Replace fake webserver with Avalanche

[saketjajoo]

Currently, the setup uses fake webserver as the metrics emitter which only supports standard metrics (e.g. the debug/* metrics and the process and go metrics). Having variable/configurable metrics would help collect good benchmarking data. Hence, I suggest replacing the fake webserver component with avalanche.

Avalanche provides several configurable options to change the configurations (e.g. metrics count, label count, series count, metric name length, etc.). Although, this may not represent production-like metrics (for e.g. metric names are like avalanche_metric_mmmmm_01 does not look like a real workload metric), having different configurations can help collect good data to benchmark and analyze the Prometheus's performance.

---

Sample / Reference Commit: https://github.com/saketjajoo/test-infra/commit/eb8678bdfa2f5404bbf3997103f20cfb18ca6877

[bwplotka]

This is still relevant, especially after the latest changes in 0.6.0 prometheus-community/avalanche#101

[bwplotka]

Did a quick research around swapping fake-webserver with avalanche.

I think I would propose we keep fake-webserver. We could simplify it a bit and move to avalanche project as another mode to improve maintainability.

WDYT @ArthurSens @bboreham @kakkoyun

Closing for now, we can reopen.

[ArthurSens]

I think we'll need avalanche to Benchmark CTs, but agree this can be done in a custom benchmark

[bwplotka]

We can (and should) add native histograms and CTs to fake-webserver IMO (pull the latest client and configure nhs)