How React Scheduler works internally?

1. Why React Scheduler is needed.

Letā€™s start with following piece of code(source), weā€™ve already covered it in our very first episode of this series.

js
function workLoopSync() {
while (workInProgress !== null) {
performUnitOfWork(workInProgress);
}
}
js
function workLoopSync() {
while (workInProgress !== null) {
performUnitOfWork(workInProgress);
}
}

In a word, React internally works on each Fiber on the Fiber tree, workInProgress is to track current position, the traversal algorithm is already explained in my previous post.

workLoopSync() is very easy to understand, since it is synchronous, there is no interrupting of our work, so React just keep working in a while loop.

Things are different in concurrent mode(source).

function workLoopConcurrent() {
while (workInProgress !== null && !shouldYield()) {
performUnitOfWork(workInProgress);
}
}
function workLoopConcurrent() {
while (workInProgress !== null && !shouldYield()) {
performUnitOfWork(workInProgress);
}
}

In concurrent mode, tasks with higher priorities could interrupt tasks with lower priority, we need a way to interrupt the tasks and resume, that is what shouldYield() does the trick, but obviously there is more than that.

2. Letā€™s first start with some background knowledge

2.1 Event Loop

To be honest, I cannot explain this very well, I suggest you read the explanation from javascript.info or watch this nice video from Jake Archibald.

Simply put, JavaScript engine would do something as below

  1. grab the task(macro task) from the task queue and run it
  2. if there are microtasks scheduled, run them
  3. check if needs to render and do it
  4. repeat 1 if there are more task or wait for more task.

The loop is very self-explanatory, because there is actually sort of loop there.

2.2 setImmediate() to schedule a new task without blocking the rendering

To schedule some task without blocking rendering(the step 3 above), weā€™re already familiar with the trick of setTimeout(callback, 0), it schedules a new macro task.

There is an event better API setImmediate() but it is only available in IE and node.js.

It is better because setTimeout() actually hav minimum of about 4ms in nested calls, setImmediate() doesnā€™t have the delay.

OK, we are ready to touch the first piece of code in React Scheduler(source).

let schedulePerformWorkUntilDeadline;
if (typeof localSetImmediate === "function") {
// Node.js and old IE.
// There's a few reasons for why we prefer setImmediate.
//
// Unlike MessageChannel, it doesn't prevent a Node.js process from exiting.
// (Even though this is a DOM fork of the Scheduler, you could get here
// with a mix of Node.js 15+, which has a MessageChannel, and jsdom.)
// https://github.com/facebook/react/issues/20756
//
// But also, it runs earlier which is the semantic we want.
// If other browsers ever implement it, it's better to use it.
// Although both of these would be inferior to native scheduling.
schedulePerformWorkUntilDeadline = () => {
localSetImmediate(performWorkUntilDeadline);
};
} else if (typeof MessageChannel !== "undefined") {
// DOM and Worker environments.
// We prefer MessageChannel because of the 4ms setTimeout clamping.
const channel = new MessageChannel();
const port = channel.port2;
channel.port1.onmessage = performWorkUntilDeadline;
schedulePerformWorkUntilDeadline = () => {
port.postMessage(null);
};
} else {
// We should only fallback here in non-browser environments.
schedulePerformWorkUntilDeadline = () => {
localSetTimeout(performWorkUntilDeadline, 0);
};
}
let schedulePerformWorkUntilDeadline;
if (typeof localSetImmediate === "function") {
// Node.js and old IE.
// There's a few reasons for why we prefer setImmediate.
//
// Unlike MessageChannel, it doesn't prevent a Node.js process from exiting.
// (Even though this is a DOM fork of the Scheduler, you could get here
// with a mix of Node.js 15+, which has a MessageChannel, and jsdom.)
// https://github.com/facebook/react/issues/20756
//
// But also, it runs earlier which is the semantic we want.
// If other browsers ever implement it, it's better to use it.
// Although both of these would be inferior to native scheduling.
schedulePerformWorkUntilDeadline = () => {
localSetImmediate(performWorkUntilDeadline);
};
} else if (typeof MessageChannel !== "undefined") {
// DOM and Worker environments.
// We prefer MessageChannel because of the 4ms setTimeout clamping.
const channel = new MessageChannel();
const port = channel.port2;
channel.port1.onmessage = performWorkUntilDeadline;
schedulePerformWorkUntilDeadline = () => {
port.postMessage(null);
};
} else {
// We should only fallback here in non-browser environments.
schedulePerformWorkUntilDeadline = () => {
localSetTimeout(performWorkUntilDeadline, 0);
};
}

Here we see the 2 different fallbacks of setImmediate(), with MessageChannel and setTimeout.

