Skip to main content

Inspecting Events

Events provide an additional layer of observability in production systems. By following a trail of events, you can inspect what was triggered, when, and by whom.

Accessing events through Client API​

You can inspect the event(s) with the Client API that exposes the MetaflowTrigger object for every event-triggered run. For instance, we can inspect the event that triggered a production run ModelRefreshFlow/argo-modelrefreshflow-rlpgc:

from metaflow import namespace, Run
namespace(None)
Run('ModelRefreshFlow/argo-modelrefreshflow-rlpgc').trigger.event

Remember that namespace(None) sets the global namespace which is required to inspect production runs e.g. in a notebook. This will print out metadata about the event:

MetaflowEvent(name='data_updated',
              id='ca75a1a4-91de-40c2-944c-0b39436c721e',
              timestamp=datetime.datetime(2023, 5, 15, 19, 50, 43),
              type='event')

Find a description of all fields in the related API documentation.

Dealing with multiple triggers​

If a run was triggered by multiple events, you can inspect specific MetaflowEvent by name:

run.trigger['data_updated'].name

Or, if a run was triggered by multiple flows when using @trigger_on_finish, you can inspect a specific triggering Run by its flow name:

run.trigger['FirstFlow'].data.model

Following a trail of runs​

When flows are connected with @trigger_on_finish, you can use MetaflowTrigger to follow a chain of triggered runs. Consider this chain of flows:

ETLFlow → TrainingFlow → PredictionFlow

We can walk backwards from the latest run of PredictionFlow back to the triggering run of ETLFlow:

etl_run = Flow('PredictionFlow').latest_run.trigger.run.trigger.run

Events in the Metaflow UI​

If you have deployed the Metaflow GUI, you can view metadata about triggers right in the UI. The circles with arrows inside indicate event-triggered runs: