How to run custom code on notifications?

Kinto is able to execute some custom code when a particular event occurs. For example, when a record is created or updated in a particular collection.

Kinto uses the same thread to trigger notifications on events, so any custom code that is executed through a notification will block the incoming request until it’s done.

This design is useful when we want to ensure that something is done on the server before we send back the result to the client, and within the database transaction. But it’s usually preferrable to run the notifications asynchronously.

For the latter, the simplest way to run our custom code asynchronously is to use separate process workers that are notified via a job queue.

This tutorial presents the basic steps to run code both ways:

  • synchronously in Python;

  • asynchronously using a Redis queue, consumed via any third-party application.

Run synchronous code

In this example, we will track the creation of new buckets.

Implement a listener

Create a file tracker.py with the following scaffold:

from kinto.core.listeners import ListenerBase

class Listener(ListenerBase):
    def __call__(self, event):
        print(event.payload)

def load_from_config(config, prefix=''):
    return Listener()

Now, every time a new event occurs, we create a record in a tracker collection.

from kinto.core.listeners import ListenerBase

class Listener(ListenerBase):
    def __call__(self, event):
        backend = event.request.registry.storage
        userid = event.request.prefixed_userid
        backend.create(obj={'userid': userid}, resource_name='tracker')

def load_from_config(config, prefix=''):
    return Listener()

In order to keep advantage of what we’ve just tracked and show it into the response:

from pyramid.events import NewRequest, NewResponse

from kinto.core import utils as core_utils
from kinto.core.listeners import ListenerBase
from kinto.core.storage import Filter

class Listener(ListenerBase):
    def __call__(self, event):
        backend = event.request.registry.storage
        userid = event.request.prefixed_userid
        backend.create(obj={'userid': userid}, resource_name='tracker')

def count_created_buckets(event):
    userid = event.request.prefixed_userid
    if userid:
        backend = event.request.registry.storage
        filters = [Filter('userid', userid, core_utils.COMPARISON.EQ)]
        _, count = backend.get_all(resource_name='tracker', filters=filters)
        event.response.headers['Buckets-Created'] = str(count)

def load_from_config(config, prefix=''):
    config.add_subscriber(count_created_buckets, NewResponse)
    return Listener()

Add it to Python path

For the simplicity in this tutorial, we will just alter the PYTHONPATH system environment variable. Specify the path to the folder containing the tracker.py:

$ export PYTHONPATH="/path/to/folder:${PYTHONPATH}"

In order to test that it works, simply try to import it from a python script:

$ python
Python 2.7.9 (default, Apr  2 2015, 15:33:21)
[GCC 4.9.2] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import tracker
>>>

Enable in configuration

As explained in the settings section, just enable a new listener pointing to your python module:

kinto.event_listeners = tracker

kinto.event_listeners.tracker.use = tracker
kinto.event_listeners.tracker.actions = create
kinto.event_listeners.tracker.resources = bucket

Kinto should load the listeners without errors:

$ kinto start
Starting subprocess with file monitor
2016-01-21 16:21:59,941 INFO  [kinto.core.initialization][MainThread] Setting up 'tracker' listener

Test it

Create a bucket (using HTTPie):

$ http --auth alice:s3cr3t --verbose PUT http://localhost:8888/v1/buckets/bid1
$ http --auth alice:s3cr3t --verbose PUT http://localhost:8888/v1/buckets/bid2

Now, every response has a Buckets-Created header:

$ http --auth alice:s3cr3t --verbose GET http://localhost:8888/v1/

HTTP/1.1 200 OK
Content-Length: 66
Content-Type: application/json
Buckets-Created: 2
...

It worked!

Run asynchronous code

Note

You will need to install kinto-redis to use this listener.

In this part, we will take advantage of the built-in listener that delivers the events into a Redis queue. Separate scripts, also as known as “workers”, then consume the queue to execute custom asynchronous code.

Run Redis

Redis is available in most Linux distributions or Mac OS brew. Using Docker it is also very easy to run a server on localhost:6379:

$ sudo docker run -p 6379:6379 redis

Setup Kinto queue

In configuration, we setup the listener to post the message to a queue named eventqueue:

kinto.event_listeners = redis

kinto.event_listeners.redis.use = kinto_redis.listeners
kinto.event_listeners.redis.url = redis://localhost:6379/0
kinto.event_listeners.redis.pool_size = 5
kinto.event_listeners.redis.listname = eventqueue

Kinto should load the listeners without errors:

$ kinto start
Starting subprocess with file monitor
2016-01-21 16:21:59,941 INFO  [kinto.core.initialization][MainThread] Setting up 'redis' listener

Run worker(s)

The simplest worker would look like that:

import time
import json

import redis

def main():
    db = redis.Redis()
    # Run indefinitely.
    while True:
        # Wait for new messages (blocking).
        key, payload = db.blpop("eventqueue")
        # Decode JSON payload.
        message = json.loads(payload)
        # Simulate long task.
        time.sleep(2)
        print(message)

if __name__ == "__main__":
    main()

Run it in a separate terminal:

$ python worker.py

Test it!

Create a record (using HTTPie):

$ echo '{"data": {"note": "kinto"}}' | \
    http --auth alice:s3cr3t --verbose POST http://localhost:8888/v1/buckets/default/collections/notes/records

The server response is returned immediately.

But 2 seconds later, look at the worker output:

{'resource_name': 'record', 'user_id': 'account:alice', 'timestamp': 1453459942672, 'uri': '/buckets/c8c94a74-5bf6-9fb0-5b72-b0777da6718e/collections/assets/records', 'bucket_id': 'c8c94a74-5bf6-9fb0-5b72-b0777da6718e', 'action': 'create', 'collection_id': 'assets'}

It worked!