govoplan-workflow-engine (0.1.17)
Installation
pip install --index-url https://git.add-ideas.de/api/packages/GovOPlaN/pypi/simple/ --extra-index-url https://pypi.org/simple govoplan-workflow-engineAbout this package
Headless, versioned workflow definition and execution engine for GovOPlaN.
GovOPlaN Workflow Engine
Repository type: module (platform).
Headless workflow definition and execution services for GovOPlaN modules.
Workflow Engine owns the canonical BPMN/native graph model, immutable
definition revisions, runtime instances and evidence, migrations, validation,
execution adapters, dispatch, retries, and recovery. It intentionally provides
no primary navigation or visual editor. Modules can depend on runtime module ID
workflow_engine and the versioned workflow.* interfaces without installing
the optional govoplan-workflow editor.
The extraction preserves the existing workflow_* tables, Alembic revision
IDs, /api/v1/workflow routes, workflow:* permissions, instance IDs, and
audit references. See the engine/editor split
for the ownership and compatibility policy.
See the module concept and
BPMN interoperability contract for runtime
semantics and adapter boundaries.
See durable runtime recovery for action,
worker, timer, and unknown-provider-outcome handling.
The optional workflow_engine.service_launcher capability lets Portal start an
authorized active workflow from an exact published Service revision. It resolves
tenant overrides before system baselines, pins the selected workflow revision,
records the Service and binding in trusted instance context, and safely replays
the same launch after an ambiguous response. Portal never accesses Workflow
Engine tables.
Active definitions reconcile durable API, one-time/interval schedule, platform-event, and parent-workflow trigger registrations. A shared worker claims due deliveries with scale-out-safe locking, rechecks the exact active revision and automation authority, and starts instances idempotently. Duration, deadline, and platform-event waits are persistent runtime state rather than human handoffs; event filters and variable mappings are bounded JSON expressions and never executable code. Cron remains an optional governed scheduler-adapter concern.
Consequential module actions are staged in Core's durable recovery ledger before provider dispatch. Conclusive effects commit with the Workflow projection. Lost acknowledgements block continuation and expose evidence-based Effect confirmed and Effect absent operator actions instead of a blind retry. Instance workers, trigger deliveries, and timers are fenced across hosts; linked Dataflow recovery states remain unresolved until Dataflow reports a conclusive outcome.
Checks
/mnt/DATA/git/govoplan/.venv/bin/python -m unittest discover -s tests