Memoization on fly/firecracker

1

I’m building an application that utilizes memoization for some heavy computation calls between function invocations. Is this pattern compatible on a serverless platform like fly? My initial testing seems to indicate no, there is no way to save in-process memory state across queries.

2

Fly is not serverless in this way. Fly deploys long-lived containers for your application, so memory state is preserved between requests to a single container, so you can cache results in memory and have them preserved across requests.

3 Likes
3

@Eric_Pauley is correct. If you need this data to persist longer than the life of a VM you can persist data in files on a volume or an external service like redis.

1 Like
4

@E.G can you describe what type of app you’re building and what you’re trying to do? What do you mean by query there? Are you looking to snapshot a VM and restore it’s memory state later?

As mentioned earlier, we’re not a “functions as a service” platform. VMs are long-lived and can handle many requests during their lifetime. Caching in memory is a common, but it is lost when a VM stops. We’re hoping to offer snapshots and restores at some point though.

5

Can you describe what type of app you’re building and what you’re trying to do? What do you mean by query there? Are you looking to snapshot a VM and restore it’s memory state later?

Thanks Michael and Eric. That makes sense. A simple python script should illustrate my question. By query I’m referring to an API query. Just asking about the lifecycle of the vm and the state of memory between invocations. I think your first answer was what I needed.

CACHE = None

def populate_cache(data_seed):
    global CACHE
    # resource intensive cache population code w/ request data seed
    # ...

def get_cached_data(request):
    global CACHE
    if CACHE is None:
        populate_cache(request.data)
    return CACHE