Event recording & summarize need cleanup with respect to self-time vs incr-load-time vs blocked-time vs total-time.

[michaelwoerister]

At the moment, we have not specified how self-time and other kinds of measurements interact with each other:

• Does the incremental cache loading time contribute to a queries self-time?
• In the incremental cache loading case, is the loading event nested inside a regular query-provider event?
• Should we distinguish between actual cache loading, metadata loading, and provider re-executions caused by missing cache entries? If so, how?
• Do query-blocked time contribute to self-time? How about total-time?
• Is a query-blocked event always followed by a cache-hit event (the latter of which would then be responsible for incrementing the invocation count) What if cache-hit events are filtered out? Is it OK to have wrong invocation counts?

[michaelwoerister]

My suggested answers to these questions are:

Does the incremental cache loading time contribute to a queries self-time?

Yes. Self-time is what one sorts for when trying to find out what is expensive. It would be unhelpful if something got sorted to the back even though 50% of total compile time is spent loading things for it from the cache.

In the incremental cache loading case, is the loading event nested inside a regular query-provider event?

Let's go with no. That causes fewer events to be generated. Also, in order to make things really uniform, in-memory cache hit events would have to wrapped with a query-provider event too, which would be lots of overhead for not much gain.

Should we distinguish between actual cache loading, metadata loading, and provider re-executions caused by missing cache entries? If so, how?

Down the line, yes. It's a useful distinction to make. These should all be separate query kinds (incr. cache loading already is). However, I think this is something to be implemented after we have query keys, so we don't make too many changes at once.

Do query-blocked time contribute to self-time? How about total-time?

Blocking should add to self-time for the same reason as incr. cache loading should add to self-time (see above). I don't remember what I meant by "How about total-time?", I think the question is not relevant anymore since we switched to @andjo403's approach of computing total time as thread_max_time - thread_min_time.

Is a query-blocked event always followed by a cache-hit event (the latter of which would then be responsible for incrementing the invocation count) What if cache-hit events are filtered out? Is it OK to have wrong invocation counts?

I haven't thought about this one yet.

[michaelwoerister]

Is a query-blocked event always followed by a cache-hit event (the latter of which would then be responsible for incrementing the invocation count) What if cache-hit events are filtered out? Is it OK to have wrong invocation counts?

It turns out the natural way of implementing this is that a cache-hit event is indeed generated after each query-blocked event. So this does not behave differently than the regular case.