[*] Reglue async and loop together

This commit is contained in:
Reece Wilson 2022-03-10 15:35:01 +00:00
parent 216587195e
commit 03bb80239c
14 changed files with 338 additions and 152 deletions

View File

@ -7,8 +7,26 @@
***/
#pragma once
namespace Aurora::Loop
{
struct ILoopQueue;
}
namespace Aurora::Async
{
AUE_DEFINE(ERunMode,
(
eLowLatencyYield, // uses high perf cond var + yield + trylock
eLowLatencyFreqKernel, // uses high perf cond var + timeout(freqency)
eEfficient // delegates sleep to the kernel once kernel objects are scheduled
));
struct RunMode
{
ERunMode mode;
AuUInt16 freqMsTick;
};
struct IThreadPool
{
// Spawning
@ -65,6 +83,10 @@ namespace Aurora::Async
virtual void AssertWorker(WorkerId_t id) = 0;
// Async subsystem glue
virtual bool ScheduleLoopSource(const AuSPtr<Loop::ILoopSource> &loopSource, WorkerId_t workerId, AuUInt32 timeout, const AuConsumer<AuSPtr<Loop::ILoopSource>, bool> &callback) = 0;
virtual AuSPtr<Loop::ILoopQueue> ToKernelWorkQueue() = 0;
virtual AuSPtr<Loop::ILoopQueue> ToKernelWorkQueue(WorkerId_t workerId) = 0;
virtual void UpdateWorkMode(WorkerId_t workerId, RunMode mode) = 0;
virtual ERunMode GetCurrentThreadRunMode() = 0;
virtual ERunMode GetThreadRunMode(WorkerId_t workerId) = 0;
};
}

View File

@ -8,7 +8,7 @@
While you may drive low perf user facing apps and small services from this, other services should divide and conquer
* Allow the network subsystem to load balance sockets across a predefined amount of workers,
* Use semaphores instead of massive lengthy arrays of mutexes
* Use grouped long-polling LoopSources to convert kernel or waitable objects to a single time-polled loop source
* Use grouped long-polling LoopSources to convert kernel or waitable objects to a single time-polled loop source (freq based thread runners, alt to eEfficient)
***/
#pragma once

View File

@ -206,4 +206,4 @@ namespace Aurora::Memory
using MemoryViewStreamRead = MemoryViewStream<true>;
using MemoryViewStreamWrite = MemoryViewStream<false>;
}
}

View File

@ -23,7 +23,7 @@ the buildscripts into your applications build pipeline to get started.
- Basic cmdline parsing from any module
- Exit and fatal save condition callbacks
- IPC [WIP]
- Network
- Network [WIP]
- Random; secure and fast
- Hardware Info; memory and cpu info
- Software Stack Info
@ -345,7 +345,7 @@ UTF-8 and attempt to read a BOM to translate any other input to UTF-8.
### Paths
We assume all paths are messy. Incorrect splitters, double splitters, relative paths, and keywords are resolved internally.
No such URL or path builder, data structure to hold a tokenized representation, or similar concept exists in the codebase.
No URL or path builder, data structure to hold a tokenized URI expression, or similar concept exists in the codebase.
All string 'paths' are simply expanded, similar to MSCRT's `fullpath` or UNIX's `realpath`, at time of usage.
<br>
Path tokens include:<br>
@ -366,13 +366,9 @@ Path tokens include:<br>
### NIO
The networking stack supports a handful of architectural paradigms<br>
- block on write<br>
- delegate write to end of network frame on write<br>
- read with an all-or-nothing flag and an async flag<br>
- read with an asynchronous stream callback
- peaking<br>
- async read/write pump whenever and/or all
- Worker thread delegated resolve using system resolver
- Callback with fence-id based asynchronous write abstraction
- Loop Source support
### CIO
@ -468,10 +464,10 @@ Utility (third party)
provides production hardening <br>
Examples:<br>
-> Use embedded crypto libraries; libtomcrypt, libtommath<br>
-> While there are some bugs in libtomcrypt and others, none appear to <br>
->> While there are some bugs in libtomcrypt and others, none appear to <br>
cryptographically cripple the library. Could you do better?<br>
-> Use portable libraries like mbedtls, O(1) heap, mimalloc<br>
-> Writing a [D]TLS/allocator stack would take too much time<br>
-> Linking against external allocators, small cross-platform utilities, and <br>
->> Writing a [D]TLS/allocator stack would take too much time<br>
->> Linking against external allocators, small cross-platform utilities, and <br>
so on is probably fine <br>
-> Shim libcurl instead of inventing yet another http stack <br>

