Page MenuHomePhabricator

[services] Rust Integration - c++ - Implement worker in bidi reactor
ClosedPublic

Authored by karol on Sep 6 2022, 7:21 AM.
Tags
None
Referenced Files
Unknown Object (File)
Sun, Nov 10, 6:57 AM
Unknown Object (File)
Thu, Nov 7, 6:05 AM
Unknown Object (File)
Tue, Nov 5, 11:29 AM
Unknown Object (File)
Tue, Nov 5, 11:29 AM
Unknown Object (File)
Sun, Nov 3, 7:01 AM
Unknown Object (File)
Tue, Oct 29, 4:23 PM
Unknown Object (File)
Sun, Oct 27, 8:13 PM
Unknown Object (File)
Sun, Oct 27, 8:13 PM

Details

Summary

Depends on D5068

In response to this issue, we should implement the code that will delegate the work away from the reactors to separate threads.

Here, I implement using the worker in the bidi reactors.

Test Plan

Services build as usual.

Diff Detail

Repository
rCOMM Comm
Lint
No Lint Coverage
Unit
No Test Coverage

Event Timeline

tomek requested changes to this revision.Sep 9 2022, 5:31 AM
tomek added inline comments.
services/lib/src/server-base-reactors/ServerBidiReactorBase.h
100 ↗(On Diff #16393)

This API looks strange. I would expect that worker is an internal entity and we interact with pool instead. We submit a task to a pool which schedules it and assigns to a worker when it is available.

101–105 ↗(On Diff #16393)

This is really confusing. Why do we have two callbacks provided to the schedule method? Looking at D5068 it seems like the second argument is a callback which is invoked at the end of the task. Can't we include the logic for callback at the end of the task?

This revision now requires changes to proceed.Sep 9 2022, 5:31 AM

I understand your concerns. The point here is that I wanted to have a possibility to schedule a task and if this task doesn't fail, then schedule a callback that is called only afterward. If you take a look at the implementations, these tasks/callbacks differ so I think that introducing a separate layer of abstraction makes sense.

Also, in this case, we can easily replace the logic for the thread pooling because it is transparent from places where Worker::schedule is being invoked.

Not really qualified yet to give this a proper review.

tomek requested changes to this revision.Sep 15 2022, 7:53 AM
tomek added inline comments.
services/lib/src/server-base-reactors/ServerBidiReactorBase.h
101 ↗(On Diff #16655)

According to https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0018r3.html it isn't safe to schedule a lambda which captures this by reference. If we can guarantee that this exists when the task is executed, it should be fine. Please check, if according to the linked document, our approach is safe.

This revision now requires changes to proceed.Sep 15 2022, 7:53 AM
services/lib/src/server-base-reactors/ServerBidiReactorBase.h
101 ↗(On Diff #16655)

This is guaranteed. this lives as long as the grpc connection is on. It gets destructed when the grpc connection ends. And basically, these lambdas call the functions that do the grpc operations. Only with these operations, the connection may terminate. So, the only way to delete this is to call the functions that are inside of these lambdas. (e.g. Finish)

This revision is now accepted and ready to land.Sep 16 2022, 10:10 AM