2.3 Priority Queue

Priority Queue is a common data structure for scheduling. I suggest you try creating a priority queue in JavaScript by yourself.

It perfectly suits the needs in React. Since events of different priorities come in, we need to quickly find the one with highest priority to process.

React implements Priority Queue with min-heap, you can find the source code here.

3. Call stack of workLoopConcurrent

Now, letā€™s take a look at how workLoopConcurrent is called.

All the code is in ReactFiberWorkLoop.js, letā€™s break it down.

Weā€™ve met ensureRootIsScheduled() a lot of times, it is used in quite a few place. As the name implies, ensureRootIsScheduled() schedules a task for React to do the updates if there are any.

Notice that it doesnā€™t call performConcurrentWorkOnRoot() directly but treats it as a callback by scheduleCallback(priority, callback). scheduleCallback() is an api in Scheduler.

Weā€™ll dive into the scheduler soon, but for now, just keep in mind that Scheduler will run the task at the right time.

3.1 performConcurrentWorkOnRoot() returns a closure of itself if interrupted.

See that performConcurrentWorkOnRoot() returns differently based on the progress?

  1. If shouldYield() is true, workLoopConcurrent will break, which results in incomplete update(RootInComplete), performConcurrentWorkOnRoot() will return performConcurrentWorkOnRoot.bind(null, root). (code)
  2. If it is complete, then it returns null.

You might wonder if a task is interrupted by shouldYield(), how would it resume? Yes, this is the answer. Scheduler looks at the return value of task callback to see if there is continuation, the return value is kind of rescheduling. Weā€™ll cover this soon.

4. Scheduler

Finally, we are in the realm of Scheduler. Donā€™t feel overwhelmed, I was intimidated at the beginning but soon realized it is unnecessary.

Message Queue is is a way to hand out control, Scheduler does exactly like that.

scheduleCallback() mentioned above is unstable_scheduleCallback in Scheduler world.

4.1 scheduleCallback() - Scheduler schedules tasks by exipriationTime

In order for Scheduler to schedule tasks, it first need to store tasks with their priorities. This is done by Priority Queue which weā€™ve already covered as background knowledge.

It uses expirationTime to indicate priority. This is fair, sooner it expires, sooner we have to process it. Below is the code inside of scheduleCallback() where a task is created

var currentTime = getCurrentTime();
var startTime;
if (typeof options === "object" && options !== null) {
var delay = options.delay;
if (typeof delay === "number" && delay > 0) {
startTime = currentTime + delay;
} else {
startTime = currentTime;
}
} else {
startTime = currentTime;
}
var timeout;
switch (priorityLevel) {
case ImmediatePriority:
timeout = IMMEDIATE_PRIORITY_TIMEOUT;
break;
case UserBlockingPriority:
timeout = USER_BLOCKING_PRIORITY_TIMEOUT;
break;
case IdlePriority:
timeout = IDLE_PRIORITY_TIMEOUT;
break;
case LowPriority:
timeout = LOW_PRIORITY_TIMEOUT;
break;
case NormalPriority:
default:
timeout = NORMAL_PRIORITY_TIMEOUT;
break;
}
var expirationTime = startTime + timeout;
var newTask = {
id: taskIdCounter++,
callback,
priorityLevel,
startTime,
expirationTime,
sortIndex: -1,
};

