Delaying a workflow

Use pd.flow.delay to delay a step in a workflow.

These docs show you how to write Python code to handle delays. If you don't need to write code, see our built-in delay actions.

Using pd.flow.delay

pd.flow.delay takes one argument: the number of milliseconds you'd like to pause your workflow until the next step executes. You can pause your workflow for as little as one millisecond, or as long as one year.

Note that delays happen at the end of the step where they're called.

import random
 
def handler(pd: 'pipedream'):
    # Delay a workflow for 60 seconds (60,000 ms)
    pd.flow.delay(60 * 1000)
 
    # Delay a workflow for 15 minutes
    pd.flow.delay(15 * 60 * 1000)
 
    # Delay a workflow based on the value of incoming event data,
    # or default to 60 seconds if that variable is undefined
    default = 60 * 1000
    delayMs = pd.steps["trigger"].get("event", {}).get("body", {}).get("delayMs", default)
    pd.flow.delay(delayMs)
 
    # Delay a workflow a random amount of time
    pd.flow.delay(random.randint(0, 999))

Credit usage

The length of time a workflow is delayed from pd.flow.delay does not impact your credit usage. For example, delaying a 256 megabyte workflow for five minutes will not incur ten credits.

However, using pd.flow.delay in a workflow will incur two credits.

One credit is used to initially start the workflow, then the second credit is used when the workflow resumes after its pause period has ended.

cancel_url and resume_url

Both the built-in Delay actions and pd.flow.delay return a cancel_url and resume_url that lets you cancel or resume paused executions.

These URLs are specific to a single execution of your workflow. While the workflow is paused, you can load these in your browser or send an HTTP request to either:

• Hitting the cancel_url will immediately cancel that execution
• Hitting the resume_url will immediately resume that execution early

Since Pipedream pauses your workflow at the end of the step where you run call pd.flow.delay, you can send these URLs to third party systems, via email, or anywhere else you'd like to control the execution of your workflow.

import requests
 
def handler(pd: 'pipedream'):
  links = pd.flow.delay(15 * 60 * 1000)
  # links contains a dictionary with two entries: resume_url and cancel_url
 
  # Send the URLs to a system you own
  requests.post("https://example.com", json=links)
 
  # Email yourself the URLs. Click on the links to cancel / resume
  pd.send.email(
    subject=f"Workflow execution {pd.steps['trigger']['context']['id']}",
    text=f"Cancel your workflow here: {links['cancel_url']} . Resume early here: {links['resume_url']}",
    html=None
  )
 
  # Delay happens at the end of this step

When delays happen

Pipedream pauses your workflow at the end of the step where you call pd.flow.delay. This lets you send the cancel_url and resume_url to third-party systems.

def handler(pd: 'pipedream'):
  urls = pd.flow.delay(15 * 60 * 1000)
  cancel_url, resume_url = urls["cancel_url"], urls["resume_url"]
  # ... run any code you want here
 
  # Delay happens at the end of this step

Delays and HTTP responses

You cannot run pd.respond after running pd.flow.delay. Pipedream ends the original execution of the workflow when pd.flow.delay is called and issues the following response to the client to indicate this state:

$.respond() not called for this invocation

If you need to set a delay on an HTTP request triggered workflow, consider using time.sleep instead.

time.sleep

Alternatively, you can use time.sleep instead of using pd.flow.delay to delay individual workflow steps.

However, there are some drawbacks to using time.sleep instead of pd.flow.delay. time.sleep will count towards your workflow's compute time, for example:

import time
 
def handler(pd: 'pipedream'):
  # delay this step for 30 seconds
  delay = 30
  
  time.sleep(delay)

The Python step above will hold the workflow's execution for this step for 30 seconds; however, 30 seconds will also contribute to your credit usage. Also consider that workflows have a hard limit of 750 seconds.