improve initialisation time
jannikmi opened this issue · comments
currently the initialisation time of a Timezonefinder instance is around .7 seconds.
This is probably mostly due to reading the polygon index file into memory.
Find ways of reducing the startup time without affecting the query time
Data point: I upgraded my script from timezonefinder 5.2.0 to 6.2.0, and execution time in AWS Lambda went from about 350ms to between 8 and 8.5 seconds. I can't say for sure that it was initialization time at fault, but the problem went away when I downgraded back to 5.2.0.
That sounds like a lot, but it could be due to low disk read throughput.
Have you checked that you are reusing your Timezonefinder instance? (Cf. #176)
Yeah, I don't know enough about Lambda to know why it's so much worse, but it makes sense that it might be more IO-bound than other environments when doing big reads like this. I am reusing my TimezoneFinder instance (or, more accurately, I'm only using it once).
At least you could make sure to reuse your instance between consequent lambda calls (e.g. as global variable).
Would you be willing to look into the improvement of the initialisation time?
This is a very low traffic personal project, so using a global variable wouldn't help much (most executions are cold starts), and I don't want to use provisioned concurrency (even though it would only be a few dollars a month, the project currently costs about a penny per month). I'm not a skilled enough library developer to help with this issue, but I'd happily test any new versions.