486 lines
22 KiB
Python
486 lines
22 KiB
Python
import re
|
|
|
|
from SpiffWorkflow.serializer.exceptions import MissingSpecError
|
|
from lxml import etree
|
|
import shlex
|
|
from datetime import datetime
|
|
from typing import List
|
|
|
|
from SpiffWorkflow import Task as SpiffTask, WorkflowException
|
|
from SpiffWorkflow.bpmn.BpmnScriptEngine import BpmnScriptEngine
|
|
from SpiffWorkflow.bpmn.parser.ValidationException import ValidationException
|
|
from SpiffWorkflow.bpmn.serializer.BpmnSerializer import BpmnSerializer
|
|
from SpiffWorkflow.bpmn.specs.EndEvent import EndEvent
|
|
from SpiffWorkflow.bpmn.workflow import BpmnWorkflow
|
|
from SpiffWorkflow.camunda.parser.CamundaParser import CamundaParser
|
|
from SpiffWorkflow.dmn.parser.BpmnDmnParser import BpmnDmnParser
|
|
from SpiffWorkflow.exceptions import WorkflowTaskExecException
|
|
from SpiffWorkflow.specs import WorkflowSpec
|
|
|
|
import crc
|
|
from crc import session, app
|
|
from crc.api.common import ApiError
|
|
from crc.models.file import FileDataModel, FileModel, FileType
|
|
from crc.models.task_event import TaskEventModel
|
|
from crc.models.user import UserModelSchema
|
|
from crc.models.workflow import WorkflowStatus, WorkflowModel, WorkflowSpecDependencyFile
|
|
from crc.scripts.script import Script
|
|
from crc.services.file_service import FileService
|
|
from crc import app
|
|
from crc.services.user_service import UserService
|
|
|
|
|
|
class CustomBpmnScriptEngine(BpmnScriptEngine):
|
|
"""This is a custom script processor that can be easily injected into Spiff Workflow.
|
|
It will execute python code read in from the bpmn. It will also make any scripts in the
|
|
scripts directory available for execution. """
|
|
|
|
def evaluate(self, task, expression):
|
|
"""
|
|
Evaluate the given expression, within the context of the given task and
|
|
return the result.
|
|
"""
|
|
return self.evaluate_expression(task, expression)
|
|
|
|
|
|
def execute(self, task: SpiffTask, script, data):
|
|
|
|
study_id = task.workflow.data[WorkflowProcessor.STUDY_ID_KEY]
|
|
if WorkflowProcessor.WORKFLOW_ID_KEY in task.workflow.data:
|
|
workflow_id = task.workflow.data[WorkflowProcessor.WORKFLOW_ID_KEY]
|
|
else:
|
|
workflow_id = None
|
|
|
|
try:
|
|
if task.workflow.data[WorkflowProcessor.VALIDATION_PROCESS_KEY]:
|
|
augmentMethods = Script.generate_augmented_validate_list(task, study_id, workflow_id)
|
|
else:
|
|
augmentMethods = Script.generate_augmented_list(task, study_id, workflow_id)
|
|
|
|
super().execute(task, script, data, externalMethods=augmentMethods)
|
|
except SyntaxError as e:
|
|
raise ApiError('syntax_error',
|
|
f'Something is wrong with your python script '
|
|
f'please correct the following:'
|
|
f' {script}, {e.msg}')
|
|
except NameError as e:
|
|
raise ApiError('name_error',
|
|
f'something you are referencing does not exist:'
|
|
f' {script}, {e}')
