Throttler / Timer ; use current scheduler by ivovandongen · Pull Request #9611 · mapbox/mapbox-gl-native

ivovandongen · 2017-07-25T15:51:05Z

This PR builds on top of #9728 to make the Timers and throttlers use the current scheduler.

ivovandongen · 2017-07-27T15:42:58Z

@jfirebaugh I've started on adding delayed scheduling to the Scheduler. I've borrowed some patterns from Akka, but would love your feedback.

The idea is to schedule a message for later delivery on the given mailbox with:

virtual std::unique_ptr<Scheduled> schedule(Duration timeout, std::weak_ptr<Mailbox>, std::unique_ptr<Message>) = 0;

Since we need the object to make the Message this needs to go through either Actor or ActorRef:

std::unique_ptr<Scheduler::Scheduled> invokeDelayed(Duration timeout, Fn fn, Args&&... args, Scheduler& scheduler = *Scheduler::GetCurrent()) {

The Throttler is already using this.

The actual implementations for Scheduler would use different strategies for the timers. On the Runloop backed schedulers, we can use the existing Timer/Alarm classes. The thread pool would have a dedicated thread/threads for controlling the timers.

The scheduler returns a handle in the form of Scheduler::Scheduled that is analogous to the Cancelable on Akka, but also has ownership of the timer so one needs to keep it alive to keep the timer alive.

The Throttler now is a proxy for the ActorRef passed in and has a similar interface to ActorRef. The throttling is now ActorRef-wide, not per method.

Note: ThreadBasedTimer is a quick implementation to try things out, it's not very fast and I want to replace it with a Runloop backed one for most, but couldn't quite get it compiling right away.

ivovandongen · 2017-07-31T13:39:35Z

I've added two implementations for the timers:

A actor based one, based on @tmpsantos' work that uses a dedicated timer thread for the default ThreadPool: ThreadPoolScheduledTask
One that uses timers directly for the different Runloop implementations: RunLoopScheduledTask

The only RunLoop that doesn't support scheduling a delayed message is the NodeThreadPool. But, no timers are used there anyway.

This should be sufficient to continue work on #9576

ivovandongen · 2017-07-31T15:21:30Z

Trying out this branch, I noticed that I overlooked one issue with the way I use timers now. They still get the current Runloop instead of the runloop I intended them to be scheduled on.

kkaefer · 2017-07-31T14:22:14Z

    }
 }

-template class ThreadLocal<RunLoop>;


Removes the template instantiation, but the corresponding code is still there?

The code that used the template was in RunLoop, but has since been removed in favor of a ThreadLocal on Scheduler

kkaefer · 2017-07-31T14:25:07Z

+
+    // Set/Get the current Scheduler for this thread
+    static Scheduler* GetCurrent();
+    static void SetCurrent(Scheduler*);


Can we make this one protected so that only subclasses can set the current scheduler?

kkaefer · 2017-07-31T15:00:00Z

+    // Invoke twice in a row
+    throttler.invoke(&Test::doIt);
+    throttler.invoke(&Test::doIt);
+    loop.runOnce(); // Let the timer do it's thing


typo -> "its" ;)

kkaefer · 2017-07-31T15:04:21Z

+        virtual void cancel() = 0;
+
+        virtual bool isFinished() = 0;
+    };


We already have an almost identical class called WorkTask (and a few others that are similar). Maybe we can reuse/unify those?

AsyncRequest is the most generic interface. It's an interface for objects that cancel themselves when they go out of scope.

I've made it extend from AsyncRequest. However, we need a way to see if the task is completed (for Throttler) that's why I made it a new class with a isFinished() method.

kkaefer · 2017-07-31T15:08:42Z

+class ThreadPoolScheduledTask : public Scheduler::Scheduled {
+public:
+    // The Impl is scheduled on the Runloop of the timer thread
+    // and uses it to start a timer. When the timer endsm it calls


kkaefer · 2017-07-31T15:17:23Z

-
-                delegate.invoke(fn, std::forward<Args>(args)...);
-            });
+            pendingInvocation = delegate.invokeDelayed(timeToNextInvocation, fn, std::forward<Args>(args)...);


std::move(fn)

kkaefer · 2017-07-31T15:19:05Z

+    }
+private:
+    std::shared_ptr<Fn> fn;
+};


We have the combination of WorkRequest/WorkTask which is very similar to this. Can this be reused here?

kkaefer · 2017-07-31T15:20:51Z

+        wrap(
+        [&, weakMailbox = std::move(weakMailbox), message = std::move(message)] () mutable {
+            if (auto mailbox = weakMailbox.lock()) {
+                mailbox->push(std::forward<std::unique_ptr<Message>>(message));


Why forward instead of move?

