Take advantage of HTTP PATCH to partially update resources and transfer less data over the wire
The HTTP verb named PATCH can be used for partial updates, i.e., when you would like to update just a few fields of a resource. While you could update a resource “partially” using HTTP PUT, by sending the entire resource with the updated values, that is potentially problematic. At the very least, you might end up consuming more network bandwidth than necessary.
For partial updates, HTTP PATCH is easier and safer, and ASP.NET Web API provides excellent support for HTTP PATCH requests. This article will discuss how we can use PATCH to perform partial updates when working with RESTful services using Web API.
PATCH vs. PUT
The HTTP PATCH method should be used whenever you would like to change or update just a small part of the state of the resource. You should use the PUT method only when you would like to replace the resource in its entirety. Note that PATCH is not a replacement for PUT or POST, but just a way of applying a delta update to a resource representation. Roy Fielding, who authored the REST architectural style and many web standards, said that he created PATCH because “partial PUT is never RESTful.”
Let’s look at a good example of a PATCH request. Imagine you have a model named Customer and not all of its members are editable. ASP.NET Web API incorporates nice support for HTTP PATCH requests using the Delta<T> class, which was included as part of the OData additions. We will cover how to use the Delta<T> class in a later post. In the section that follows, we will examine how we can implement HTTP PATCH support in Web API in the simplest way.
Implementing HTTP PATCH in WebAPI
Let’s now get into a bit of code. Create a new Web API project in Visual Studio. In the new project you have created, create a controller named CustomerController (you can leave it as empty for now) and the following Customer model class.
Now, suppose only the two properties City and Country in the Customermodel are editable. Let’s create a delta request class to update the Customer model. We’ll call it the CustomerPatchRequest class.
Note how no state is changed and a HTTP 404 is returned if the customer being searched is not found in the list. If the customer is found, a delta change is performed on the customer object and then a HTTP 200 is returned. Ideally, you should persist the data after altering the object’s state.
The customersList object is a list of customer instances. It is populated as shown in the code snippet below.
You can call this Web API method via Fiddler. To do this, select PATCH as the HTTP Verb, specify the URL in the Composer window, and specify the following in the request body.
{”Id”:19,”City”:”NewYork”,”Country”:”USA”}
Finally, specify the content type in the Composer window as shown in the code snippet below.
Content-Type: application/json
Here is the complete code listing of the CustomerController for your reference.
HTTP PATCH is a nice way to update a resource by specifying only the properties that have changed. There are many ways in which HTTP PATCH can be implemented in a RESTful web API. This article presented the simplest way. In a future post, we will examine how we can take advantage of the JsonPatch library to implement HTTP PATCH support.
GET access a resource in a read-only mode POST Normally used to send a new resource into the server (create action) PUT Normally used to update a given resource (update action) DELETE Used to delete a resource HEAD Not part of the CrUD actions, but the verb is used to ask if a given resource exists without returning any of its representations OPTIONS Not part of the CrUD actions, but used to retrieve a list of available verbs on a given resource (i.e., What can the client do with a specifc resource?)
Source: https://medium.com/@tkssharma/writing-neat-asynchronous-node-js-code-with-promises-async-await-fa8d8b0bcd7c Async functions return a Promise. If the function throws an error, the Promise will be rejected. If the function returns a value, the Promise will be resolved.
Await
Async functions can make use of the await expression. This will pause the async function and wait for the Promise to resolve prior to moving on.
We have a sequence of asynchronous tasks to be done one after another. For instance, loading scripts.
How to code it well?
Promises provide a couple of recipes to do that.
In this chapter we cover promise chaining.
It looks like this:
newPromise(function(resolve, reject){setTimeout(()=>resolve(1),1000);// (*)}).then(function(result){// (**)alert(result);// 1return result *2;}).then(function(result){// (***)alert(result);// 2return result *2;}).then(function(result){alert(result);// 4return result *2;});
The idea is that the result is passed through the chain of .then handlers.
Here the flow is:
The initial promise resolves in 1 second (*),
Then the .then handler is called (**).
The value that it returns is passed to the next .then handler (***)
…and so on.
