Tracing: What do we want to measure and how

[o0Ignition0o]

Followup to a comment, now that we have stable tracing we might wanna sit back and figure out what we would like to trace, and at what level.

IMO this calls for a group discussion, I hope this issue provide such a space.

Do we really need to measure so much? @BrynCooke any opinion on this?

[Geal]

I think we should measure a lot. We are still learning how this software works, so we need to see clearly its internal behaviour to build a good mental model. This may not be the case later, when we're more confident and know which measurements are really useful.
In the meantime, we can reduce the amount of data by carefully setting the log levels (keeping debug or trace for our usage) and sampling traces. Example: sending all traces for failing queries, but 1 in 10 successful queries, because the successful ones are less interesting and the common case

[abernix]

This issue is getting a bit stale at this point and i'm not sure if we have anything actionable on it today, particularly given that we've done a lot of measuring (over measuring, in a lot of cases!) in our tracing and have observed — in most cases —  the performance and verbosity implications of doing so.

Do we want to do anything else with this right now, concretely?

[o0Ignition0o]

I don't think we do, we have changed this quite a bit over time, so i m closing this.

Happy to reopen later if needs be