we can see the task data in the logs without fully enabling debugging.
ProductInfo = namedtuple('ProductInfo', ['color', 'size', 'style', 'price']) The workflow runners take an `-l` argument that can be used to specify the logging level used
when running the example workflows.
INVENTORY = {
'product_a': ProductInfo(False, False, False, 15.00),
'product_b': ProductInfo(False, False, False, 15.00),
'product_c': ProductInfo(True, False, False, 25.00),
'product_d': ProductInfo(True, True, False, 20.00),
'product_e': ProductInfo(True, True, True, 25.00),
'product_f': ProductInfo(True, True, True, 30.00),
'product_g': ProductInfo(False, False, True, 25.00),
}
def lookup_product_info(product_name):
return INVENTORY[product_name]
def lookup_shipping_cost(shipping_method):
return 25.00 if shipping_method == 'Overnight' else 5.00
additions = {
'lookup_product_info': lookup_product_info,
'lookup_shipping_cost': lookup_shipping_cost
}
CustomScriptEngine = PythonScriptEngine(scriptingAdditions=additions)
We pass the script engine we created to the workflow when we load it.
.. code:: python
return BpmnWorkflow(parser.get_spec(process), script_engine=CustomScriptEngine)
We can use the custom functions in script tasks like any normal function:
.. figure:: figures/custom_script_usage.png
:scale: 30%
:align: center
Workflow with lanes
And we can simplify our 'Call Activity' flows:
.. figure:: figures/call_activity_script_flow.png
:scale: 30%
:align: center
Workflow with lanes
To run this workflow:
.. code-block:: console
./run.py -p order_product -b bpmn/call_activity_script.bpmn bpmn/top_level_script.bpmn
We have also done some work using `Restricted Python <https://restrictedpython.readthedocs.io/en/latest/>`_
to provide more secure alternatives to standard python functions.
Serialization Serialization
------------- -------------
.. warning:: .. warning::
Serialization Changed in Version 1.1.7. Support for pre-1.1.7 serialization will be dropped in 1.2. Serialization Changed in Version 1.1.7.
Support for pre-1.1.7 serialization will be dropped in a future release.
The old serialization method still works but it is deprecated. The old serialization method still works but it is deprecated.
To migrate your system to the new version, see "Migrating between To migrate your system to the new version, see "Migrating between
serialization versions" below. serialization versions" below.
@ -242,37 +85,32 @@ setting. This may not always be the case, we may be executing the workflow in th
may have a user request a web page where we open a specific workflow that we may be in the middle of, do one step of may have a user request a web page where we open a specific workflow that we may be in the middle of, do one step of
that workflow and then the user may be back in a few minutes, or maybe a few hours depending on the application. that workflow and then the user may be back in a few minutes, or maybe a few hours depending on the application.
To accomplish this, we can import the serializer The :code:`BpmnWorkflowSerializer` class contains a serializer for a workflow containing only standard BPMN Tasks.
Since we are using custom task classes (the Camunda :code:`UserTask` and the DMN :code:`BusinessRuleTask`),
.. code:: python we'll need to supply serializers for those task specs as well.
from SpiffWorkflow.bpmn.serializer import BpmnWorkflowSerializer
This class contains a serializer for a workflow containing only standard BPMN Tasks. Since we are using custom task
classes (the Camunda :code:`UserTask` and the DMN :code:`BusinessRuleTask`), we'll need to import serializers for those task s
pecs as well.
.. code:: python
from SpiffWorkflow.camunda.serializer import UserTaskConverter
from SpiffWorkflow.dmn.serializer import BusinessRuleTaskConverter
Strictly speaking, these are not serializers per se: they actually convert the tasks into dictionaries of Strictly speaking, these are not serializers per se: they actually convert the tasks into dictionaries of
JSON-serializable objects. Conversion to JSON is done only as the last step and could easily be replaced with some JSON-serializable objects. Conversion to JSON is done only as the last step and could easily be replaced with some
other output format. other output format.
We'll need to configure a Workflow Spec Converter with our custom classes: We'll need to configure a Workflow Spec Converter with our custom classes, as well as an optional
custom data converter.
.. code:: python .. code:: python
wf_spec_converter = BpmnWorkflowSerializer.configure_workflow_spec_converter( def create_serializer(task_types, data_converter=None):
[ UserTaskConverter, BusinessRuleTaskConverter ])
We create a serializer that can handle our extended task specs: wf_spec_converter = BpmnWorkflowSerializer.configure_workflow_spec_converter(task_types)
return BpmnWorkflowSerializer(wf_spec_converter, data_converter)