A task is the unit of work that Scheduler handles

var currentTime = getCurrentTime();
var startTime;
if (typeof options === "object" && options !== null) {
var delay = options.delay;
if (typeof delay === "number" && delay > 0) {
startTime = currentTime + delay;
} else {
startTime = currentTime;
}
} else {
startTime = currentTime;
}
var timeout;
switch (priorityLevel) {
case ImmediatePriority:
timeout = IMMEDIATE_PRIORITY_TIMEOUT;
break;
case UserBlockingPriority:
timeout = USER_BLOCKING_PRIORITY_TIMEOUT;
break;
case IdlePriority:
timeout = IDLE_PRIORITY_TIMEOUT;
break;
case LowPriority:
timeout = LOW_PRIORITY_TIMEOUT;
break;
case NormalPriority:
default:
timeout = NORMAL_PRIORITY_TIMEOUT;
break;
}
var expirationTime = startTime + timeout;
var newTask = {
id: taskIdCounter++,
callback,
priorityLevel,
startTime,
expirationTime,
sortIndex: -1,
};

A task is the unit of work that Scheduler handles

The code is very straightforward, for each priority we have different timeout, they are defined here.

// Times out immediately
var IMMEDIATE_PRIORITY_TIMEOUT = -1;
// Eventually times out
var USER_BLOCKING_PRIORITY_TIMEOUT = 250;
var NORMAL_PRIORITY_TIMEOUT = 5000;

Default is 5 second timeout

var LOW_PRIORITY_TIMEOUT = 10000;
// Never times out
var IDLE_PRIORITY_TIMEOUT = maxSigned31BitInt;
// Times out immediately
var IMMEDIATE_PRIORITY_TIMEOUT = -1;
// Eventually times out
var USER_BLOCKING_PRIORITY_TIMEOUT = 250;
var NORMAL_PRIORITY_TIMEOUT = 5000;

Default is 5 second timeout

var LOW_PRIORITY_TIMEOUT = 10000;
// Never times out
var IDLE_PRIORITY_TIMEOUT = maxSigned31BitInt;

So default it has 5 seconds timeout, and for user blocking it has 250ms. Weā€™ll soon see some examples of these priorities.

Task is created, now it is time to put it in the Priority Queue.

if (startTime > currentTime) {
// This is a delayed task.
newTask.sortIndex = startTime;
push(timerQueue, newTask);
if (peek(taskQueue) === null && newTask === peek(timerQueue)) {
// All tasks are delayed, and this is the task with the earliest delay.
if (isHostTimeoutScheduled) {
// Cancel an existing timeout.
cancelHostTimeout();
} else {
isHostTimeoutScheduled = true;
}
// Schedule a timeout.
requestHostTimeout(handleTimeout, startTime - currentTime);
}
} else {
newTask.sortIndex = expirationTime;
push(taskQueue, newTask);
// Schedule a host callback, if needed. If we're already performing work,
// wait until the next time we yield.
if (!isHostCallbackScheduled && !isPerformingWork) {
isHostCallbackScheduled = true;
requestHostCallback(flushWork);
}
}
if (startTime > currentTime) {
// This is a delayed task.
newTask.sortIndex = startTime;
push(timerQueue, newTask);
if (peek(taskQueue) === null && newTask === peek(timerQueue)) {
// All tasks are delayed, and this is the task with the earliest delay.
if (isHostTimeoutScheduled) {
// Cancel an existing timeout.
cancelHostTimeout();
} else {
isHostTimeoutScheduled = true;
}
// Schedule a timeout.
requestHostTimeout(handleTimeout, startTime - currentTime);
}
} else {
newTask.sortIndex = expirationTime;
push(taskQueue, newTask);
// Schedule a host callback, if needed. If we're already performing work,
// wait until the next time we yield.
if (!isHostCallbackScheduled && !isPerformingWork) {
isHostCallbackScheduled = true;
requestHostCallback(flushWork);
}
}

Oh right, when we schedule a task, it could have a delay option like setTimeout() does. Letā€™s keep it aside and come back later.

Just focus on the else branch. We can see two important calls

  1. push(taskQueue, newTask) - add the task into the queue, this is just the priority queue API, Iā€™ll just skip.
  2. requestHostCallback(flushWork) - process them!

requestHostCallback(flushWork) is necessary, because Scheduler is host agnostic, it should be just some independent black box which could be run on any host, so it needs to be requested.

