Cache dictionary

[kaykhancheckpoint]

Can you use this to cache a large dictionary?

[keithrozario]

Depends how large -- and how would the dictionary be populated? Via file from S3?

[kaykhancheckpoint]

Hmm not quite, lets say i have built an object which has taken a long time due to some complex computation, now i wish to cache this object because running that computation every time the lambda function fires is time consuming.

cache = {}

if cache["data"]:
    do_some_work()
else:
   some_class_obj = some_large_computation()
   cache["data"] = some_class_obj 
   do_some_work()

Do you have any suggestions on how i can deal with this

[keithrozario]

I see. The solution I suggest is to setup the code to generate the dictionary outside of the handler.

# everything here will execute during init time
cache = {}
some_class_obj = some_large_computation()
cache["data"] = some_class_obj 
do_some_work()
# end of section

def handler(event, context):
    other_stuff()
    return

just be careful if the cache dictionary requires more than 10 seconds, this might not work.

[kaykhancheckpoint]

I think initially it does take more than 10 seconds. So like you said this probably would not work for my usecase. Out of curiosity why is there a 10~ second limit?

The other problem i see with this approach is that you have to pass the cached object around. The function that actually requires the cached object may be 3-4 levels down from the handler function.

[keithrozario]

ok, I'll look into it.

The other option you can do while I work on this, is to simply generate the dictionary, and save it into global variable. Global variables remain constant across invocations of a single execution context :).

[kaykhancheckpoint]

Ye i have used cachetools https://pypi.org/project/cachetools/ to save the obj into a dictionary but when the aws lambda container shuts down the cache is reset. The container shuts down if there are no calls made to it after 30~ minutes (I think)

Right now im thinking of pickling the obj and storing it onto s3 and then retrieving it and unpickling it

[keithrozario]

Yes, that seems the most reasonable approach. Although I wouldn't use pickle 👍

Because the lambda container will expire in minutes if not used (typically 10-30), and you'd lose everything in the container including any cache items you had.

[kaykhancheckpoint]

Yeah thats why i was going to pickle and store on s3 and retrieve from s3 so the cache wouldint be in the container.

Anyway after trying the above i decided it wasen't worth it, it took a long time to unpickle the file once i recieved it from s3.

I decided to move away from serverless for this application im working on and instead im deploying it on kubernetes as an api, where i can store it in cache that will last for how ever long i set the TTL to be.