View File

@ -175,8 +175,28 @@ namespace Aurora::Async
ThreadPool::AssertWorker(id);
}
bool AsyncApp::ScheduleLoopSource(const AuSPtr<Loop::ILoopSource> &loopSource, WorkerId_t workerId, AuUInt32 timeout, const AuConsumer<AuSPtr<Loop::ILoopSource>, bool> &callback)
AuSPtr<Loop::ILoopQueue> AsyncApp::ToKernelWorkQueue()
{
return ThreadPool::ScheduleLoopSource(loopSource, workerId, timeout, callback);
return ThreadPool::ToKernelWorkQueue();
}
AuSPtr<Loop::ILoopQueue> AsyncApp::ToKernelWorkQueue(WorkerId_t workerId)
{
return ThreadPool::ToKernelWorkQueue(workerId);
}
void AsyncApp::UpdateWorkMode(WorkerId_t workerId, RunMode mode)
{
ThreadPool::UpdateWorkMode(workerId, mode);
}
ERunMode AsyncApp::GetCurrentThreadRunMode()
{
return ThreadPool::GetCurrentThreadRunMode();
}
ERunMode AsyncApp::GetThreadRunMode(WorkerId_t workerId)
{
return ThreadPool::GetThreadRunMode(workerId);
}
}

View File

@ -37,8 +37,12 @@ namespace Aurora::Async
void AddFeature(WorkerId_t id, AuSPtr<Threading::Threads::IThreadFeature> feature, bool async) override;
void AssertInThreadGroup(ThreadGroup_t group) override;
void AssertWorker(WorkerId_t id) override;
bool ScheduleLoopSource(const AuSPtr<Loop::ILoopSource> &loopSource, WorkerId_t workerId, AuUInt32 timeout, const AuConsumer<AuSPtr<Loop::ILoopSource>, bool> &callback) override;
AuSPtr<Loop::ILoopQueue> ToKernelWorkQueue() override;
AuSPtr<Loop::ILoopQueue> ToKernelWorkQueue(WorkerId_t workerId) override;
void UpdateWorkMode(WorkerId_t workerId, RunMode mode) override;
ERunMode GetCurrentThreadRunMode() override;
ERunMode GetThreadRunMode(WorkerId_t workerId) override;
// Main thread logic
void Start() override;