We'll call this from our main script:
.. code:: python .. code:: python
serializer = BpmnWorkflowSerializer(wf_spec_converter) serializer = create_serializer([ UserTaskConverter, BusinessRuleTaskConverter ], custom_data_converter)
We first configure a workflow spec converter that uses our custom task converters, and then we create
a :code:`BpmnWorkflowSerializer` from our workflow spec and data converters.
We'll give the user the option of dumping the workflow at any time. We'll give the user the option of dumping the workflow at any time.
@ -300,15 +138,15 @@ two components:
- a data converter (which handles workflow and task data). - a data converter (which handles workflow and task data).
The default workflow spec converter likely to meet your needs, either on its own, or with the inclusion of The default workflow spec converter likely to meet your needs, either on its own, or with the inclusion of
:code:`UserTask` and :code:`BusinessRuleTask` in the :code:`camnuda` and :code:`dmn` subpackages of this :code:`UserTask` and :code:`BusinessRuleTask` in the :code:`camnuda` or :code:`spiff` and :code:`dmn` subpackages
library, and all you'll need to do is add them to the list of task converters, as we did above. of this library, and all you'll need to do is add them to the list of task converters, as we did above.
However, he default data converter is very simple, adding only JSON-serializable conversions of :code:`datetime` However, he default data converter is very simple, adding only JSON-serializable conversions of :code:`datetime`
and :code:`timedelta` objects (we make these available in our default script engine) and UUIDs. If your and :code:`timedelta` objects (we make these available in our default script engine) and UUIDs. If your
workflow or task data contains objects that are not JSON-serializable, you'll need to extend ours, or extend workflow or task data contains objects that are not JSON-serializable, you'll need to extend ours, or extend
its base class to create one of your own. its base class to create one of your own.
To do extend ours: To extend ours:
1. Subclass the base data converter 1. Subclass the base data converter
2. Register classes along with functions for converting them to and from dictionaries 2. Register classes along with functions for converting them to and from dictionaries
@ -421,3 +259,163 @@ new 1.1 format.
If you've overridden the serializer version, you may need to incorporate our serialization changes with If you've overridden the serializer version, you may need to incorporate our serialization changes with
your own. You can find our conversions in your own. You can find our conversions in
`version_migrations.py <https://github.com/sartography/SpiffWorkflow/blob/main/SpiffWorkflow/bpmn/serializer/version_migration.py>`_ `version_migrations.py <https://github.com/sartography/SpiffWorkflow/blob/main/SpiffWorkflow/bpmn/serializer/version_migration.py>`_
Custom Script Engines
---------------------
You may need to modify the default script engine, whether because you need to make additional
functionality available to it, or because you might want to restrict its capabilities for
security reasons.
.. warning::
The default script engine does little to no sanitization and uses :code:`eval`
and :code:`exec`! If you have security concerns, you should definitely investigate
replacing the default with your own implementation.
We'll cover a simple extension of custom script engine here. There is also an examples of
a similar engine based on `RestrictedPython <https://restrictedpython.readthedocs.io/en/latest/>`_
included alongside this example.
The default script engine imports the following objects:
- :code:`timedelta`
- :code:`datetime`
- :code:`dateparser`
- :code:`pytz`
You could add other functions or classes from the standard python modules or any code you've
implemented yourself. Your global environment can be passed in using the `default_globals`
argument when initializing the script engine. In our RestrictedPython example, we use their
`safe_globals` which prevents users from executing some potentially unsafe operations.
In our example models so far, we've been using DMN tables to obtain product information. DMN
tables have a **lot** of uses so we wanted to feature them prominently, but in a simple way.
If a customer was selecting a product, we would surely have information about how the product
could be customized in a database somewhere. We would not hard code product information in
our diagram (although it is much easier to modify the BPMN diagram than to change the code
itself!). Our shipping costs would not be static, but would depend on the size of the order and
where it was being shipped -- maybe we'd query an API provided by our shipper.
SpiffWorkflow is obviously **not** going to know how to make a call to **your** database or
