message_loop.h 文件参考

Single-threaded task dispatcher with three queue backends, timers and scheduling envelopes. 更多...

#include <atomic>
#include <functional>
#include <future>
#include <limits>
#include <memory>
#include <mutex>
#include <string>
#include <tuple>
#include <type_traits>
#include <utility>
#include "./functional.h"
#include "./memory_resource.h"
#include "./schedule.h"
#include "./task_handle.h"
#include "./timer.h"

message_loop.h 的引用(Include)关系图:

此图展示该文件直接或间接的被哪些文件引用了:

浏览源代码.

类
class	vlink::MessageLoop
	Serial task dispatcher with selectable queue backend and bounded timer registry. 更多...
struct	vlink::MessageLoop::AliveState
	Lifetime gate shared with cross-thread observers. 更多...

命名空间
	vlink

详细描述

Single-threaded task dispatcher with three queue backends, timers and scheduling envelopes.

MessageLoop is VLink's serial task executor. It owns a bounded task queue and the timer registry attached to it; tasks posted from any thread execute in order on the loop's own thread. Three queue implementations are available; the choice is fixed at construction.

Dispatch mode table

Queue type	Backend	Capacity	Properties
kNormalType	mutex + std::deque (or pmr)	10000	FIFO, lock-based (default)
kLockfreeType	MpmcQueue<Callback>	10000	Lock-free MPMC
kPriorityType	Two pmr priority queues	10000	Droppable / protected split

Back-pressure on a full queue is selected by Strategy: kOptimizationStrategy retries up to ten times with 1 ms sleeps before dropping an eligible task, kPopStrategy drops immediately, kBlockStrategy retries indefinitely. Idle dispatch is always condition-variable driven and independent of Strategy.

Lifecycle diagram

*                          run() / async_run()
*   +--------+                        |
*   |  idle  | ----- async_run -----> | spawn thread
*   +--------+                        |
*      ^                              v
*      |                          +--------+
*      |   wait_for_quit          | active |  <-- post_task / exec_task
*      |       (joined)           +--------+
*      |                              |
*      |                       quit() | quit(true)
*      |                              v
*      |                          +--------+
*      +------- on_end() -------- | drain  | -- pending tasks --> dropped
*                                 +--------+
* 

Priority levels

Symbol	Value	Notes
kNoPriority	0	FIFO sentinel
kLowestPriority	1	Lowest real priority
kTimerPriority	50	Reserved for timer callbacks
kNormalPriority	100	Default user-task priority
kHighestPriority	65535	Highest priority

Example

vlink::MessageLoop loop;
loop.async_run();
 
loop.post_task([] { do_work(); });
loop.post_task_with_priority([] { urgent(); }, vlink::MessageLoop::kHighestPriority);
 
auto fut = loop.invoke_task([]() -> int { return compute(); });
int result = fut.get();   // wait from a different thread
 
loop.quit();
loop.wait_for_quit();

注解

Maximum queue depth is 10000 (kMaxTaskSize) and maximum timer count is 100 (kMaxTimerSize). Blocking .get() on the future returned by invoke_task from inside the loop thread deadlocks; always call it from another thread.