As the result is passed along the chain of handlers, we can see a sequence of alert calls: 1 → 2 → 4.
The whole thing works, because a call to promise.then returns a promise, so that we can call the next .then on it.
When a handler returns a value, it becomes the result of that promise, so the next .then is called with it.
To make these words more clear, here’s the start of the chain:
newPromise(function(resolve, reject){setTimeout(()=>resolve(1),1000);}).then(function(result){alert(result);return result *2;// <-- (1)})// <-- (2)// .then…
The value returned by .then is a promise, that’s why we are able to add another .then at (2). When the value is returned in (1), that promise becomes resolved, so the next handler runs with the value.
Unlike the chaining, technically we can also add many .then to a single promise, like this:
let promise =newPromise(function(resolve, reject){setTimeout(()=>resolve(1),1000);});
promise.then(function(result){alert(result);// 1return result *2;});
promise.then(function(result){alert(result);// 1return result *2;});
promise.then(function(result){alert(result);// 1return result *2;});
…But that’s a totally different thing. Here’s the picture (compare it with the chaining above):
All .then on the same promise get the same result – the result of that promise. So in the code above all alert show the same: 1. There is no result-passing between them.
In practice we rarely need multiple handlers for one promise. Chaining is used much more often.
Normally, a value returned by a .then handler is immediately passed to the next handler. But there’s an exception.
If the returned value is a promise, then the further execution is suspended until it settles. After that, the result of that promise is given to the next .then handler.
Here the first .then shows 1 returns new Promise(…) in the line (*). After one second it resolves, and the result (the argument of resolve, here it’s result*2) is passed on to handler of the second .then in the line (**). It shows 2 and does the same thing.
So the output is again 1 → 2 → 4, but now with 1 second delay between alert calls.
Returning promises allows us to build chains of asynchronous actions.
Let’s use this feature with loadScript to load scripts one by one, in sequence:
loadScript("/article/promise-chaining/one.js").then(function(script){returnloadScript("/article/promise-chaining/two.js");}).then(function(script){returnloadScript("/article/promise-chaining/three.js");}).then(function(script){// use functions declared in scripts// to show that they indeed loadedone();two();three();});
Here each loadScript call returns a promise, and the next .then runs when it resolves. Then it initiates the loading of the next script. So scripts are loaded one after another.
We can add more asynchronous actions to the chain. Please note that code is still “flat”, it grows down, not to the right. There are no signs of “pyramid of doom”.
Please note that technically it is also possible to write .then directly after each promise, without returning them, like this:
loadScript("/article/promise-chaining/one.js").then(function(script1){loadScript("/article/promise-chaining/two.js").then(function(script2){loadScript("/article/promise-chaining/three.js").then(function(script3){// this function has access to variables script1, script2 and script3one();two();three();});});});
This code does the same: loads 3 scripts in sequence. But it “grows to the right”. So we have the same problem as with callbacks. Use chaining (return promises from .then) to evade it.
Sometimes it’s ok to write .then directly, because the nested function has access to the outer scope (here the most nested callback has access to all variables scriptX), but that’s an exception rather than a rule.
Thenables
To be precise, .then may return an arbitrary “thenable” object, and it will be treated the same way as a promise.
A “thenable” object is any object with a method .then.
The idea is that 3rd-party libraries may implement “promise-compatible” objects of their own. They can have extended set of methods, but also be compatible with native promises, because they implement .then.
Here’s an example of a thenable object:
classThenable{constructor(num){this.num = num;}then(resolve, reject){alert(resolve);// function() { native code }// resolve with this.num*2 after the 1 secondsetTimeout(()=>resolve(this.num *2),1000);// (**)}}newPromise(resolve =>resolve(1)).then(result =>{returnnewThenable(result);// (*)}).then(alert);// shows 2 after 1000ms
JavaScript checks the object returned by .then handler in the line (*): if it has a callable method named then, then it calls that method providing native functions resolve, reject as arguments (similar to executor) and waits until one of them is called. In the example above resolve(2) is called after 1 second (**). Then the result is passed further down the chain.