4.2 requestHostCallback()

function requestHostCallback(callback) {
scheduledHostCallback = callback;
if (!isMessageLoopRunning) {
isMessageLoopRunning = true;
schedulePerformWorkUntilDeadline();
}
}
const performWorkUntilDeadline = () => {
if (scheduledHostCallback !== null) {
const currentTime = getCurrentTime();
// Keep track of the start time so we can measure how long the main thread
// has been blocked.
startTime = currentTime;
const hasTimeRemaining = true;
// If a scheduler task throws, exit the current browser task so the
// error can be observed.
//
// Intentionally not using a try-catch, since that makes some debugging
// techniques harder. Instead, if `scheduledHostCallback` errors, then
// `hasMoreWork` will remain true, and we'll continue the work loop.
let hasMoreWork = true;
try {
hasMoreWork = scheduledHostCallback(hasTimeRemaining, currentTime);
} finally {
if (hasMoreWork) {
// If there's more work, schedule the next message event at the end
// of the preceding one.
schedulePerformWorkUntilDeadline();

we can see that Scheduler keeps processing tasks in the queue by scheduling

Here it gives browser a chance to paint

} else {
isMessageLoopRunning = false;
scheduledHostCallback = null;
}
}
} else {
isMessageLoopRunning = false;
}
// Yielding to the browser will give it a chance to paint, so we can
// reset this.
needsPaint = false;
};
function requestHostCallback(callback) {
scheduledHostCallback = callback;
if (!isMessageLoopRunning) {
isMessageLoopRunning = true;
schedulePerformWorkUntilDeadline();
}
}
const performWorkUntilDeadline = () => {
if (scheduledHostCallback !== null) {
const currentTime = getCurrentTime();
// Keep track of the start time so we can measure how long the main thread
// has been blocked.
startTime = currentTime;
const hasTimeRemaining = true;
// If a scheduler task throws, exit the current browser task so the
// error can be observed.
//
// Intentionally not using a try-catch, since that makes some debugging
// techniques harder. Instead, if `scheduledHostCallback` errors, then
// `hasMoreWork` will remain true, and we'll continue the work loop.
let hasMoreWork = true;
try {
hasMoreWork = scheduledHostCallback(hasTimeRemaining, currentTime);
} finally {
if (hasMoreWork) {
// If there's more work, schedule the next message event at the end
// of the preceding one.
schedulePerformWorkUntilDeadline();

we can see that Scheduler keeps processing tasks in the queue by scheduling

Here it gives browser a chance to paint

} else {
isMessageLoopRunning = false;
scheduledHostCallback = null;
}
}
} else {
isMessageLoopRunning = false;
}
// Yielding to the browser will give it a chance to paint, so we can
// reset this.
needsPaint = false;
};

As mentioned in 2.2 schedulePerformWorkUntilDeadline() is just a wrapper on performWorkUntilDeadline().

scheduledHostCallback is set in requestHostCallback() and called right away in performWorkUntilDeadline(), this is to give main thread a chance to render because of async nature.

Ignoring some details, here is the most important line.

js
hasMoreWork = scheduledHostCallback(hasTimeRemaining, currentTime)
js
hasMoreWork = scheduledHostCallback(hasTimeRemaining, currentTime)

This means flushWork() will be called with (true, currentTime).

I donā€™t know why it is hardcoded as true here. maybe because of refactoring mistakes.