make API calls to **your** vendors. However, you can implement the calls yourself and make them
available as a method that can be used within a script task.
We are not going to actually include a database or API and write code for connecting to and querying
it, but we can model our database with a simple dictionary lookup since we only have 7 products
and just return the same static info for shipping for the purposes of the tutorial.
.. code:: python
from collections import namedtuple
from SpiffWorkflow.bpmn.PythonScriptEngine import PythonScriptEngine
ProductInfo = namedtuple('ProductInfo', ['color', 'size', 'style', 'price'])
INVENTORY = {
'product_a': ProductInfo(False, False, False, 15.00),
'product_b': ProductInfo(False, False, False, 15.00),
'product_c': ProductInfo(True, False, False, 25.00),
'product_d': ProductInfo(True, True, False, 20.00),
'product_e': ProductInfo(True, True, True, 25.00),
'product_f': ProductInfo(True, True, True, 30.00),
'product_g': ProductInfo(False, False, True, 25.00),
}
def lookup_product_info(product_name):
return INVENTORY[product_name]
def lookup_shipping_cost(shipping_method):
return 25.00 if shipping_method == 'Overnight' else 5.00
additions = {
'lookup_product_info': lookup_product_info,
'lookup_shipping_cost': lookup_shipping_cost
}
CustomScriptEngine = PythonScriptEngine(scripting_additions=additions)
We pass the script engine we created to the workflow when we load it.
.. code:: python
return BpmnWorkflow(parser.get_spec(process), script_engine=CustomScriptEngine)
We can use the custom functions in script tasks like any normal function:
.. figure:: figures/custom_script_usage.png
:scale: 30%
:align: center
Workflow with lanes
And we can simplify our 'Call Activity' flows:
.. figure:: figures/call_activity_script_flow.png
:scale: 30%
:align: center
Workflow with lanes
To run this workflow:
.. code-block:: console
./run.py -p order_product -b bpmn/call_activity_script.bpmn bpmn/top_level_script.bpmn
It is also possible to completely replace `exec` and `eval` with something else, or to
execute or evaluate statements in a completely separate environment by subclassing the
:code:`PythonScriptEngine` and overriding `_execute` and `_evaluate`. We have examples of
executing code inside a docker container or in a celery task i this repo.
MultiInstance Notes
-------------------
**loopCardinality** - This variable can be a text representation of a
number - for example '2' or it can be the name of a variable in
task.data that resolves to a text representation of a number.
It can also be a collection such as a list or a dictionary. In the
case that it is a list, the loop cardinality is equal to the length of
the list and in the case of a dictionary, it is equal to the list of
the keys of the dictionary.
If loopCardinality is left blank and the Collection is defined, or if
loopCardinality and Collection are the same collection, then the
MultiInstance will loop over the collection and update each element of
that collection with the new information. In this case, it is assumed
that the incoming collection is a dictionary, currently behavior for
working with a list in this manner is not defined and will raise an error.
**Collection** This is the name of the collection that is created from
the data generated when the task is run. Examples of this would be
form data that is generated from a UserTask or data that is generated
from a script that is run. Currently the collection is built up to be
a dictionary with a numeric key that corresponds to the place in the
loopCardinality. For example, if we set the loopCardinality to be a
list such as ['a','b','c] the resulting collection would be {1:'result
from a',2:'result from b',3:'result from c'} - and this would be true
even if it is a parallel MultiInstance where it was filled out in a
different order.
**Element Variable** This is the variable name for the current
iteration of the MultiInstance. In the case of the loopCardinality
being just a number, this would be 1,2,3, . . . If the
loopCardinality variable is mapped to a collection it would be either
the list value from that position, or it would be the value from the
dictionary where the keys are in sorted order. It is the content of the
element variable that should be updated in the task.data. This content
will then be added to the collection each time the task is completed.
Example:
In a sequential MultiInstance, loop cardinality is ['a','b','c'] and elementVariable is 'myvar'
then in the case of a sequential multiinstance the first call would
have 'myvar':'a' in the first run of the task and 'myvar':'b' in the
second.
Example:
In a Parallel MultiInstance, Loop cardinality is a variable that contains
{'a':'A','b':'B','c':'C'} and elementVariable is 'myvar' - when the multiinstance is ready, there
will be 3 tasks. If we choose the second task, the task.data will
contain 'myvar':'B'.