This feature allows to integrate custom objects with promise chains without having to inherit from Promise.
In frontend programming promises are often used for network requests. So let’s see an extended example of that.
We’ll use the fetch method to load the information about the user from the remote server. The method is quite complex, it has many optional parameters, but the basic usage is quite simple:
let promise =fetch(url);
This makes a network request to the url and returns a promise. The promise resolves with a response object when the remote server responds with headers, but before the full response is downloaded.
To read the full response, we should call a method response.text(): it returns a promise that resolves when the full text downloaded from the remote server, with that text as a result.
The code below makes a request to user.json and loads its text from the server:
fetch('/article/promise-chaining/user.json')// .then below runs when the remote server responds.then(function(response){// response.text() returns a new promise that resolves with the full response text// when we finish downloading itreturn response.text();}).then(function(text){// ...and here's the content of the remote filealert(text);// {"name": "iliakan", isAdmin: true}});
There is also a method response.json() that reads the remote data and parses it as JSON. In our case that’s even more convenient, so let’s switch to it.
We’ll also use arrow functions for brevity:
// same as above, but response.json() parses the remote content as JSONfetch('/article/promise-chaining/user.json').then(response => response.json()).then(user =>alert(user.name));// iliakan
Now let’s do something with the loaded user.
For instance, we can make one more request to github, load the user profile and show the avatar:
// Make a request for user.jsonfetch('/article/promise-chaining/user.json')// Load it as json.then(response => response.json())// Make a request to github.then(user =>fetch(`https://api.github.com/users/${user.name}`))// Load the response as json.then(response => response.json())// Show the avatar image (githubUser.avatar_url) for 3 seconds (maybe animate it).then(githubUser =>{let img = document.createElement('img');
img.src = githubUser.avatar_url;
img.className ="promise-avatar-example";
document.body.append(img);setTimeout(()=> img.remove(),3000);// (*)});
The code works, see comments about the details, but it should be quite self-descriptive. Although, there’s a potential problem in it, a typical error of those who begin to use promises.
Look at the line (*): how can we do something after the avatar has finished showing and gets removed? For instance, we’d like to show a form for editing that user or something else. As of now, there’s no way.
To make the chain extendable, we need to return a promise that resolves when the avatar finishes showing.
Now right after setTimeout runs img.remove(), it calls resolve(githubUser), thus passing the control to the next .thenin the chain and passing forward the user data.
As a rule, an asynchronous action should always return a promise.
That makes possible to plan actions after it. Even if we don’t plan to extend the chain now, we may need it later.
Finally, we can split the code into reusable functions:
Asynchronous actions may sometimes fail: in case of an error the corresponding promise becomes rejected. For instance, fetchfails if the remote server is not available. We can use .catch to handle errors (rejections).
Promise chaining is great at that aspect. When a promise rejects, the control jumps to the closest rejection handler down the chain. That’s very convenient in practice.
For instance, in the code below the URL is wrong (no such server) and .catch handles the error:
fetch('https://no-such-server.blabla')// rejects.then(response => response.json()).catch(err =>alert(err))// TypeError: failed to fetch (the text may vary)
Or, maybe, everything is all right with the server, but the response is not a valid JSON:
fetch('/')// fetch works fine now, the server responds successfully.then(response => response.json())// rejects: the page is HTML, not a valid json.catch(err =>alert(err))// SyntaxError: Unexpected token < in JSON at position 0
In the example below we append .catch to handle all errors in the avatar-loading-and-showing chain:
The "invisible try..catch" around the executor automatically catches the error and treats it as a rejection.
That’s so not only in the executor, but in handlers as well. If we throw inside .then handler, that means a rejected promise, so the control jumps to the nearest error handler.
Here’s an example:
newPromise(function(resolve, reject){resolve("ok");}).then(function(result){thrownewError("Whoops!");// rejects the promise}).catch(alert);// Error: Whoops!
That’s so not only for throw, but for any errors, including programming errors as well:
newPromise(function(resolve, reject){resolve("ok");}).then(function(result){blabla();// no such function}).catch(alert);// ReferenceError: blabla is not defined
As a side effect, the final .catch not only catches explicit rejections, but also occasional errors in the handlers above.
