161 lines
7.6 KiB
Python
161 lines
7.6 KiB
Python
from SpiffWorkflow.bpmn.specs.ManualTask import ManualTask
|
|
from SpiffWorkflow.bpmn.specs.MultiInstanceTask import MultiInstanceTask
|
|
from SpiffWorkflow.bpmn.specs.NoneTask import NoneTask
|
|
from SpiffWorkflow.bpmn.specs.ScriptTask import ScriptTask
|
|
from SpiffWorkflow.bpmn.specs.UserTask import UserTask
|
|
from SpiffWorkflow.bpmn.workflow import BpmnWorkflow
|
|
from SpiffWorkflow.dmn.specs.BuisnessRuleTask import BusinessRuleTask
|
|
from SpiffWorkflow.specs import CancelTask, StartTask
|
|
from pandas import ExcelFile
|
|
|
|
from crc.api.common import ApiError
|
|
from crc.models.api_models import Task, MultiInstanceType
|
|
import jinja2
|
|
from jinja2 import Template
|
|
|
|
from crc.services.file_service import FileService
|
|
from crc.services.workflow_processor import WorkflowProcessor, CustomBpmnScriptEngine
|
|
from SpiffWorkflow import Task as SpiffTask, WorkflowException
|
|
|
|
|
|
class WorkflowService(object):
|
|
"""Provides tools for processing workflows and tasks. This
|
|
should at some point, be the only way to work with Workflows, and
|
|
the workflow Processor should be hidden behind this service.
|
|
This will help maintain a structure that avoids circular dependencies.
|
|
But for now, this contains tools for converting spiff-workflow models into our
|
|
own API models with additional information and capabilities."""
|
|
|
|
@classmethod
|
|
def test_spec(cls, spec_id):
|
|
"""Runs a spec through it's paces to see if it results in any errors. Not full proof, but a good
|
|
sanity check."""
|
|
|
|
spec = WorkflowProcessor.get_spec(spec_id)
|
|
bpmn_workflow = BpmnWorkflow(spec, script_engine=CustomBpmnScriptEngine())
|
|
bpmn_workflow.data[WorkflowProcessor.STUDY_ID_KEY] = 1
|
|
bpmn_workflow.data[WorkflowProcessor.WORKFLOW_ID_KEY] = spec_id
|
|
bpmn_workflow.data[WorkflowProcessor.VALIDATION_PROCESS_KEY] = True
|
|
|
|
while not bpmn_workflow.is_completed():
|
|
try:
|
|
bpmn_workflow.do_engine_steps()
|
|
tasks = bpmn_workflow.get_tasks(SpiffTask.READY)
|
|
for task in tasks:
|
|
task_api = WorkflowService.spiff_task_to_api_task(
|
|
task) # Assure we try to process the documenation, and raise those errors.
|
|
WorkflowProcessor.populate_form_with_random_data(task)
|
|
task.complete()
|
|
except WorkflowException as we:
|
|
raise ApiError.from_task_spec("workflow_execution_exception", str(we),
|
|
we.sender)
|
|
|
|
@staticmethod
|
|
def spiff_task_to_api_task(spiff_task):
|
|
task_type = spiff_task.task_spec.__class__.__name__
|
|
|
|
if isinstance(spiff_task.task_spec, UserTask):
|
|
task_type = "UserTask"
|
|
elif isinstance(spiff_task.task_spec, ManualTask):
|
|
task_type = "ManualTask"
|
|
elif isinstance(spiff_task.task_spec, BusinessRuleTask):
|
|
task_type = "BusinessRuleTask"
|
|
elif isinstance(spiff_task.task_spec, CancelTask):
|
|
task_type = "CancelTask"
|
|
elif isinstance(spiff_task.task_spec, ScriptTask):
|
|
task_type = "ScriptTask"
|
|
elif isinstance(spiff_task.task_spec, StartTask):
|
|
task_type = "StartTask"
|
|
else:
|
|
task_type = "NoneTask"
|
|
|
|
info = spiff_task.task_info()
|
|
if info["is_looping"]:
|
|
mi_type = MultiInstanceType.looping
|
|
elif info["is_sequential_mi"]:
|
|
mi_type = MultiInstanceType.sequential
|
|
elif info["is_parallel_mi"]:
|
|
mi_type = MultiInstanceType.parallel
|
|
else:
|
|
mi_type = MultiInstanceType.none
|
|
|
|
task = Task(spiff_task.id,
|
|
spiff_task.task_spec.name,
|
|
spiff_task.task_spec.description,
|
|
task_type,
|
|
spiff_task.get_state_name(),
|
|
None,
|
|
"",
|
|
spiff_task.data,
|
|
mi_type,
|
|
info["mi_count"],
|
|
info["mi_index"])
