2. Dataset Binding with AWS¶
We will create an AWS Lambda to continuously sync a bucket with Dataloop’s dataset
If you want to catch events from the AWS bucket and update the Dataloop Dataset you need to set up a Lambda.The Lambda will catch the AWS bucket events and will reflect them into the Dataloop Platform.
We have prepared an environment zip file with our SDK for python3.8, so you don’t need to create anything else to use dtlpy in the lambda.
NOTE: For any other custom use (e.g. other python version or more packages) try creating your own layer (We used this tutorial and the python:3.8 docker image).
2.1. Create the Lambda¶
Create a new Lambda
The default timeout is 3[s] so we’ll need to change to 1[m] (1 Minute):Configuration → General configuration → Edit → Timeout
Go to the Lambda console -> Select your function -> Configuration -> (Left-side panel) Environment variables -> Edit -> Add environment variable* Add the 3 secrets vars
DTLPY_PASSWORDTo populate the values for the vars:
DTLPY_PASSWORDyou’ll need to create a DataLoop Bot on your Dataloop project using the following code:
import dtlpy as dl dl.login() project = dl.projects.get(project_name='project name') bot = project.bots.create(name='serviceAccount', return_credentials=True) print('username: ', bot.id) print('password: ', bot.password)
Copy the following code:
import os import urllib.parse # Set dataloop path to tmp (to read/write from the lambda) os.environ["DATALOOP_PATH"] = "/tmp" import dtlpy as dl DATASET_ID = os.environ.get('DATASET_ID') DTLPY_USERNAME = os.environ.get('DTLPY_USERNAME') DTLPY_PASSWORD = os.environ.get('DTLPY_PASSWORD') def lambda_handler(event, context): dl.login_m2m(email=DTLPY_USERNAME, password=DTLPY_PASSWORD) dataset = dl.datasets.get(dataset_id=DATASET_ID) driver_path = dl.drivers.get(driver_id=dataset.driver).path for record in event['Records']: # Get the bucket name bucket = record['s3']['bucket']['name'] # Get the file name filename = urllib.parse.unquote_plus(record['s3']['object']['key'], encoding='utf-8') remote_path = None if driver_path == '/': driver_path = None if driver_path is not None and driver_path not in filename: return if driver_path: remote_path = filename.replace(driver_path, '') else: remote_path = filename if 'ObjectRemoved' in record['eventName']: # On delete event - delete the item from Dataloop try: dtlpy_filename = '/' + remote_path filters = dl.Filters(field='filename', values=dtlpy_filename) dataset.items.delete(filters=filters) except Exception as e: raise e elif 'ObjectCreated' in record['eventName']: # On create event - add a new item to the Dataset try: # upload the file path = 'external://' + filename # dataset.items.upload(local_path=path, overwrite=True) # if overwrite is required dataset.items.upload(local_path=path, remote_path=remote_path) except Exception as e: raise e
2.2. Add a Layer to the Lambda¶
We have created an AWS Layer with the Dataloop SDK ready. Click here to download the zip file.Because the layer’s size is larger than 50MB you cannot use it directly (AWS restrictions), but need to upload it to a bucket first.Once uploaded, create a new layer for the dtlpy env:
Go to the layers screen and “click Add Layer”.add_layer
Choose a name (dtlpy-env).
Use the link to the bucket layer.zip.
Select the env (x86_64, python3.8).
Click “Create” and the bottom on the page.
Go back to your lambda and add the layer:
Select the “Add Layer”.add_layer
Choose “Custom layer” and select the Layer you’ve added and the version.
click “Add” at the bottom.
2.3. Create the Bucket Events¶
Go to the bucket you are using, and create the event:
Go to Properties → Event notifications → Create event notification
Choose a name for the Event
For Event types choose: All object create events, All object delete events
Destination - Lambda function → Choose from your Lambda functions → choose the function you build → SAVE
Deploy and you’re good to go!