Since the lambda is mutable, this should also protect against double-invocation when message is empty.

kkaefer · 2017-07-31T15:21:37Z

+};
+
+} // namespace util
+} // namespace mbgl


Missing newline

jfirebaugh

Can you submit the RunLoop::Get → Scheduler::GetCurrent changes as a separate PR? I'm 👍 on those ones but much less certain of the throttler / timer changes. Maybe we should just rip out the throttler dependency and rethink it entirely.

jfirebaugh · 2017-07-31T22:09:17Z


 RunLoop* RunLoop::Get() {
-    return current.get();
+    assert(static_cast<RunLoop*>(Scheduler::GetCurrent()));


Check the type too: assert(dynamic_cast<RunLoop*>(Scheduler::GetCurrent())) (x4).

jfirebaugh · 2017-07-31T22:09:45Z

@@ -4,6 +4,7 @@
 #include <mbgl/util/thread_local.hpp>


No longer used (x2).

jfirebaugh · 2017-07-31T22:26:26Z

+        virtual void cancel() = 0;
+
+        virtual bool isFinished() = 0;
+    };


AsyncRequest is the most generic interface. It's an interface for objects that cancel themselves when they go out of scope.

jfirebaugh · 2017-07-31T22:27:08Z

        push(task);
        return std::make_unique<WorkRequest>(task);
    }
+


Nit: excess whitespace here.

jfirebaugh · 2017-07-31T22:47:43Z

+class Message;

 /*
    A `Scheduler` is responsible for coordinating the processing of messages by


Update the references to RunLoop in this comment.

…e static library

- This doesn't work for asynchronous rendering - see #9611

ivovandongen · 2017-09-01T14:36:27Z

As discussed in chat; closing this and just removing the throttling all together.

- This doesn't work for asynchronous rendering - see #9611

ivovandongen self-assigned this Jul 25, 2017

ivovandongen force-pushed the ivd-current-scheduler branch 4 times, most recently from 63f1b25 to d0f27bf Compare July 27, 2017 15:30

ivovandongen force-pushed the ivd-current-scheduler branch 5 times, most recently from 0431ca6 to 4ddc3ca Compare July 31, 2017 13:31

ivovandongen requested review from jfirebaugh and tmpsantos July 31, 2017 13:34

ivovandongen removed the ⚠️ DO NOT MERGE Work in progress, proof of concept, or on hold label Jul 31, 2017

kkaefer reviewed Jul 31, 2017

View reviewed changes

jfirebaugh reviewed Jul 31, 2017

View reviewed changes

ivovandongen force-pushed the ivd-current-scheduler branch 3 times, most recently from 9df6cda to b0c3a82 Compare August 7, 2017 17:32

ivovandongen added 5 commits August 9, 2017 12:13

[default] add delayed message delivery on thread pool

38b6bfa

[core] add invokeDelayed to actor ref

b1b42cf

[core] Throttler - use invokeDelayed instead of timer

cd2b29a

[node] include platform files directly instead of using a intermediat…

66ce50d

…e static library

[all] add delayed message delivery on run loop

617759d

ivovandongen force-pushed the ivd-current-scheduler branch from 278526b to 617759d Compare August 9, 2017 09:16

ivovandongen added a commit that referenced this pull request Aug 9, 2017

[core] STOPGAP - remove throttling from geometry tile

c809e44

- This doesn't work for asynchronous rendering - see #9611

ivovandongen mentioned this pull request Aug 10, 2017

Android snapshotter #9748

Merged

6 tasks done

ivovandongen added a commit that referenced this pull request Aug 10, 2017

[core] STOPGAP - remove throttling from geometry tile

c788f0d

- This doesn't work for asynchronous rendering - see #9611

ivovandongen added a commit that referenced this pull request Aug 18, 2017

[core] STOPGAP - remove throttling from geometry tile

f73ff7a

- This doesn't work for asynchronous rendering - see #9611

ivovandongen closed this Sep 1, 2017

ivovandongen added a commit that referenced this pull request Sep 4, 2017

[core] remove throttling from geometry tile

7e11617

- This doesn't work for asynchronous rendering - see #9611

ivovandongen added a commit that referenced this pull request Sep 8, 2017

[core] remove throttling from geometry tile

bdfc761

- This doesn't work for asynchronous rendering - see #9611

ivovandongen added a commit that referenced this pull request Sep 12, 2017

[core] remove throttling from geometry tile

2c02c69

- This doesn't work for asynchronous rendering - see #9611

ivovandongen added a commit that referenced this pull request Sep 13, 2017

[core] remove throttling from geometry tile

931a899

- This doesn't work for asynchronous rendering - see #9611

ivovandongen added a commit that referenced this pull request Sep 15, 2017

[core] remove throttling from geometry tile

358a66b

- This doesn't work for asynchronous rendering - see #9611

ivovandongen added a commit that referenced this pull request Sep 18, 2017

[core] remove throttling from geometry tile

1132f18

- This doesn't work for asynchronous rendering - see #9611

ivovandongen added a commit that referenced this pull request Sep 19, 2017

[core] remove throttling from geometry tile

bcb14df

- This doesn't work for asynchronous rendering - see #9611

ivovandongen added a commit that referenced this pull request Sep 21, 2017

[core] remove throttling from geometry tile

8bb7ce6

- This doesn't work for asynchronous rendering - see #9611

ivovandongen added a commit that referenced this pull request Sep 22, 2017

[core] remove throttling from geometry tile

9168f8a

- This doesn't work for asynchronous rendering - see #9611

ivovandongen added a commit that referenced this pull request Sep 22, 2017

[core] remove throttling from geometry tile

e18da38

- This doesn't work for asynchronous rendering - see #9611

jfirebaugh deleted the ivd-current-scheduler branch October 24, 2017 20:40

Conversation

ivovandongen commented Jul 25, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ivovandongen commented Jul 27, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ivovandongen commented Jul 31, 2017

Uh oh!

ivovandongen commented Jul 31, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jfirebaugh left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ivovandongen commented Sep 1, 2017

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

ivovandongen commented Jul 25, 2017 •

edited

Loading

ivovandongen commented Jul 27, 2017 •

edited

Loading