23
doc/bpmn/events.rst
View File

@ -209,27 +209,20 @@ reach the event.
Message Events Message Events
^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^
.. sidebar:: QA Lane In BPMN, Messages are used to communicate across processes. Technically, Messages are not
intended to be used inside a single process, but Spiff does support this use.
Ideally, this lane would be a process independent from the ordering process (we don't want
it to be cancelled just because an order eventually completes). However, limitations of how
SpiffWorkflow handles processes precludes multiple top-level processes.
In BPMN, Messages are used to communicate across processes and cannot be used within a
workflow, but SpiffWorkflow allows message communication between lanes as well as between
parent and child workflows. We'll use the first scenario in our example.
We've added a QA lane to out ordering process, whose job is investigating order order delays
and recommending improvements. This portion of our process will only be started when an
appropriate message is received.
Messages are similar to signals, in that they are referenced by name, but they have the Messages are similar to signals, in that they are referenced by name, but they have the
additional property that they may contain a payload. additional property that they may contain a payload.
We've added a QA process to our model, which will be initiated whenever an order takes to long
to fulfill. We'll send the reason for the delay in the message.
.. note:: .. note::
We currently depend on some Camunda-specific features in our implementation, but we This example depends on some Camunda-specific features in our implementation; there is
intend to replace this with our own. an alternate messaging implementation in the Spiff extensions package, described in
:doc:`spiff-extensions`.
.. figure:: figures/throw_message_event.png .. figure:: figures/throw_message_event.png
:scale: 30% :scale: 30%

BIN
doc/bpmn/figures/correlation.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 29 KiB

BIN
doc/bpmn/figures/data_object_configuration.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 17 KiB

BIN
doc/bpmn/figures/data_objects.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 242 KiB

BIN
doc/bpmn/figures/spiff_message_catch.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 28 KiB

BIN
doc/bpmn/figures/spiff_message_throw.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 38 KiB

1
doc/bpmn/index.rst
View File

@ -59,6 +59,7 @@ Supported BPMN Elements
organization organization
events events
multiinstance multiinstance
spiff-extensions
Putting it All Together Putting it All Together
----------------------- -----------------------