|
|
|
|
# else:
|
|
# self.run_predefined_script(task, script[2:], data) # strip off the first two characters.
|
|
|
|
# def run_predefined_script(self, task: SpiffTask, script, data):
|
|
# commands = shlex.split(script)
|
|
# path_and_command = commands[0].rsplit(".", 1)
|
|
# if len(path_and_command) == 1:
|
|
# module_name = "crc.scripts." + self.camel_to_snake(path_and_command[0])
|
|
# class_name = path_and_command[0]
|
|
# else:
|
|
# module_name = "crc.scripts." + path_and_command[0] + "." + self.camel_to_snake(path_and_command[1])
|
|
# class_name = path_and_command[1]
|
|
# try:
|
|
# mod = __import__(module_name, fromlist=[class_name])
|
|
# klass = getattr(mod, class_name)
|
|
# study_id = task.workflow.data[WorkflowProcessor.STUDY_ID_KEY]
|
|
# if WorkflowProcessor.WORKFLOW_ID_KEY in task.workflow.data:
|
|
# workflow_id = task.workflow.data[WorkflowProcessor.WORKFLOW_ID_KEY]
|
|
# else:
|
|
# workflow_id = None
|
|
#
|
|
# if not isinstance(klass(), Script):
|
|
# raise ApiError.from_task("invalid_script",
|
|
# "This is an internal error. The script '%s:%s' you called " %
|
|
# (module_name, class_name) +
|
|
# "does not properly implement the CRC Script class.",
|
|
# task=task)
|
|
# if task.workflow.data[WorkflowProcessor.VALIDATION_PROCESS_KEY]:
|
|
# """If this is running a validation, and not a normal process, then we want to
|
|
# mimic running the script, but not make any external calls or database changes."""
|
|
# klass().do_task_validate_only(task, study_id, workflow_id, *commands[1:])
|
|
# else:
|
|
# klass().do_task(task, study_id, workflow_id, *commands[1:])
|
|
# except ModuleNotFoundError:
|
|
# raise ApiError.from_task("invalid_script",
|
|
# "Unable to locate Script: '%s:%s'" % (module_name, class_name),
|
|
# task=task)
|
|
|
|
def evaluate_expression(self, task, expression):
|
|
"""
|
|
Evaluate the given expression, within the context of the given task and
|
|
return the result.
|
|
"""
|
|
study_id = task.workflow.data[WorkflowProcessor.STUDY_ID_KEY]
|
|
if WorkflowProcessor.WORKFLOW_ID_KEY in task.workflow.data:
|
|
workflow_id = task.workflow.data[WorkflowProcessor.WORKFLOW_ID_KEY]
|
|
else:
|
|
workflow_id = None
|
|
|
|
try:
|
|
if task.workflow.data[WorkflowProcessor.VALIDATION_PROCESS_KEY]:
|
|
augmentMethods = Script.generate_augmented_validate_list(task, study_id, workflow_id)
|
|
else:
|
|
augmentMethods = Script.generate_augmented_list(task, study_id, workflow_id)
|
|
exp, valid = self.validateExpression(expression)
|
|
return self._eval(exp, externalMethods=augmentMethods, **task.data)
|
|
|
|
except Exception as e:
|
|
raise WorkflowTaskExecException(task,
|
|
"Error evaluating expression "
|
|
"'%s', %s" % (expression, str(e)))
|
|
|
|
|
|
@staticmethod
|
|
def camel_to_snake(camel):
|
|
camel = camel.strip()
|
|
return re.sub(r'(?<!^)(?=[A-Z])', '_', camel).lower()
|
|
|
|
|
|
class MyCustomParser(BpmnDmnParser):
|
|
"""
|
|
A BPMN and DMN parser that can also parse Camunda forms.
|
|
"""
|
|
OVERRIDE_PARSER_CLASSES = BpmnDmnParser.OVERRIDE_PARSER_CLASSES
|
|
OVERRIDE_PARSER_CLASSES.update(CamundaParser.OVERRIDE_PARSER_CLASSES)
|
|
|
|
|
|
class WorkflowProcessor(object):
|
|
_script_engine = CustomBpmnScriptEngine()
|
|
_serializer = BpmnSerializer()
|
|
|
|
WORKFLOW_ID_KEY = "workflow_id"
|
|
STUDY_ID_KEY = "study_id"
|
|
VALIDATION_PROCESS_KEY = "validate_only"
|
|
|
|
def __init__(self, workflow_model: WorkflowModel, validate_only=False):
|
|
"""Create a Workflow Processor based on the serialized information available in the workflow model."""
|
|
|
|
self.workflow_model = workflow_model
|
|
|
|
if workflow_model.bpmn_workflow_json is None: # The workflow was never started.