4.3 flushWork()

try {
// No catch in prod code path.
return workLoop(hasTimeRemaining, initialTime);
} finally {
//
}
try {
// No catch in prod code path.
return workLoop(hasTimeRemaining, initialTime);
} finally {
//
}

flushWork just wraps up workLoop()

4.4 workLoop() - the core of Scheduler

As workLoopConcurrent() in reconciliation, workLoop() is the core of Scheduler. They have similar name because they have similar process.

js
if (
currentTask.expirationTime > currentTime &&
// ( )
(!hasTimeRemaining || shouldYieldToHost())
) {
// This currentTask hasn't expired, and we've reached the deadline.
break;
}
js
if (
currentTask.expirationTime > currentTime &&
// ( )
(!hasTimeRemaining || shouldYieldToHost())
) {
// This currentTask hasn't expired, and we've reached the deadline.
break;
}

Just as workLoopConcurrent(), shouldYieldToHost() is checked here. Weā€™ll cover it later.

const callback = currentTask.callback;
if (typeof callback === "function") {
currentTask.callback = null;
currentPriorityLevel = currentTask.priorityLevel;
const didUserCallbackTimeout = currentTask.expirationTime <= currentTime;
const continuationCallback = callback(didUserCallbackTimeout);
currentTime = getCurrentTime();
if (typeof continuationCallback === "function") {

Here we see why the return value of tasks matter

Notice that under this branch, the task is NOT popped!

currentTask.callback = continuationCallback;
} else {
if (currentTask === peek(taskQueue)) {
pop(taskQueue);
}
}
advanceTimers(currentTime);
} else {
pop(taskQueue);
}
const callback = currentTask.callback;
if (typeof callback === "function") {
currentTask.callback = null;
currentPriorityLevel = currentTask.priorityLevel;
const didUserCallbackTimeout = currentTask.expirationTime <= currentTime;
const continuationCallback = callback(didUserCallbackTimeout);
currentTime = getCurrentTime();
if (typeof continuationCallback === "function") {

Here we see why the return value of tasks matter

Notice that under this branch, the task is NOT popped!

currentTask.callback = continuationCallback;
} else {
if (currentTask === peek(taskQueue)) {
pop(taskQueue);
}
}
advanceTimers(currentTime);
} else {
pop(taskQueue);
}

Letā€™s break it down.

currentTask.callback, which is actually performConcurrentWorkOnRoot() in this case.

js
const didUserCallbackTimeout = currentTask.expirationTime <= currentTime;
const continuationCallback = callback(didUserCallbackTimeout);
js
const didUserCallbackTimeout = currentTask.expirationTime <= currentTime;
const continuationCallback = callback(didUserCallbackTimeout);

It is called with a flag to indicate if it is expired or not.

performConcurrentWorkOnRoot() will fall back on sync mode if timeout (code). This means from now on there should not be any interruption.

js
const shouldTimeSlice =
!includesBlockingLane(root, lanes) &&
!includesExpiredLane(root, lanes) &&
(disableSchedulerTimeoutInWorkLoop || !didTimeout);
let exitStatus = shouldTimeSlice
? renderRootConcurrent(root, lanes)
: renderRootSync(root, lanes);
js
const shouldTimeSlice =
!includesBlockingLane(root, lanes) &&
!includesExpiredLane(root, lanes) &&
(disableSchedulerTimeoutInWorkLoop || !didTimeout);
let exitStatus = shouldTimeSlice
? renderRootConcurrent(root, lanes)
: renderRootSync(root, lanes);

Ok, back to workLoop()

js
if (typeof continuationCallback === "function") {
currentTask.callback = continuationCallback;
} else {
if (currentTask === peek(taskQueue)) {
pop(taskQueue);
}
}
js
if (typeof continuationCallback === "function") {
currentTask.callback = continuationCallback;
} else {
if (currentTask === peek(taskQueue)) {
pop(taskQueue);
}
}

Important here, we can see the the task is only popped when the return value of callback is not function. If it is function, it will just update the callback of the task, since it is not popped, next tick of workLoop() would result in the same task again.

This means if the return value of this callback is a function, then this task is not done, we should work on it again.

js
advanceTimers(currentTime)
js
advanceTimers(currentTime)

This is for delayed tasks, weā€™ll come back later.

