refactor: [caches/memory] drop "referenced object" support

[crandles]

Changes:

• refactor: [caches/memory] drop "referenced object" support
  • no noticeable performance differences; this optimization did not seem to be worth the complexity it caused within the overall codebase
  • calling code should be simpler now; should make it simpler to optimize the overall codebase (no more provider specific code paths)
• chore: [tests] use t.Context() over context.Background()
• • chore: [docs/dashboards] add redis datasource to grafana
• chore: [docs/dashboards] add more memory & request duration panels
  • added: "GOMAXPROCS" stat, 2 extra memory stats, cpu usage, go gc duration, processed bytes, frontend request duration histogram, + overall request duration histogram
  • (processed bytes not working via darwin at this time)

---

new panels:

more new panels:

[jranson]

re: references. For the memory cache, this will add a marshal immediately before every Store and an unmarshal immediately after every Retrieve, and those get more expensive as a time series object gets bigger (has more timestamps and/or series). Just want to double check that the CPU utilization doesn't go nuts as a result. The goal was that if many users are viewing the same dashboard with a fast auto-refresh that we're not repeatedly serializing/deserializing the identical payload over and over.

[crandles]

sure, let me spin up a before and after load test (along with a before ristretto)

I had something going with k6 somewhere, let me revisit that.

with any luck the singleflight layer assists us there, but we could perhaps use that at another level

[crandles]

good call, seeing some difference:

though I should increase my durations & ensure this load test is acting correctly