View File

@ -10,6 +10,7 @@
#include "ThreadPool.hpp"
#include "WorkItem.hpp"
#include "Schedular.hpp"
#include "ThreadWorkerQueueShim.hpp"
namespace Aurora::Async
{
@ -221,16 +222,64 @@ namespace Aurora::Async
auto state = GetThreadState();
bool success {};
auto runMode = GetCurrentThreadRunMode();
do
{
if (state->inLoopSourceMode)
auto asyncLoop = state->asyncLoop;
asyncLoop->OnFrame();
if (asyncLoop->GetSourceCount() > 1)
{
success = PollLoopSource(block);
bool bShouldTrySleepForKernel {};
if (runMode == ERunMode::eLowLatencyFreqKernel)
{
if (state->rateLimiter.CheckExchangePass())
{
bShouldTrySleepForKernel = asyncLoop->IsSignaled();
}
else
{
if (!PollInternal(false))
{
AuThreading::ContextYield();
}
else
{
success = true;
}
}
}
else if (runMode == ERunMode::eLowLatencyYield)
{
AuThreading::ContextYield();
block = false;
bShouldTrySleepForKernel = asyncLoop->IsSignaled();
}
else if (runMode == ERunMode::eEfficient)
{
bShouldTrySleepForKernel = block;
if (!block)
{
bShouldTrySleepForKernel = asyncLoop->IsSignaled();
}
}
if (bShouldTrySleepForKernel && asyncLoop->WaitAny(0))
{
PollInternal(block);
success = true;
}
else
{
success |= PollInternal(block);
}
}
else
{
success = PollInternal(block);
success |= state->inLoopSourceMode;
}
} while (success);
@ -251,7 +300,7 @@ namespace Aurora::Async
// TODO: reimplement this
// this is stupid and gross
if (group->workQueue.size() > 2)
if (group->workQueue.size() > group->workers.size()*3)
{
if (!group->sorted)
{
@ -374,6 +423,8 @@ namespace Aurora::Async
if (itr->second->GetPrio() < 0.25)
{
group->sorted = false;
if (lowPrioCont) continue;
if (!lowPrioContCached)
@ -430,102 +481,6 @@ namespace Aurora::Async
return true;
}
bool ThreadPool::PollLoopSource(bool block)
{
auto state = GetThreadState();
auto group = state->parent.lock();
//state->pendingWorkItems.clear();
auto magic = CtxPollPush();
bool retValue {};
// TODO (reece): This function isn't very efficient
{
AU_LOCK_GUARD(group->cvWorkMutex);
AuList<AsyncAppWaitSourceRequest> curLoopReq = state->loopSources;
AuList<AuSPtr<Loop::ILoopSource>> curLoopSources;
auto lenLoopReqs = curLoopReq.size();
curLoopSources.resize(lenLoopReqs + 1);
for (auto i = 0; i < lenLoopReqs; i++)
{
curLoopSources[i] = curLoopReq[i].loopSource;
}
curLoopSources[lenLoopReqs] = group->eventLs;
AuList<AuSPtr<Loop::ILoopSource>> nextLoopSources;
if (block)
{
// TODO (reece): work on async epoll like abstraction
nextLoopSources = Loop::WaitMultipleOrObjects(curLoopSources, 0);
}
else
{
nextLoopSources.reserve(curLoopSources.size());
for (const auto &source : curLoopSources)
{
if (source->IsSignaled())
{
nextLoopSources.push_back(source);
}
}
}
auto time = Time::CurrentClockMS();
state->loopSources.clear();
state->loopSources.reserve(curLoopReq.size());
if (AuExists(nextLoopSources, group->eventLs))
{
PollInternal(false);
}
for (const auto &request : curLoopReq)
{
bool remove {};
bool removeType {};
if (AuExists(nextLoopSources, request.loopSource))
{
remove = true;
removeType = true;
}
else
{
if (request.requestedOffset)
{
if (request.endTime < time)
{
remove = true;
removeType = false;
}
}
}
if (!remove)
{
state->loopSources.push_back(request);
}
else
{
request.callback(request.loopSource, removeType);
retValue |= removeType;
}
}
state->inLoopSourceMode = state->loopSources.size();
}
return retValue;
}
void ThreadPool::Shutdown()
{
// Nested shutdowns can happen; prevent a write lock
@ -755,38 +710,60 @@ namespace Aurora::Async
SysAssert(static_cast<WorkerId_t>(tlsWorkerId) == id);
}
bool ThreadPool::ScheduleLoopSource(const AuSPtr<Loop::ILoopSource> &loopSource, WorkerId_t workerId, AuUInt32 timeout, const AuConsumer<AuSPtr<Loop::ILoopSource>, bool> &callback)
AuSPtr<Loop::ILoopQueue> ThreadPool::ToKernelWorkQueue()
{
auto thread = this->GetThreadHandle(workerId);
if (!thread)
return this->GetThreadState()->asyncLoop;
}
AuSPtr<Loop::ILoopQueue> ThreadPool::ToKernelWorkQueue(WorkerId_t workerId)
{
auto worker = this->GetThreadHandle(workerId);
if (!worker)
{
return false;
SysPushErrorGen("Couldn't find requested worker");
return {};
}
auto group = thread->parent.lock();
return worker->asyncLoop;
}
void ThreadPool::UpdateWorkMode(WorkerId_t workerId, RunMode mode)
{
auto states = this->GetThreadHandles(workerId);
if (!states.size())
{
AU_LOCK_GUARD(group->cvWorkMutex);
AsyncAppWaitSourceRequest req {};
req.startTime = Time::CurrentClockMS();
if (timeout)
{
req.requestedOffset = timeout;
req.endTime = req.startTime + timeout;
}
req.loopSource = loopSource;
req.callback = callback;
if (!AuTryInsert(thread->loopSources, req))
{
return false;
}
thread->inLoopSourceMode = thread->loopSources.size();
SysPushErrorGen("Couldn't find requested worker");
return;
}
return true;
for (const auto &state : states)
{
state->runMode = mode.mode;
if (mode.freqMsTick)
{
state->rateLimiter.SetNextStep(mode.freqMsTick * 1'000'000);
}
}
}
ERunMode ThreadPool::GetCurrentThreadRunMode()
{
auto state = this->GetThreadState();
if (!state)
{
return ERunMode::eEfficient;
}
return state->runMode;
}
ERunMode ThreadPool::GetThreadRunMode(WorkerId_t workerId)
{
auto worker = this->GetThreadHandle(workerId);
if (!worker)
{
SysPushErrorGen("Couldn't find requested worker");
return {};
}
return worker->runMode;
}
// Unimplemented fiber hooks, 'twas used for science
@ -869,7 +846,28 @@ namespace Aurora::Async
threadState->running = AuThreadPrimitives::EventUnique(true, false, true);
threadState->syncSema = AuThreadPrimitives::SemaphoreUnique(0);
threadState->id = workerId;
//threadState->eventDriven = runner;
threadState->asyncLoop = AuMakeShared<AsyncLoop>();
threadState->rateLimiter.SetNextStep(1'000'000); // 1MS in nanoseconds
if (!threadState->asyncLoop)
{
SysPushErrorMem();
return {};
}
if (!threadState->syncSema)
{
SysPushErrorMem();
return {};
}
if (!threadState->syncSema)
{
SysPushErrorMem();
return {};
}
threadState->asyncLoop->SourceAdd(group->eventLs);
if (!create)
{
@ -1064,6 +1062,38 @@ namespace Aurora::Async
return *ret;
}
AuList<AuSPtr<ThreadState>> ThreadPool::GetThreadHandles(WorkerId_t id)
{
AU_LOCK_GUARD(this->rwlock_->AsReadable());
auto group = GetGroup(id.first);
if (!group)
{
return {};
}
AuList<AuSPtr<ThreadState>> ret;
if (id.second != Async::kThreadIdAny)
{
AuSPtr<ThreadState> *ptr;
if (!AuTryFind(group->workers, id.second, ptr))
{
return {};
}
ret.push_back(*ptr);
}
else
{
for (const auto &[key, value] : group->workers)
{
ret.push_back(value);
}
}
return ret;
}
AUKN_SYM AuSPtr<IThreadPool> NewThreadPool()
{
// apps that don't require async shouldn't be burdened with the overhead of this litl spiner

