Reuse session in rest requests

[hackenbergstefan]

Changelog: BugFix: Reuse session in ConanRequester to speed up requests.
Docs: omit

fixes #14794

[CLAassistant]

All committers have signed the CLA.

[memsharded]

Hi @hackenbergstefan

Thanks for your contribution.
If this is to be considered for 2.0.X releases, please target the release/2.0 branch in your PR.

It seems that the change broke some integrations tests that have some mocking, if you can please try to have a look, and if it makes sense to fix the mocking for the tests to pass, or something else is happening. If not, don't worry, we can try to have a look ourselves.

I am curious, how did you get that deep into the Conan code to realize this? What effects were you seeing, like slower than expected http calls?

[hackenbergstefan]

Hey!
First of all, thank's for the quick reply! 👍

I am curious, how did you get that deep into the Conan code to realize this? What effects were you seeing, like slower than expected http calls?

I'm currently using Conan 1 migrating to Conan 2. During this migration I realized that Conan 2 was much slower. This disappointed me and my colleagues and so I started digging into ;-)

[hackenbergstefan]

Thanks for your contribution. If this is to be considered for 2.0.X releases, please target the release/2.0 branch in your PR.

Should I open two PRs? One for develop2 and one for release/2.0?

[memsharded]

Interesting! That much slower than caching the session really makes a big difference?
Besides some slowness introduced by the compatibility checks of cppstd Conan 2.0 shouldn't be slower than 1.X (and of course besides this possible bug). The cppstd thing it is in our todo-list.

If this bugfix improves noticeable the speed, that would be great, thanks again for looking into this!

[memsharded]

Should I open two PRs? One for develop2 and one for release/2.0?

Not necessary, we merge release/2.0->develop2 regularly.

[hackenbergstefan]

If this bugfix improves noticeable the speed, that would be great, thanks again for looking into this!

On my client and infrastructure I had these differences:

• Conan 1 (Session): 0.4-0.1ms per ConanRequester._call_method
• Conan 2 (no Session): 100-300ms per ConanRequester._call_method

[AbrilRBS]

This is great! Thanks a lot for going above and beyond and investigating the root cause of the slowdown :)

[hackenbergstefan]

This is great! Thanks a lot for going above and beyond and investigating the root cause of the slowdown :)

Sure! This is how Open-Source should work in my opinion ;-)

[AbrilRBS]

Seeing a (subjective) improvement in compatibility resolution with this change too, linking #14799 to not lose track of improvements on that area

[memsharded]

Amazing. Installing some large graph (only recipes + package_id computation), reduced from 22 seconds to 11 seconds.

Thanks again for contributing this! It will be in next 2.0.12

[SpaceIm]

Thanks a lot, it's way way way faster when conan checks new RREV, PREV and compatible packages.