Event-Driven Pipelines

Trigger pipelines automatically based on platform events

Overview

Event-driven pipelines execute automatically when specific platform events occur. React to model deployments, cluster changes, or custom events without manual intervention.

Available Event Types

Event Type	When Triggered	Use Case
model.added	New model registered	Auto-deploy new models
model.updated	Model metadata changed	Update deployments
deployment.created	New deployment started	Configure monitoring
deployment.failed	Deployment failed	Send alerts, rollback
cluster.ready	Cluster becomes available	Deploy waiting models
cluster.unhealthy	Cluster health degraded	Trigger diagnostics

Creating Event Triggers

In the UI

1. Open your pipeline
2. Go to Triggers tab
3. Click Add Event Trigger
4. Configure:
   • Event Type: model.added
   • Filter (optional): JSON condition
   • Enabled: Toggle on
5. Click Save

Using the SDK

from bud import BudClient

client = BudClient()

# Create event trigger
trigger = client.event_triggers.create(
    pipeline_id="pipe_abc123",
    event_type="model.added",
    filter={
        "model_source": "hugging_face"
    },
    enabled=True
)

print(f"Event trigger created: {trigger.id}")

Event Filtering

Filter events to only trigger for specific conditions:

# Only trigger for production deployments
trigger = client.event_triggers.create(
    pipeline_id="pipe_monitoring",
    event_type="deployment.created",
    filter={
        "environment": "production",
        "cluster_id": "cluster_prod"
    }
)

Event Data Access

Access event data in your pipeline using the event context:

# In your pipeline actions, reference event data
{
    "id": "notify",
    "action": "notification",
    "params": {
        "message": "Model {{event.model_name}} was added",
        "recipients": ["[e-mail]"]
    }
}

Example: Auto-Deploy on Model Add

Automatically deploy models when they’re added to the registry:

from bud import BudClient

client = BudClient()

# Create auto-deployment pipeline
pipeline = client.pipelines.create(
    name="Auto-Deploy New Models",
    definition={
        "steps": [
            {
                "id": "health_check",
                "action": "cluster_health",
                "params": {
                    "cluster_id": "cluster_prod"
                }
            },
            {
                "id": "deploy",
                "action": "deployment_create",
                "params": {
                    "model_id": "{{event.model_id}}",
                    "cluster_id": "cluster_prod",
                    "deployment_name": "{{event.model_name}}-auto"
                },
                "depends_on": ["health_check"]
            },
            {
                "id": "notify",
                "action": "notification",
                "params": {
                    "message": "Deployed {{event.model_name}} successfully",
                    "channel": "slack"
                },
                "depends_on": ["deploy"]
            }
        ]
    }
)

# Create event trigger
trigger = client.event_triggers.create(
    pipeline_id=pipeline.id,
    event_type="model.added",
    filter={
        "model_source": "hugging_face",
        "auto_deploy": True
    },
    enabled=True
)

Example: Failure Alert Pipeline

Send alerts when deployments fail:

pipeline = client.pipelines.create(
    name="Deployment Failure Alerts",
    definition={
        "steps": [
            {
                "id": "log_failure",
                "action": "log",
                "params": {
                    "message": "Deployment failed: {{event.deployment_id}}",
                    "level": "error"
                }
            },
            {
                "id": "send_alert",
                "action": "notification",
                "params": {
                    "message": "🚨 Deployment {{event.deployment_name}} failed: {{event.error_message}}",
                    "channel": "slack",
                    "priority": "high"
                }
            }
        ]
    }
)

trigger = client.event_triggers.create(
    pipeline_id=pipeline.id,
    event_type="deployment.failed",
    enabled=True
)

Managing Event Triggers

List All Triggers

# Get all event triggers for a pipeline
triggers = client.event_triggers.list(pipeline_id="pipe_abc123")

for trigger in triggers:
    print(f"{trigger.event_type}: {trigger.enabled}")

Disable a Trigger

# Temporarily disable
client.event_triggers.disable(trigger_id="trig_xyz789")

Update Filter

# Update event filter conditions
client.event_triggers.update(
    trigger_id="trig_xyz789",
    filter={
        "environment": "production",
        "priority": "high"
    }
)

Best Practices

Use Filters: Avoid triggering on every event - filter for relevance

Idempotent Actions: Ensure pipeline actions are safe to retry

Add Logging: Track which events triggered executions

Test Events: Manually trigger test events before enabling

Monitor Executions: Watch for unexpected trigger frequency

Event Data Structure

Each event includes standard fields:

{
  "event_type": "model.added",
  "event_id": "evt_abc123",
  "timestamp": "2024-01-29T12:00:00Z",
  "source": "budmodel",
  "data": {
    "model_id": "model_xyz789",
    "model_name": "Llama-3.2-1B",
    "model_source": "hugging_face",
    "created_by": "[e-mail]"
  }
}

Troubleshooting

Pipeline not triggering

Cause: Event filter too restrictive or trigger disabled

Solution: Check filter conditions, verify trigger is enabled

Too many executions

Cause: Events firing more frequently than expected

Solution: Add more specific filters, check event source

Missing event data

Cause: Event payload doesn’t include expected fields

Solution: Check event schema, add fallback values in pipeline

Next Steps

Scheduled Pipelines

Run pipelines on a schedule

Troubleshooting

Common issues and solutions