The variable outside lambda handler method is not preserved in the subsequences requests

[phuclh]

I have a python script like this:

model = None

def get_model():
    global model

    if model is not None:
        return model

    model = SentenceTransformer('all-MiniLM-L6-v2')

    return model

def lambda_handler(event, context):
    model = get_model()

When the model = SentenceTransformer('all-MiniLM-L6-v2') is initialized it will download the library and this process takes a while. So I put it in the get_model method and return the model from the previous request for subsequence requests to save time.

However, when I execute the lambda function via sidecar within a route.

\App\Sidecar\Functions\Test::execute()

I got timeout for that request, so I put that code into a job and run it in a queue. This time, it ran successfully and I expected the get_model method would return the model from the 1st run in the queue when I re-visit the above route, but it didn't. However, it did return when I dispatched another job to queue (I knew that because this time the job ran much faster).

So I wonder if AWS Lambda creates a different instance for QUEUE and HTTP Request?

[aarondfrancis]

The variable is guaranteed to stay populated if it's global, but you have no guarantees that you'll get the same container ever again unfortunately.

You could try warming, which might help!