feat: add fixed_window bucket option

[pubalokta]

Original PR by Panga

Add

• fixed_window bucket option when enabled refill at specified interval instead of granular.
• fixed_window optional parameter in the take and takeElevated methods. The parameter takes precedence over the bucket option, making it easier to activate/deactivate using a feature flag on the client side.

Used for enforcing a rate limit policy that does not allow allow request rate to go over a the size in a given interval.

Examples

Default behavior

buckets = {
  ip: {
    size: 1000,
    per_second: 1000
  }
}

Result: Refill 1 token each 1 millisecond.

New fixed_window: true

buckets = {
  ip: {
    size: 1000,
    per_second: 1000,
    fixed_window: true
  }
}

Result: Refill 1000 tokens each 1000 milliseconds.

[kampde]

couldn't really review all, will continue tomorrow.

[kampde]

What would be the benefit or use case of being able to enable this on demand? What would happen if clients are doing requests with fixed window on and off all the time?

[pubalokta]

What would be the benefit or use case of being able to enable this on demand?

We would be able to do a controlled rollout. Nowadays everyone is using sliding window; we want to introduce the fixed window algorithm but wanna make sure the transition is controlled. If something doesn't work as expected sending the fixed_window=false (feature flag) param will fallback to sliding window

What would happen if clients are doing requests with fixed window on and off all the time?

That shouldn't happen. It's the same risk for clients calling take with and without overrides.

[kampde]

typo

Suggested change

	On top of that, you can use the `fixed_window` property in the `configOverride` parameter to safely activate/deactivate the new fixed_window algorithm from the client on-demand. This parameter acts as a feature flag to safely deploy the change to produciton.
	On top of that, you can use the `fixed_window` property in the `configOverride` parameter to safely activate/deactivate the new fixed_window algorithm from the client on-demand. This parameter acts as a feature flag to safely deploy the change to production.

[pubalokta]

fixed