View File

@ -69,7 +69,13 @@ namespace Aurora::Async
virtual void AssertInThreadGroup(ThreadGroup_t group) override;
virtual void AssertWorker(WorkerId_t id) override;
virtual bool ScheduleLoopSource(const AuSPtr<Loop::ILoopSource> &loopSource, WorkerId_t workerId, AuUInt32 timeout, const AuConsumer<AuSPtr<Loop::ILoopSource>, bool> &callback) override;
virtual AuSPtr<Loop::ILoopQueue> ToKernelWorkQueue() override;
virtual AuSPtr<Loop::ILoopQueue> ToKernelWorkQueue(WorkerId_t workerId) override;
virtual void UpdateWorkMode(WorkerId_t workerId, RunMode mode) override;
virtual ERunMode GetCurrentThreadRunMode() override;
virtual ERunMode GetThreadRunMode(WorkerId_t workerId) override;
//virtual bool ScheduleLoopSource(const AuSPtr<Loop::ILoopSource> &loopSource, WorkerId_t workerId, AuUInt32 timeout, const AuConsumer<AuSPtr<Loop::ILoopSource>, bool> &callback) override;
// Internal API
@ -77,7 +83,6 @@ namespace Aurora::Async
bool InternalRunOne(bool block);
bool PollInternal(bool block);
bool PollLoopSource(bool block);
size_t GetThreadWorkersCount(ThreadGroup_t group);
@ -116,7 +121,8 @@ namespace Aurora::Async
AuSPtr<GroupState> GetGroup(ThreadGroup_t type);
AuSPtr<ThreadState> GetThreadState();
AuSPtr<ThreadState> GetThreadHandle(WorkerId_t id);
AuList<AuSPtr<ThreadState>> GetThreadHandles(WorkerId_t id);
using ThreadDb_t = AuBST<ThreadGroup_t, AuSPtr<GroupState>>;
ThreadDb_t threads_;

View File

@ -24,6 +24,7 @@ namespace Aurora::Async
};
struct GroupState;
struct AsyncLoop;
struct ThreadState
{
@ -36,11 +37,13 @@ namespace Aurora::Async
AuList<AuSPtr<AuThreads::IThreadFeature>> features;
bool rejecting {};
bool exiting {};
bool inLoopSourceMode {};
bool shuttingdown {};
AuThreadPrimitives::EventUnique_t running;
//bool running;
AuList<AsyncAppWaitSourceRequest> loopSources;
AuList<WorkEntry_t> pendingWorkItems;
AuSPtr<AsyncLoop> asyncLoop;
Utility::RateLimiter rateLimiter;
ERunMode runMode;
};
}