112
doc/bpmn/spiff-extensions.rst Normal file
View File

@ -0,0 +1,112 @@
Spiff Extensions
================
BPMN Model
----------
We'll be using the following files from `spiff-example-cli <https://github.com/sartography/spiff-example-cli>`_.
- `bpmn-spiff/events <https://github.com/sartography/spiff-example-cli/blob/master/bpmn-spiff/events.bpmn>`_ workflow
- `bpmn-spiff/call activity <https://github.com/sartography/spiff-example-cli/blob/master/bpmn-spiff/call_activity.bpmn>`_ workflow
- `product_prices <https://github.com/sartography/spiff-example-cli/blob/master/bpmn/product_prices.dmn>`_ DMN table
- `shipping_costs <https://github.com/sartography/spiff-example-cli/blob/master/bpmn/shipping_costs.dmn>`_ DMN table
We'll also be using the `run-spiff.py <https://github.com/sartography/spiff-example-clie/blob/master/run-spiff.py>`_ script
instead of the `run.py <https://github.com/sartography/spiff-example-clie/blob/master/run.py>`_ script
Camunda's BPMN editor does not handle data objects in the expected way. You can create data object
references, but there is no way to re-use data objects.
It also does not support Message Correlations, and the inteface for generating a message payload doesn't work
well in a Python environment.
We have extended BPMN.js to correct some of these issues. The examples in this section were created using our
custom BPMN editor, `bpmn-js-spiffworkflow <https://github.com/sartography/bpmn-js-spiffworkflow>`_.
Data Objects
^^^^^^^^^^^^
Data objects exist at a process level and are not visible in the diagram, but when you create a data object
reference, you can choose what data object it points to.
.. figure:: figures/data_object_configuration.png
:scale: 50%
:align: center
Configuring a data object reference
When a data output association (a line) is drawn from a task to a data object reference, the value is copied
from the task data to the workflow data and removed from the task. If a data input association is created from
a data object reference, the value is temporarily copied into the task data while the task is being executed,
and immediate removed afterwards.
This allows sensitive data to be removed from individual tasks (in our example, the customer's credit card
number). It can also be used to prevent large objects from being repeatedly copied from task to task.
Multiple data object references can point to the same underlying data. In our example, we use to references
to the same data object to pass the credit card info to both tasks that require it. On the right panel, we can
see that only one data object exists in the process.
.. figure:: figures/data_objects.png
:scale: 30%
:align: center
Data objects in a process
If you step through this workflow, you'll see that the card number is not contained in the task data after
the 'Enter Payment Info' has been completed.
Configuring Messages
^^^^^^^^^^^^^^^^^^^^
Messages are handled slightly differently in Spiff Message Events. On an Message Throw Event or Send Task,
we define a payload, which is simply a bit of python code that will be evaluated against the task data and
sent along with the message. In the corresponding Message Catch Event or Receive Task, we define a
variable name where we'll store the result.
Spiff Messages can also optionally use correlation keys. The correlation key is an expression or set of
expressions that are evaluated against a message payload to create an additional identifier for associating
messages with processes.
In our example, it is possible that multiple QA processes could be started (the timer event will fire every
minute until the order fulfillment process is complete). In this case, the message name is insufficient, as
there will be multiple processes that can accept messages based on the name.
.. figure:: figures/correlation.png
:scale: 50%
:align: center
Defining a correlation key
We use the timestamp of the message creation as a unique key that can be used to distinguish between multiple
QA processes.
.. figure:: figures/spiff_message_throw.png
:scale: 50%
:align: center
Configuring a message throw event
When we receive the event, we assign the payload to :code:`order_info`.
.. figure:: figures/spiff_message_catch.png
:scale: 50%
:align: center
Configuring a message catch event
The correlation is visible on both the Throw and Catch Events, but it is associated with the message rather
than the tasks themselves; if you update the expression on either event, the changes will appear in both places.
Running The Model
^^^^^^^^^^^^^^^^^
If you have set up our example repository, this model can be run with the
following command:
.. code-block:: console
./run-spiff.py -p order_product \
-d bpmn/product_prices.dmn bpmn/shipping_costs.dmn \
-b bpmn-spiffevents.bpmn bpmn-spiff/call_activity.bpmn

