Use per-scheduler stack pools (let's recycle) #14100
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The stack pool was only used to create new stacks when MT was enabled. The stacks were then pushed to each scheduler's free stacks, and eventually dropped on reschedule (after context swap, so we're sure to only ever deallocate stacks that are no longer used). This led the stacks to never be reused with MT. Only allocated, put back into a pool, then deallocated.
This patch changes the behavior to have a stack pool running on each scheduler, and to use it to create and collect the stacks, and reuse them when possible. It also drops the mutex since the stack pool can never be accessed in parallel (in fact it never was).
Also starts a collecting fiber on each thread.
Stacks won't necessarily return to the pool that created them. For example if thread A allocates a stack, then the scheduler sends the fiber to thread B, then the stack will be returned to thread B's scheduler.
It may thus only lead to better performance if there are different fibers, running on multiple threads that are spawning fibers. It won't have much (or any) impact if there is only one fiber spawning other fibers (e.g. a HTTP::Server).