Status of MarkusSintonen's speedup PRs?

[jaddison]

Unless I'm misreading the situation, @MarkusSintonen worked with maintainers to improve concurrent request performance through several PRs over the course of a few weeks.

Many of those PRs currently sit in an open state, months after the last comment - they seemed to be trending in a positive direction. I'm curious if there are plans to restart efforts and move them forward?

[tomchristie]

My views on this have changed since those conversations started... the smart resolution here is to reduce the complexity of the connection pooling implementation until there simply isn't space left for any serious performance issues. I'm not able to expand on this just yet, tho will do so when possible.

[MarkusSintonen]

My views on this have changed since those conversations started... the smart resolution here is to reduce the complexity of the connection pooling implementation until there simply isn't space left for any serious performance issues. I'm not able to expand on this just yet, tho will do so when possible.

How about fixing the perf bugs first and then later refactoring into a better state? As the proposed fixes work without issues.

[tomchristie]

As the proposed fixes work without issues.

That's not true. Review is non-trivial and changes introduce trade-offs that aren't obviously beneficial.

As one example... the benchmarking has introduced dependencies into the httpcore requirements that're currently breaking the build and that shouldn't be part of our CI.

Eg see encode/httpcore#1001 (comment) and encode/httpcore#964

[MarkusSintonen]

introduce trade-offs that aren't obviously beneficial.

What are these for example?

As one example... the benchmarking has introduced dependencies into the httpcore requirements that're currently breaking the build and that shouldn't be part of our CI.

Well its not exactly related to the performance fixes. Benchmark was already run and showed how badly its working with non trivial concurrency levels. Benchmark can be just removed. Although CI here not running any perf checks is not great...

[MarkusSintonen]

I'm like constantly getting pinged about the issues in httpx. People keep on asking what is the status as fixes are not moving to any direction.

[T-256]

As one example... the benchmarking has introduced dependencies into the httpcore requirements that're currently breaking the build and that shouldn't be part of our CI.

fwiw, there is an open PR for that: encode/httpcore#965