Scheduled Triggers

Use flexible schedules to launch pipelines and executions automatically

Schedule triggers to run workloads automatically on a recurring basis. Perfect for nightly training runs, batch predictions, data pipeline updates, or any workflow that needs to happen regularly.

Why Schedule Your Workflows?

Consistency: Run training at the same time every day with fresh data

Efficiency: Launch batch jobs during off-peak hours to optimize costs

Hands-free: No manual intervention needed, just set it and forget it

Flexibility: Use simple recurring schedules or advanced cron expressions

---

How Scheduling Works

Scheduled triggers require two components:

1. Scheduled (Cron) condition: Defines when the trigger fires

2. Action: Defines what to launch (execution, pipeline, etc.)

Scheduled triggers launch no earlier than the specified time and no more frequently than once per minute.

If multiple conditions are added to a trigger, all conditions must be met at the same time for the trigger to launch.

---

Quick Start Example

Let's create a simple scheduled execution that runs every weekday at noon UTC.

Step 1: Add the Execution Step

Add this step to your valohai.yaml:

- step:
    name: Scheduled Execution
    image: python:3.11
    command:
    - pip install valohai-utils
    - python ./scheduled.py

Create scheduled.py:

import valohai
import datetime

valohai.prepare(
    step="Scheduled Execution",
)

month_day = datetime.datetime.now().strftime("%-d %b")

print(f"Today is {month_day}")

Commit and push these changes to Git, then fetch the new commit in Valohai.

Step 2: Create the Trigger

1. Go to Project Settings → Triggers

2. Click Create Trigger

3. Set Trigger Type to "Scheduled" (default)

4. In the Conditions column, click + Add → select Scheduled (Cron)

   • Select "On Weekdays"

   • Select "12:00"

5. In the Actions column, click + Add → select Run Execution

   • Source Commit Reference: main (or your primary branch)

   • Execution Step Name: Scheduled Execution

   • Execution Title: Daily execution at 12 UTC

6. Click Save Trigger

The trigger will now launch automatically every weekday at noon UTC.

---

Scheduling Options

Simple Recurring Schedules

For common patterns, use the visual scheduler:

• Daily at a specific time

• Weekdays only (Monday–Friday)

• Weekly on specific days

• Monthly on specific dates

These are easy to configure and cover most use cases.

Advanced Cron Expressions

For complex schedules, use cron syntax directly:

┌───────────── minute (0 - 59)
│ ┌───────────── hour (0 - 23)
│ │ ┌───────────── day of month (1 - 31)
│ │ │ ┌───────────── month (1 - 12)
│ │ │ │ ┌───────────── day of week (0 - 6, Sunday = 0)
│ │ │ │ │
* * * * *

Common cron patterns:

Pattern	Cron Expression	Description
Every hour	0 * * * *	On the hour, every hour
Every 6 hours	0 */6 * * *	At 00:00, 06:00, 12:00, 18:00
Every Monday at 9 AM	0 9 * * 1	Weekly on Mondays
First of the month	0 0 1 * *	Monthly at midnight
Every 15 minutes	*/15 * * * *	Four times per hour

Use cronmaker.com to generate and validate complex cron expressions.

---

Common Scheduling Patterns

Pattern: Nightly Model Training

Goal: Retrain your model every night at 2 AM UTC with the latest data

# valohai.yaml
- step:
    name: nightly-training
    image: pytorch/pytorch:2.0.0-cuda11.7-cudnn8-runtime
    command:
      - python train.py {parameters}
    inputs:
      - name: dataset
        default: datum://latest-training-data

Trigger setup:

• Type: Scheduled

• Condition: Cron 0 2 * * * (daily at 2 AM)

• Action: Run execution nightly-training

---

Pattern: Hourly Batch Predictions

Goal: Run inference on new data every hour

# valohai.yaml
- pipeline:
    name: hourly-inference
    nodes:
      - name: fetch-new-data
        step: fetch-data
        type: execution
      - name: run-predictions
        step: predict
        type: execution
    edges:
      - [fetch-new-data.output.*, run-predictions.input.data]

Trigger setup:

• Type: Scheduled

• Condition: Cron 0 * * * * (every hour on the hour)

• Action: Run pipeline hourly-inference

---

Pattern: Weekly Performance Report

Goal: Generate model performance metrics every Monday morning

# valohai.yaml
- step:
    name: weekly-report
    image: python:3.11
    command:
      - pip install -r requirements.txt
      - python generate_report.py
    outputs:
      - name: report.pdf

Trigger setup:

• Type: Scheduled

• Condition: Cron 0 9 * * 1 (Mondays at 9 AM)

• Action: Run execution weekly-report

---

Monitoring and Debugging

View Trigger Logs

Check if your scheduled triggers are firing correctly:

1. Go to Project Settings → Triggers

2. Find your trigger in the list

3. Click the ... menu → View Logs

Logs show every trigger attempt, whether it succeeded or failed, and why.

Common Issues

Trigger not firing?

• Check that the trigger is Enabled in the trigger settings

• Verify the cron expression is correct

• Ensure you have at least one condition set

Trigger firing at wrong time?

• All times in Valohai are UTC. Convert your local time to UTC.

• Use time.is/UTC to check current UTC time

Multiple triggers firing at once?

• This is normal for triggers with the same schedule

• Each trigger launches independently

• Consider consolidating into a single trigger if they should be coordinated

Triggers that repeatedly fail will auto-disable. Check logs and re-enable the trigger after fixing the issue.

---

Next Steps

• More trigger types? Explore Webhook Triggers and Notification Triggers

• Need custom logic? Check out the REST API

• Building complex workflows? Learn about Pipelines