|
|
|
|
# Only process the form and documentation if this is something that is ready or completed.
|
|
if not (spiff_task._is_predicted()):
|
|
if hasattr(spiff_task.task_spec, "form"):
|
|
task.form = spiff_task.task_spec.form
|
|
for field in task.form.fields:
|
|
WorkflowService._process_options(spiff_task, field)
|
|
|
|
task.documentation = WorkflowService._process_documentation(spiff_task)
|
|
return task
|
|
|
|
@staticmethod
|
|
def _process_documentation(spiff_task):
|
|
"""Runs the given documentation string through the Jinja2 processor to inject data
|
|
create loops, etc... - If a markdown file exists with the same name as the task id,
|
|
it will use that file instead of the documentation. """
|
|
|
|
documentation = spiff_task.task_spec.documentation if hasattr(spiff_task.task_spec, "documentation") else ""
|
|
|
|
try:
|
|
doc_file_name = spiff_task.task_spec.name + ".md"
|
|
data_model = FileService.get_workflow_file_data(spiff_task.workflow, doc_file_name)
|
|
raw_doc = data_model.data.decode("utf-8")
|
|
except ApiError:
|
|
raw_doc = documentation
|
|
|
|
if not raw_doc:
|
|
return ""
|
|
|
|
try:
|
|
template = Template(raw_doc)
|
|
return template.render(**spiff_task.data)
|
|
except jinja2.exceptions.TemplateError as ue:
|
|
raise ApiError(code="template_error", message="Error processing template for task %s: %s" %
|
|
(spiff_task.task_spec.name, str(ue)), status_code=500)
|
|
# TODO: Catch additional errors and report back.
|
|
|
|
@staticmethod
|
|
def _process_options(spiff_task, field):
|
|
""" Checks to see if the options are provided in a separate lookup table associated with the
|
|
workflow, and populates these if possible. """
|
|
if field.has_property(Task.ENUM_OPTIONS_FILE_PROP):
|
|
if not field.has_property(Task.EMUM_OPTIONS_VALUE_COL_PROP) or \
|
|
not field.has_property(Task.EMUM_OPTIONS_LABEL_COL_PROP):
|
|
raise ApiError.from_task("invalid_emum",
|
|
"For emumerations based on an xls file, you must include 3 properties: %s, "
|
|
"%s, and %s, you supplied %s" % (Task.ENUM_OPTIONS_FILE_PROP,
|
|
Task.EMUM_OPTIONS_VALUE_COL_PROP,
|
|
Task.EMUM_OPTIONS_LABEL_COL_PROP),
|
|
task=spiff_task)
|
|
|
|
# Get the file data from the File Service
|
|
file_name = field.get_property(Task.ENUM_OPTIONS_FILE_PROP)
|
|
value_column = field.get_property(Task.EMUM_OPTIONS_VALUE_COL_PROP)
|
|
label_column = field.get_property(Task.EMUM_OPTIONS_LABEL_COL_PROP)
|
|
data_model = FileService.get_workflow_file_data(spiff_task.workflow, file_name)
|
|
xls = ExcelFile(data_model.data)
|
|
df = xls.parse(xls.sheet_names[0])
|
|
if value_column not in df:
|
|
raise ApiError("invalid_emum",
|
|
"The file %s does not contain a column named % s" % (file_name, value_column))
|
|
if label_column not in df:
|
|
raise ApiError("invalid_emum",
|
|
"The file %s does not contain a column named % s" % (file_name, label_column))
|
|
|
|
for index, row in df.iterrows():
|
|
field.options.append({"id": row[value_column],
|
|
"name": row[label_column]})
|