|
|
self.spec_data_files = FileService.get_spec_data_files(
|
|
workflow_spec_id=workflow_model.workflow_spec_id)
|
|
spec = self.get_spec(self.spec_data_files, workflow_model.workflow_spec_id)
|
|
else:
|
|
self.spec_data_files = FileService.get_spec_data_files(
|
|
workflow_spec_id=workflow_model.workflow_spec_id,
|
|
workflow_id=workflow_model.id)
|
|
spec = None
|
|
|
|
self.workflow_spec_id = workflow_model.workflow_spec_id
|
|
|
|
try:
|
|
self.bpmn_workflow = self.__get_bpmn_workflow(workflow_model, spec, validate_only)
|
|
self.bpmn_workflow.script_engine = self._script_engine
|
|
|
|
if UserService.has_user():
|
|
current_user = UserService.current_user(allow_admin_impersonate=True)
|
|
current_user_data = UserModelSchema().dump(current_user)
|
|
tasks = self.bpmn_workflow.get_tasks(SpiffTask.READY)
|
|
for task in tasks:
|
|
task.data['current_user'] = current_user_data
|
|
|
|
if self.WORKFLOW_ID_KEY not in self.bpmn_workflow.data:
|
|
if not workflow_model.id:
|
|
session.add(workflow_model)
|
|
session.commit()
|
|
# If the model is new, and has no id, save it, write it into the workflow model
|
|
# and save it again. In this way, the workflow process is always aware of the
|
|
# database model to which it is associated, and scripts running within the model
|
|
# can then load data as needed.
|
|
self.bpmn_workflow.data[WorkflowProcessor.WORKFLOW_ID_KEY] = workflow_model.id
|
|
workflow_model.bpmn_workflow_json = WorkflowProcessor._serializer.serialize_workflow(
|
|
self.bpmn_workflow, include_spec=True)
|
|
self.save()
|
|
|
|
except MissingSpecError as ke:
|
|
raise ApiError(code="unexpected_workflow_structure",
|
|
message="Failed to deserialize workflow"
|
|
" '%s' version %s, due to a mis-placed or missing task '%s'" %
|
|
(self.workflow_spec_id, self.get_version_string(), str(ke)))
|
|
|
|
# set whether this is the latest spec file.
|
|
if self.spec_data_files == FileService.get_spec_data_files(workflow_spec_id=workflow_model.workflow_spec_id):
|
|
self.is_latest_spec = True
|
|
else:
|
|
self.is_latest_spec = False
|
|
|
|
def reset(self, workflow_model, clear_data=False):
|
|
print('WorkflowProcessor: reset: ')
|
|
|
|
self.cancel_notify()
|
|
workflow_model.bpmn_workflow_json = None
|
|
if clear_data:
|
|
# Clear form_data from task_events
|
|
task_events = session.query(TaskEventModel). \
|
|
filter(TaskEventModel.workflow_id == workflow_model.id).all()
|
|
for task_event in task_events:
|
|
task_event.form_data = {}
|
|
session.add(task_event)
|
|
session.commit()
|
|
return self.__init__(workflow_model)
|
|
|
|
def __get_bpmn_workflow(self, workflow_model: WorkflowModel, spec: WorkflowSpec, validate_only=False):
|
|
if workflow_model.bpmn_workflow_json:
|
|
bpmn_workflow = self._serializer.deserialize_workflow(workflow_model.bpmn_workflow_json,
|
|
workflow_spec=spec)
|
|
else:
|
|
bpmn_workflow = BpmnWorkflow(spec, script_engine=self._script_engine)
|
|
bpmn_workflow.data[WorkflowProcessor.STUDY_ID_KEY] = workflow_model.study_id
|
|
bpmn_workflow.data[WorkflowProcessor.VALIDATION_PROCESS_KEY] = validate_only
|
|
# try:
|
|
# bpmn_workflow.do_engine_steps()
|
|
# except WorkflowException as we:
|
|
# raise ApiError.from_task_spec("error_loading_workflow", str(we), we.sender)
|
|
return bpmn_workflow
|
|
|
|
def save(self):
|
|
"""Saves the current state of this processor to the database """
|
|
self.workflow_model.bpmn_workflow_json = self.serialize()
|
|
complete_states = [SpiffTask.CANCELLED, SpiffTask.COMPLETED]
|
|
tasks = list(self.get_all_user_tasks())
|
|
self.workflow_model.status = self.get_status()
|
|
self.workflow_model.total_tasks = len(tasks)
|
|
self.workflow_model.completed_tasks = sum(1 for t in tasks if t.state in complete_states)
|
|
self.workflow_model.last_updated = datetime.now()
|
|
self.update_dependencies(self.spec_data_files)
|
|
session.add(self.workflow_model)
|
|
session.commit()
|
|
|
|
def get_version_string(self):