143
doc/bpmn/synthesis.rst
View File

@ -4,83 +4,112 @@ Putting it All Together
In this section we'll be discussing the overall structure of the workflow In this section we'll be discussing the overall structure of the workflow
runner we developed in `spiff-example-cli <https://github.com/sartography/spiff-example-cli>`_. runner we developed in `spiff-example-cli <https://github.com/sartography/spiff-example-cli>`_.
Our example application contains two different workflow runners, one that uses tasks with
Camunda extensions
(`run.py <https://github.com/sartography/spiff-example-cli/blob/main/run.py>`_) and one
that uses tasks with Spiff extensions
(`run-spiff.py <https://github.com/sartography/spiff-example-cli/blob/main/run.py>`_).
Most of the workflow operations will not change, so shared functions are defined in
`utils.py <https://github.com/sartography/spiff-example-cli/blob/main/utils.py>`_.
The primary difference is handling user tasks. Spiff User Tasks define an extensions
property that stores a filename containing a JSON schema used to define a web form. We
use `react-jsonschema-form <https://react-jsonschema-form.readthedocs.io/en/latest/>`_
to define our forms. This doesn't necessarily make a lot of sense in terms of a command
line UI, so we'll focus on the Camunda workflow runner in this document.
Loading a Workflow Loading a Workflow
------------------- -------------------
We'll need the following imports: The :code:`CamundaParser` extends the base :code:`BpmnParser`, adding functionality for
parsing forms defined in Camunda User Tasks and decision tables defined in Camunda
Business Rule Tasks. (There is a similar :code:`SpiffBpmnParser` used by the alternate
runner.)
We create the parser and use it to load our workflow.
.. code:: python .. code:: python
from SpiffWorkflow.bpmn.workflow import BpmnWorkflow parser = CamundaParser()
from SpiffWorkflow.camunda.parser.CamundaParser import CamundaParser wf = parse_workflow(parser, args.process, args.bpmn, args.dmn)
from SpiffWorkflow.dmn.parser.BpmnDmnParser import BpmnDmnParser
from custom_script_engine import CustomScriptEngine Our workflow parser looks like this;
We need to create a parser. We could have imported :code:`BpmnParser`, which
these parsers inherit from, but we need some additional features that the base
parser does not provide.
.. code:: python .. code:: python
class Parser(BpmnDmnParser): def parse_workflow(parser, process, bpmn_files, dmn_files, load_all=True):
OVERRIDE_PARSER_CLASSES = BpmnDmnParser.OVERRIDE_PARSER_CLASSES
OVERRIDE_PARSER_CLASSES.update(CamundaParser.OVERRIDE_PARSER_CLASSES)
We'll use :code:`BpmnDmnParser` as our base class, because we would like the ability
to use DMN tables in our application. The :code:`BpmnDmnParser` provides a task
parser for Business Rule Tasks, which the underlying :code:`BpmnParser` it inherits from
does not contain.
We also imported the :code:`CamundaParser` so that we can parse some Camunda
specific features we'll use (forms in User Tasks). The :code:`CamundaParser` User
Task parser will override the default parser.
In general, any task parser can be replaced with a custom parser of your
own design if you have a BPMN modeller that produces XML not handled by the
BPMN parsers in SpiffWorkflow.
.. code:: python
def parse(process, bpmn_files, dmn_files):
parser = Parser()
parser.add_bpmn_files(bpmn_files) parser.add_bpmn_files(bpmn_files)
if dmn_files: if dmn_files:
parser.add_dmn_files(dmn_files) parser.add_dmn_files(dmn_files)
top_level = parser.get_spec(process) top_level = parser.get_spec(process)
subprocesses = parser.get_process_specs() if load_all:
subprocesses = parser.find_all_specs()
else:
subprocesses = parser.get_subprocess_specs(process)
return BpmnWorkflow(top_level, subprocesses, script_engine=CustomScriptEngine) return BpmnWorkflow(top_level, subprocesses, script_engine=CustomScriptEngine)
We create an instance of our previously defined parser, add the BPMN files to it, and
optionally add any DMN files, if they were supplied.
We'll obtain the workflow specification from the parser for the top level process We'll obtain the workflow specification from the parser for the top level process
using :code:`parser.get_spec()`. using :code:`parser.get_spec()`.
We'll get the specs of all the processes that were parsed with :code:`parser.get_process_specs()` We have two options for finding subprocess specs. The method :code:`parser.find_all_specs()`
and provide these to the workflow as well. If your entire workflow is contained in your will create specs for all executable processes found in every file supplied. The method
top-level process, you can omit this argument, but if your workflow contains call activities, :code:`parser.get_subprocess_specs(process)` will create specs only for processes used by
you'll need to include it. the specified process. Both search recursively for subprocesses; the only difference is
the latter method limits the search start to the specified process.
Our examples are pretty simple and we're not loading any extraneous stuff, so we'll
just always load everything. If your entire workflow is contained in your top-level
process, you can omit the :code:`subprocess` argument, but if your workflow contains
call activities, you'll need to use one of these methods to find the models for any
called processes.
We also provide an enhanced script engine to our workflow. More information about how and We also provide an enhanced script engine to our workflow. More information about how and
