Memory store clean command is a blocker about limiter HOT 9 CLOSED

Sure!

from limiter.

Hello,

Do you have any suggestions by any chances ?
Also, have you tried using redis store ? The in-memory store was more or less designed for debug and/or development.

Cheers,

from limiter.

Could I have an example of your expiration duration ?

from limiter.

Hello,

Do you have any suggestions by any chances ?
Also, have you tried using redis store ? The in-memory store was more or less designed for debug and/or development.

Cheers,

I think redis store would remove the the bottleneck since there is no cleaner on the app but for a single deployment it would create unneeded latency.
For the in-memroy the deletion of expired records could be done like this https://github.com/gadelkareem/cachita/blob/master/memory.go#L82 because of this issue golang/go#20135 but that would not be the best solution yet since cleaner will block again. I am not sure how to tackle the in-memory cleaner but maybe if there is a replica of the counters hash map that is used only for reading and the writing could be done through channels in go routines. But I am not sure how to do increment in this way.

Could I have an example of your expiration duration ?

generalLimiter = newLimiter("4-S")
apiLimiter = newLimiter("50-M")
searchLimiter = newLimiter("30-M")
func newLimiter(rate string) *limiter.Limiter {
    r, err := limiter.NewRateFromFormatted(rate)
    h.PanicOnError(err)
    return limiter.New(memory.NewStore(), r, limiter.WithTrustForwardHeader(true))
}

from limiter.

I may have a solution using a ring buffer, would you like to experiment it ?

from limiter.

@gadelkareem Could I have an example on how many keys you have in your store, so I could have a good telemetry on how my idea improve (or not) things ?!

Thank you.

from limiter.

@novln I am afraid that would be hard to calculate because it is production data but we can estimate it through the number of requests which is around 1.2 Million request per day.

from limiter.

Hello,

Could you try this commit 3cd0e5a?

I haven't done a benchmark nor a ring buffer in order to try to keep the initial architecture.
Also, it was quicker to try this and find a quick and good enough solution for your problem.

from limiter.

I tested it with stack impact and I do not see that blocker anymore but i made a PR so you can test it more intensely.

from limiter.