[question][guidance] bulk increment with ttl about cachex

[whitfin]

@xward hi! Sorry for the slow response, busy month.

I took a look at your code and there's not much you can do about the incr call themselves, however you can speed it up a little bit by re-using a static state:

Cachex.execute(@store, fn cache ->
    Enum.reduce(checkers, %{}, fn {id, key, period, allowed, decision_if_above}, acc ->
        {:ok, n} = Cachex.incr(cache, key)

        # new ? set ttl
        if n == 1, do: Cachex.expire(cache, key, period)

        Map.put(acc, (n > allowed && decision_if_above) || :pass, id)
    end)
end)

This will skip an ETS table lookup on each call to your cache, which actually is most of the overhead involved here. If you want to try this out, you should see a fair amount of speed up - it might be enough for your use case.

Let me know! Happy to discuss further if you need further improvements, maybe we can figure out something in the API that might help.

[xward]

It indeed improved performances, thanks ! ~25% with 500 checkers (cf benchmarks)

I guess there is no way around calling incr/2 for each checkers I have.

[whitfin]

@xward there's a potential that something this might also be a little faster.

I haven't been able to test this, not on my work machine - but hopefully you get the idea. Let me know if it helps!

    Cachex.execute(@store, fn cache ->
      Enum.reduce(checkers, %{}, fn {id, key, period, allowed, decision_if_above}, acc ->
        Cachex.get_and_update(cache, key, fn
          (nil) -> { :commit, 1, ttl: period }
          (val) -> { :commit, val + 1 }
        end)
        Map.put(acc, (n > allowed && decision_if_above) || :pass, id)
      end)
    end)

The idea is that it combines it all into one single cache call, rather than multiple - although you still need to call per key.