every_x_ms sampling?

[z4jiang]

I have limited experience with tracing and I'm looking for ideas/pointers.

I'm trying to do an every_x_ms sampling (per site). This is a little more useful than random coinflip sampling.
Semantics: sample in if no samples yet from a specific site in the last x ms.
Subscriber can do a shared synchronized collections of Metadata -> last sample timestamp to achieve this, but this seems prone to contention.
It seems that an extension like thing on Metadata would be very useful, in that it's per site, and persists across spans of the same Metadata.
More generally, and extra bonus if somehow can hook in custom "accumulation". e.g. sum how many times sampled out in between, or add some custom values from fields.