View File

@ -0,0 +1,60 @@
/***
Copyright (C) 2022 J Reece Wilson (a/k/a "Reece"). All rights reserved.
File: ThreadWorkerQueueShim.cpp
Date: 2022-3-9
Author: Reece
***/
#include <Source/RuntimeInternal.hpp>
#include "Async.hpp"
#include "ThreadWorkerQueueShim.hpp"
namespace Aurora::Async
{
void AsyncLoop::OnFrame()
{
if (this->commitPending_)
{
if (LoopQueue::Commit())
{
this->commitPending_ = false;
}
}
}
bool AsyncLoop::AddCallback(const AuSPtr<Loop::ILoopSource> &source, const AuSPtr<Loop::ILoopSourceSubscriber> &subscriber)
{
auto ret = LoopQueue::AddCallback(source, subscriber);
if (ret)
{
this->commitPending_ = true;
}
return ret;
}
bool AsyncLoop::AddCallbackEx(const AuSPtr<Loop::ILoopSource> &source, const AuSPtr<Loop::ILoopSourceSubscriberEx> &subscriber)
{
auto ret = LoopQueue::AddCallbackEx(source, subscriber);
if (ret)
{
this->commitPending_ = true;
}
return ret;
}
bool AsyncLoop::AddCallback(const AuSPtr<Loop::ILoopSourceSubscriber> &subscriber)
{
auto ret = LoopQueue::AddCallback(subscriber);
if (ret)
{
this->commitPending_ = true;
}
return ret;
}
bool AsyncLoop::Commit()
{
this->commitPending_ = true;
return true;
}
}

View File

@ -0,0 +1,27 @@
/***
Copyright (C) 2022 J Reece Wilson (a/k/a "Reece"). All rights reserved.
File: ThreadWorkerQueueShim.hpp
Date: 2022-3-9
Author: Reece
***/
#pragma once
#include <Source/Loop/Loop.hpp>
#include <Source/Loop/LoopQueue.hpp>
namespace Aurora::Async
{
struct AsyncLoop : public Loop::LoopQueue
{
void OnFrame();
virtual bool AddCallback (const AuSPtr<Loop::ILoopSource> &source, const AuSPtr<Loop::ILoopSourceSubscriber> &subscriber) override;
virtual bool AddCallbackEx(const AuSPtr<Loop::ILoopSource> &source, const AuSPtr<Loop::ILoopSourceSubscriberEx> &subscriber) override;
virtual bool AddCallback (const AuSPtr<Loop::ILoopSourceSubscriber> &subscriber) override;
virtual bool Commit () override;
private:
bool commitPending_ {};
};
}

View File

@ -7,6 +7,8 @@
***/
#pragma once
#include "ILoopSourceEx.hpp"
namespace Aurora::Loop
{
struct LoopQueue : ILoopQueue
@ -20,14 +22,14 @@ namespace Aurora::Loop
AuUInt32 GetSourceCount() override;
bool AddCallback(const AuSPtr<ILoopSource> &source, const AuSPtr<ILoopSourceSubscriber> &subscriber) override;
bool AddCallbackEx(const AuSPtr<ILoopSource> &source, const AuSPtr<ILoopSourceSubscriberEx> &subscriber) override;
bool AddCallback(const AuSPtr<ILoopSourceSubscriber> &subscriber) override;
virtual bool AddCallback(const AuSPtr<ILoopSource> &source, const AuSPtr<ILoopSourceSubscriber> &subscriber) override;
virtual bool AddCallbackEx(const AuSPtr<ILoopSource> &source, const AuSPtr<ILoopSourceSubscriberEx> &subscriber) override;
virtual bool AddCallback(const AuSPtr<ILoopSourceSubscriber> &subscriber) override;
void ChugPathConfigure(AuUInt32 sectionTickTime, AuSInt sectionDequeCount) override;
void ChugHint(bool value) override;
bool Commit() override;
virtual bool Commit() override;
bool HasFinished() override;
bool IsSignaled() override;

14
Source/Loop/LoopQueue.hpp Normal file
View File

@ -0,0 +1,14 @@
/***
Copyright (C) 2022 J Reece Wilson (a/k/a "Reece"). All rights reserved.
File: LoopQueue.cpp
Date: 2022-3-9
Author: Reece
***/
#pragma once
#if defined(AURORA_IS_MODERNNT_DERIVED)
#include "LoopQueue.NT.hpp"
#elif defined(AURORA_PLATFORM_LINX)
#include "LoopQueue.Linux.hpp"
#endif

View File

@ -22,6 +22,8 @@
//_WIN32_WINNT=0x0601
#include <SDKDDKVer.h>
#include <winsock2.h>
#if defined(_AUHAS_ASIO)
#include <asio.hpp>
#endif