[zzylol]

Part 2 of the telemetry-functions stack.

The official OpenTelemetry traffic generator cannot yet shape attribute cardinality, key popularity or value distributions, or change them over time: synthetic mode uses fixed value pools, and semantic_conventions mode sends every metric value as 1.0 (its TODO: "add configurable value range, distribution"). These options are planned upstream. Until then, this receiver stands in, with a config shaped like the upstream proposal so it can be swapped out.

What it does

urn:asap:receiver:scenario_generator sends Gauge points at signals_per_second, in max_batch_size batches spread evenly over each second, with the official generator's control handling.

config:
  seed: 1
  signals_per_second: 20000
  metric_name: http.server.request.duration
  attributes:
    - {key: http.route, cardinality: 200, zipf: 1.1}
    - {key: user.id, unique: true}
    - {key: k8s.pod.name, cardinality: 50, growth_per_minute: 0.05}
  value: {lognormal: {median: 50.0, sigma: 0.6}}
  changes:
    - {after_seconds: 600, attribute: http.route, equals: http.route-3,
       value: {lognormal: {median: 150.0, sigma: 0.6}}}
    - {after_seconds: 600, attribute: http.route, inject: /new/hot-route, share: 0.15}

Values are seeded, so a seed and config reproduce the same stream. Unique values include the seed, so two sources never share them.

The points depend only on scenario time. Batch i of the n batches of second s after the scenario start is sent at s + i/n seconds, stamped with that time, and drawn from an RNG seeded by (seed, s, i). The start is the Unix second in ASAP_SCENARIO_START (the next whole second when unset), and changes count from it. Generators with the same config and start therefore send identical points even if they start a little apart; a generator that falls behind skips batches rather than shifting later ones. This is what lets #25 feed three separate pipelines the same input.

Validation

Unit tests cover cardinality (fixed and unique), Zipf skew, changes after their time, reproducibility, and that a batch depends only on its second and index. cargo fmt --check, cargo clippy --all-targets --features otap-engine,stream-benchmark -- -D warnings and cargo test --features otap-engine,stream-benchmark pass at this PR's head.