why you might want to do this is covered in :doc:`advanced`. The :code:`script_engine` why you might want to do this is covered in :doc:`advanced`. The :code:`script_engine`
argument is optional and the default will be used if none is supplied. argument is optional and the default will be used if none is supplied.
We return a :code:`BpmnWorkflow` based on the specs that uses the our custom script engine We return :code:`BpmnWorkflow` that runs our top-level workflow and contains specs for any
to execute script tasks and evaluate expressions. subprocesses defined by that workflow.
Defining Task Handlers
----------------------
In :code:`run.py`, we define the function :code:`complete_user_task`. This has code specific
to Camunda User Task specs (in :code:`run-spiff.py`, we do something different).
We also import the shared function :code:`complete_manual_task` for handling Manual
Tasks as there is no difference.
We create a mapping of task type to handler, which we'll pass to our workflow runner.
.. code:: python
handlers = {
ManualTask: complete_manual_task,
UserTask: complete_user_task,
}
This might not be a step you would need to do in an application you build, since
you would likely have only one set of task specs that need to be parsed, handled, and
serialized; however our `run` method is an awful lot of code to maintain in two separate
files.
Running a Workflow Running a Workflow
------------------ ------------------
This is our application's :code:`run()` method. This is our application's :code:`run` method.
We pass our workflow, the task handlers, a serializer (creating a serializer is covered in
more depth in :doc:`advanced`).
The :code:`step` argument is a boolean that indicates whether we want the option of seeing The :code:`step` argument is a boolean that indicates whether we want the option of seeing
a more detailed representation of the state at each step, which we'll discuss in the a more detailed representation of the state at each step, which we'll discuss in the
section following this one. section following this one. The :code:`display_types` argument controls what types of
tasks should be included in a detailed list when stepping through a process.
.. code:: python .. code:: python
def run(workflow, step): def run(workflow, task_handlers, serializer, step, display_types):
workflow.do_engine_steps() workflow.do_engine_steps()
@ -105,19 +134,15 @@ section following this one.
dump.write(state) dump.write(state)
elif selected != '': elif selected != '':
next_task = options[selected] next_task = options[selected]
if isinstance(next_task.task_spec, UserTask): handler = task_handlers.get(type(next_task.task_spec))
complete_user_task(next_task) if handler is not None:
next_task.complete() handler(next_task)
elif isinstance(next_task.task_spec, ManualTask): next_task.complete()
complete_manual_task(next_task)
next_task.complete()
else:
next_task.complete()
workflow.refresh_waiting_tasks() workflow.refresh_waiting_tasks()
workflow.do_engine_steps() workflow.do_engine_steps()
if step: if step:
print_state(workflow) print_state(workflow, next_task, display_types)
print('\nWorkflow Data') print('\nWorkflow Data')
print(json.dumps(workflow.data, indent=2, separators=[ ', ', ': ' ])) print(json.dumps(workflow.data, indent=2, separators=[ ', ', ': ' ]))
@ -186,16 +211,14 @@ Here is the code we use for examining the workflow state.
.. code:: python .. code:: python
def print_state(workflow): def print_state(workflow, task, display_types):
task = workflow.last_task
print('\nLast Task') print('\nLast Task')
print(format_task(task)) print(format_task(task))
print(json.dumps(task.data, indent=2, separators=[ ', ', ': ' ])) print(json.dumps(task.data, indent=2, separators=[ ', ', ': ' ]))
display_types = (UserTask, ManualTask, ScriptTask, ThrowingEvent, CatchingEvent)
all_tasks = [ task for task in workflow.get_tasks() if isinstance(task.task_spec, display_types) ] all_tasks = [ task for task in workflow.get_tasks() if isinstance(task.task_spec, display_types) ]
upcoming_tasks = [ task for task in all_tasks if task.state in [Task.READY, Task.WAITING] ] upcoming_tasks = [ task for task in all_tasks if task.state in [TaskState.READY, TaskState.WAITING] ]
print('\nUpcoming Tasks') print('\nUpcoming Tasks')
for idx, task in enumerate(upcoming_tasks): for idx, task in enumerate(upcoming_tasks):
@ -205,8 +228,7 @@ Here is the code we use for examining the workflow state.
for idx, task in enumerate(all_tasks): for idx, task in enumerate(all_tasks):
print(format_task(task)) print(format_task(task))
We can find out what the last task was with :code:`workflow.last_task`. We'll print We'll print information about our task as described above, as well as a dump of its data.
its information as described above, as well as a dump of its data.
We can get a list of all tasks regardless of type or state with :code:`workflow.get_tasks()`. We can get a list of all tasks regardless of type or state with :code:`workflow.get_tasks()`.
@ -216,8 +238,3 @@ the tasks to only display the ones that would have salience to a user here.
We'll further filter those tasks for :code:`READY` and :code:`WAITING` tasks for a more We'll further filter those tasks for :code:`READY` and :code:`WAITING` tasks for a more
compact display, and only show all tasks when explicitly called for. compact display, and only show all tasks when explicitly called for.
This is a very simple application, so our interactions with tasks are very basic. You will
definitely want to see the 'Navigation List' section of :doc:`advanced` for more sophisticated
ways of managing workflow state.