|
|
# this could potentially become expensive to load all the data in the data models.
|
|
# in which case we might consider using a deferred loader for the actual data, but
|
|
# trying not to pre-optimize.
|
|
file_data_models = FileService.get_spec_data_files(self.workflow_model.workflow_spec_id,
|
|
self.workflow_model.id)
|
|
return WorkflowProcessor.__get_version_string_for_data_models(file_data_models)
|
|
|
|
@staticmethod
|
|
def get_latest_version_string_for_spec(spec_id):
|
|
file_data_models = FileService.get_spec_data_files(spec_id)
|
|
return WorkflowProcessor.__get_version_string_for_data_models(file_data_models)
|
|
|
|
@staticmethod
|
|
def __get_version_string_for_data_models(file_data_models):
|
|
"""Version is in the format v[VERSION] (FILE_ID_LIST)
|
|
For example, a single bpmn file with only one version would be
|
|
v1 (12) Where 12 is the id of the file data model that is used to create the
|
|
specification. If multiple files exist, they are added on in
|
|
dot notation to both the version number and the file list. So
|
|
a Spec that includes a BPMN, DMN, an a Word file all on the first
|
|
version would be v1.1.1 (12.45.21)"""
|
|
|
|
major_version = 0 # The version of the primary file.
|
|
minor_version = [] # The versions of the minor files if any.
|
|
file_ids = []
|
|
for file_data in file_data_models:
|
|
file_ids.append(file_data.id)
|
|
if file_data.file_model.primary:
|
|
major_version = file_data.version
|
|
else:
|
|
minor_version.append(file_data.version)
|
|
minor_version.insert(0, major_version) # Add major version to beginning.
|
|
version = ".".join(str(x) for x in minor_version)
|
|
files = ".".join(str(x) for x in file_ids)
|
|
full_version = "v%s (%s)" % (version, files)
|
|
return full_version
|
|
|
|
def update_dependencies(self, spec_data_files):
|
|
existing_dependencies = FileService.get_spec_data_files(
|
|
workflow_spec_id=self.workflow_model.workflow_spec_id,
|
|
workflow_id=self.workflow_model.id)
|
|
|
|
# Don't save the dependencies if they haven't changed.
|
|
if existing_dependencies == spec_data_files:
|
|
return
|
|
|
|
# Remove all existing dependencies, and replace them.
|
|
self.workflow_model.dependencies = []
|
|
for file_data in spec_data_files:
|
|
self.workflow_model.dependencies.append(WorkflowSpecDependencyFile(file_data_id=file_data.id))
|
|
|
|
@staticmethod
|
|
def run_master_spec(spec_model, study):
|
|
"""Executes a BPMN specification for the given study, without recording any information to the database
|
|
Useful for running the master specification, which should not persist. """
|
|
spec_data_files = FileService.get_spec_data_files(spec_model.id)
|
|
spec = WorkflowProcessor.get_spec(spec_data_files, spec_model.id)
|
|
try:
|
|
bpmn_workflow = BpmnWorkflow(spec, script_engine=WorkflowProcessor._script_engine)
|
|
bpmn_workflow.data[WorkflowProcessor.STUDY_ID_KEY] = study.id
|
|
bpmn_workflow.data[WorkflowProcessor.VALIDATION_PROCESS_KEY] = False
|
|
bpmn_workflow.do_engine_steps()
|
|
except WorkflowException as we:
|
|
raise ApiError.from_task_spec("error_running_master_spec", str(we), we.sender)
|
|
|
|
if not bpmn_workflow.is_completed():
|
|
raise ApiError("master_spec_not_automatic",
|
|
"The master spec should only contain fully automated tasks, it failed to complete.")