4.5 how shouldYield() work?

source

js
function shouldYieldToHost() {
const timeElapsed = getCurrentTime() - startTime;
if (timeElapsed < frameInterval) {
// The main thread has only been blocked for a really short amount of time;
// smaller than a single frame. Don't yield yet.
return false;
}
// The main thread has been blocked for a non-negligible amount of time. We
// may want to yield control of the main thread, so the browser can perform
// high priority tasks. The main ones are painting and user input. If there's
// a pending paint or a pending input, then we should yield. But if there's
// neither, then we can yield less often while remaining responsive. We'll
// eventually yield regardless, since there could be a pending paint that
// wasn't accompanied by a call to `requestPaint`, or other main thread tasks
// like network events.
if (enableIsInputPending) {
if (needsPaint) {
// There's a pending paint (signaled by `requestPaint`). Yield now.
return true;
}
if (timeElapsed < continuousInputInterval) {
// We haven't blocked the thread for that long. Only yield if there's a
// pending discrete input (e.g. click). It's OK if there's pending
// continuous input (e.g. mouseover).
if (isInputPending !== null) {
return isInputPending();
}
} else if (timeElapsed < maxInterval) {
// Yield if there's either a pending discrete or continuous input.
if (isInputPending !== null) {
return isInputPending(continuousOptions);
}
} else {
// We've blocked the thread for a long time. Even if there's no pending
// input, there may be some other scheduled work that we don't know about,
// like a network event. Yield now.
return true;
}
}
// `isInputPending` isn't available. Yield now.
return true;
}
js
function shouldYieldToHost() {
const timeElapsed = getCurrentTime() - startTime;
if (timeElapsed < frameInterval) {
// The main thread has only been blocked for a really short amount of time;
// smaller than a single frame. Don't yield yet.
return false;
}
// The main thread has been blocked for a non-negligible amount of time. We
// may want to yield control of the main thread, so the browser can perform
// high priority tasks. The main ones are painting and user input. If there's
// a pending paint or a pending input, then we should yield. But if there's
// neither, then we can yield less often while remaining responsive. We'll
// eventually yield regardless, since there could be a pending paint that
// wasn't accompanied by a call to `requestPaint`, or other main thread tasks
// like network events.
if (enableIsInputPending) {
if (needsPaint) {
// There's a pending paint (signaled by `requestPaint`). Yield now.
return true;
}
if (timeElapsed < continuousInputInterval) {
// We haven't blocked the thread for that long. Only yield if there's a
// pending discrete input (e.g. click). It's OK if there's pending
// continuous input (e.g. mouseover).
if (isInputPending !== null) {
return isInputPending();
}
} else if (timeElapsed < maxInterval) {
// Yield if there's either a pending discrete or continuous input.
if (isInputPending !== null) {
return isInputPending(continuousOptions);
}
} else {
// We've blocked the thread for a long time. Even if there's no pending
// input, there may be some other scheduled work that we don't know about,
// like a network event. Yield now.
return true;
}
}
// `isInputPending` isn't available. Yield now.
return true;
}

It is not complicated actually, the comments explains everything. The very basic lines are as below

js
const timeElapsed = getCurrentTime() - startTime;
if (timeElapsed < frameInterval) {
// The main thread has only been blocked for a really short amount of time;
// smaller than a single frame. Don't yield yet.
return false;
}
return true;
js
const timeElapsed = getCurrentTime() - startTime;
if (timeElapsed < frameInterval) {
// The main thread has only been blocked for a really short amount of time;
// smaller than a single frame. Don't yield yet.
return false;
}
return true;

So each task is given 5ms (frameInterval), if time is up then it should yield.

Notice that this is for running the task in Scheduler, not for each performUnitOfWork(). We can see that startTime is only set in performWorkUntilDeadline(), which means it is reset for each flushWork(), and if multiple tasks could be processed in on flushWork(), there is no yield in between.

This should help you understand this React quiz.

5. Summary

Phew, this is a lot. Letā€™s draw an overall diagram.

Still a few missing parts, but weā€™ve made a huge progress. Hope it helps you understand React internals better. It is already a big diagram to digest, letā€™s put other things into future episodes.

Want to know more about how React works internally?
Check out my series - React Internals Deep Dive!

šŸ˜³ Share my post ?    
or sponsor me

ā® Prev: React advanced patterns - Reusable behavior hooks through Ref

Next: Join our discord server "Hardcore React" to learn React internals together āÆ