|
|
|
|
return bpmn_workflow.last_task.data
|
|
|
|
@staticmethod
|
|
def get_parser():
|
|
parser = MyCustomParser()
|
|
return parser
|
|
|
|
@staticmethod
|
|
def get_spec(file_data_models: List[FileDataModel], workflow_spec_id):
|
|
"""Returns a SpiffWorkflow specification for the given workflow spec,
|
|
using the files provided. The Workflow_spec_id is only used to generate
|
|
better error messages."""
|
|
parser = WorkflowProcessor.get_parser()
|
|
process_id = None
|
|
|
|
for file_data in file_data_models:
|
|
if file_data.file_model.type == FileType.bpmn:
|
|
bpmn: etree.Element = etree.fromstring(file_data.data)
|
|
if file_data.file_model.primary:
|
|
process_id = FileService.get_process_id(bpmn)
|
|
parser.add_bpmn_xml(bpmn, filename=file_data.file_model.name)
|
|
elif file_data.file_model.type == FileType.dmn:
|
|
dmn: etree.Element = etree.fromstring(file_data.data)
|
|
parser.add_dmn_xml(dmn, filename=file_data.file_model.name)
|
|
if process_id is None:
|
|
raise (ApiError(code="no_primary_bpmn_error",
|
|
message="There is no primary BPMN model defined for workflow %s" % workflow_spec_id))
|
|
try:
|
|
spec = parser.get_spec(process_id)
|
|
except ValidationException as ve:
|
|
raise ApiError(code="workflow_validation_error",
|
|
message="Failed to parse Workflow Specification '%s'. \n" % workflow_spec_id +
|
|
"Error is %s. \n" % str(ve),
|
|
file_name=ve.filename,
|
|
task_id=ve.id,
|
|
tag=ve.tag)
|
|
return spec
|
|
|
|
@staticmethod
|
|
def status_of(bpmn_workflow):
|
|
if bpmn_workflow.is_completed():
|
|
return WorkflowStatus.complete
|
|
user_tasks = bpmn_workflow.get_ready_user_tasks()
|
|
if len(user_tasks) > 0:
|
|
return WorkflowStatus.user_input_required
|
|
else:
|
|
return WorkflowStatus.waiting
|
|
|
|
# def hard_reset(self):
|
|
# """Recreate this workflow. This will be useful when a workflow specification changes.
|
|
# """
|
|
# self.spec_data_files = FileService.get_spec_data_files(workflow_spec_id=self.workflow_spec_id)
|
|
# new_spec = WorkflowProcessor.get_spec(self.spec_data_files, self.workflow_spec_id)
|
|
# new_bpmn_workflow = BpmnWorkflow(new_spec, script_engine=self._script_engine)
|
|
# new_bpmn_workflow.data = self.bpmn_workflow.data
|
|
# try:
|
|
# new_bpmn_workflow.do_engine_steps()
|
|
# except WorkflowException as we:
|
|
# raise ApiError.from_task_spec("hard_reset_engine_steps_error", str(we), we.sender)
|
|
# self.bpmn_workflow = new_bpmn_workflow
|
|
|
|
def get_status(self):
|
|
return self.status_of(self.bpmn_workflow)
|
|
|
|
def do_engine_steps(self):
|
|
try:
|
|
self.bpmn_workflow.do_engine_steps()
|
|
except WorkflowTaskExecException as we:
|
|
raise ApiError.from_task("task_error", str(we), we.task)
|
|
|
|
def cancel_notify(self):
|
|
try:
|
|
self.bpmn_workflow.cancel_notify()
|
|
except WorkflowTaskExecException as we:
|
|
raise ApiError.from_task("task_error", str(we), we.task)
|
|
|
|
def serialize(self):
|
|
return self._serializer.serialize_workflow(self.bpmn_workflow,include_spec=True)
|
|
|
|
def next_user_tasks(self):
|
|
return self.bpmn_workflow.get_ready_user_tasks()
|
|
|
|
def next_task(self):
|
|
"""Returns the next task that should be completed
|
|
even if there are parallel tasks and multiple options are
|
|
available.
|
|
If the workflow is complete
|
|
it will return the final end task.
|
|
"""
|
|
|
|
# If the whole blessed mess is done, return the end_event task in the tree
|
|
if self.bpmn_workflow.is_completed():
|
|
for task in SpiffTask.Iterator(self.bpmn_workflow.task_tree, SpiffTask.ANY_MASK):
|
|
if isinstance(task.task_spec, EndEvent):
|
|
return task
|
|
|
|
# If there are ready tasks to complete, return the next ready task, but return the one
|
|
# in the active parallel path if possible.
|
|
ready_tasks = self.bpmn_workflow.get_tasks(SpiffTask.READY)
|
|
if len(ready_tasks) > 0:
|
|
for task in ready_tasks:
|
|
if task.parent == self.bpmn_workflow.last_task:
|
|
return task
|
|
return ready_tasks[0]
|
|
|
|
# If there are no ready tasks, but the thing isn't complete yet, find the first non-complete task
|
|
# and return that
|
|
next_task = None
|
|
for task in SpiffTask.Iterator(self.bpmn_workflow.task_tree, SpiffTask.NOT_FINISHED_MASK):
|
|
next_task = task
|
|
return next_task
|
|
|
|
def previous_task(self):
|
|
return None
|
|
|
|
def complete_task(self, task):
|
|
self.bpmn_workflow.complete_task_from_id(task.id)
|
|
|
|
def get_data(self):
|
|
return self.bpmn_workflow.data
|
|
|
|
def get_workflow_id(self):
|
|
return self.workflow_model.id
|
|
|
|
def get_study_id(self):
|
|
return self.bpmn_workflow.data[self.STUDY_ID_KEY]
|
|
|
|
def get_ready_user_tasks(self):
|
|
return self.bpmn_workflow.get_ready_user_tasks()
|
|
|
|
def get_current_user_tasks(self):
|
|
"""Return a list of all user tasks that are READY or
|
|
COMPLETE and are parallel to the READY Task."""
|
|
ready_tasks = self.bpmn_workflow.get_ready_user_tasks()
|
|
additional_tasks = []
|
|
if len(ready_tasks) > 0:
|
|
for child in ready_tasks[0].parent.children:
|
|
if child.state == SpiffTask.COMPLETED:
|
|
additional_tasks.append(child)
|
|
return ready_tasks + additional_tasks
|
|
|
|
def get_all_user_tasks(self):
|
|
all_tasks = self.bpmn_workflow.get_tasks(SpiffTask.ANY_MASK)
|
|
return [t for t in all_tasks if not self.bpmn_workflow._is_engine_task(t.task_spec)]
|
|
|
|
def get_all_completed_tasks(self):
|
|
all_tasks = self.bpmn_workflow.get_tasks(SpiffTask.ANY_MASK)
|
|
return [t for t in all_tasks
|
|
if not self.bpmn_workflow._is_engine_task(t.task_spec) and t.state in [t.COMPLETED, t.CANCELLED]]
|
|
|
|
def get_nav_item(self, task):
|
|
for nav_item in self.bpmn_workflow.get_nav_list():
|
|
if nav_item['task_id'] == task.id:
|
|
return nav_item
|
|
|
|
def find_spec_and_field(self, spec_name, field_id):
|
|
"""Tracks down a form field by name in the workflow spec,
|
|
only looks at ready tasks. Returns a tuple of the task, and form"""
|
|
for spec in self.bpmn_workflow.spec.task_specs.values():
|
|
if spec.name == spec_name and hasattr(spec, "form"):
|
|
for field in spec.form.fields:
|
|
if field.id == field_id:
|
|
return spec, field
|
|
raise ApiError("invalid_field",
|
|
"Unable to find a task in the workflow with a lookup field called: %s" % field_id)
|