As we already noticed, .catch behaves like try..catch. We may have as many .then as we want, and then use a single .catch at the end to handle errors in all of them.
In a regular try..catch we can analyze the error and maybe rethrow it if can’t handle. The same thing is possible for promises. If we throw inside .catch, then the control goes to the next closest error handler. And if we handle the error and finish normally, then it continues to the closest successful .then handler.
In the example below the .catch successfully handles the error:
// the execution: catch -> thennewPromise(function(resolve, reject){thrownewError("Whoops!");}).catch(function(error){alert("The error is handled, continue normally");}).then(()=>alert("Next successful handler runs"));
Here the .catch block finishes normally. So the next successful handler is called. Or it could return something, that would be the same.
…And here the .catch block analyzes the error and throws it again:
// the execution: catch -> catch -> thennewPromise(function(resolve, reject){thrownewError("Whoops!");}).catch(function(error){// (*)if(error instanceofURIError){// handle it}else{alert("Can't handle such error");throw error;// throwing this or another error jumps to the next catch}}).then(function(){/* never runs here */}).catch(error =>{// (**)alert(`The unknown error has occurred: ${error}`);// don't return anything => execution goes the normal way});
The handler (*) catches the error and just can’t handle it, because it’s not URIError, so it throws it again. Then the execution jumps to the next .catch down the chain (**).
In the section below we’ll see a practical example of rethrowing.
Let’s improve error handling for the user-loading example.
The promise returned by fetch rejects when it’s impossible to make a request. For instance, a remote server is not available, or the URL is malformed. But if the remote server responds with error 404, or even error 500, then it’s considered a valid response.
What if the server returns a non-JSON page with error 500 in the line (*)? What if there’s no such user, and github returns a page with error 404 at (**)?
fetch('no-such-user.json')// (*).then(response => response.json()).then(user =>fetch(`https://api.github.com/users/${user.name}`))// (**).then(response => response.json()).catch(alert);// SyntaxError: Unexpected token < in JSON at position 0// ...
As of now, the code tries to load the response as JSON no matter what and dies with a syntax error. You can see that by running the example above, as the file no-such-user.json doesn’t exist.
That’s not good, because the error just falls through the chain, without details: what failed and where.
So let’s add one more step: we should check the response.status property that has HTTP status, and if it’s not 200, then throw an error.
classHttpErrorextendsError{// (1)constructor(response){super(`${response.status} for ${response.url}`);this.name ='HttpError';this.response = response;}}functionloadJson(url){// (2)returnfetch(url).then(response =>{if(response.status ==200){return response.json();}else{thrownewHttpError(response);}})}loadJson('no-such-user.json')// (3).catch(alert);// HttpError: 404 for .../no-such-user.json
We make a custom class for HTTP Errors to distinguish them from other types of errors. Besides, the new class has a constructor that accepts the response object and saves it in the error. So error-handling code will be able to access it.
Then we put together the requesting and error-handling code into a function that fetches the urland treats any non-200 status as an error. That’s convenient, because we often need such logic.
Now alert shows better message.
The great thing about having our own class for errors is that we can easily check for it in error-handling code.
For instance, we can make a request, and then if we get 404 – ask the user to modify the information.
The code below loads a user with the given name from github. If there’s no such user, then it asks for the correct name:
functiondemoGithubUser(){let name =prompt("Enter a name?","iliakan");returnloadJson(`https://api.github.com/users/${name}`).then(user =>{alert(`Full name: ${user.name}.`);// (1)return user;}).catch(err =>{if(err instanceofHttpError&& err.response.status ==404){// (2)alert("No such user, please reenter.");returndemoGithubUser();}else{throw err;}});}demoGithubUser();
Here:
If loadJson returns a valid user object, then the name is shown (1), and the user is returned, so that we can add more user-related actions to the chain. In that case the .catch below is ignored, everything’s very simple and fine.
Otherwise, in case of an error, we check it in the line (2). Only if it’s indeed the HTTP error, and the status is 404 (Not found), we ask the user to reenter. For other errors – we don’t know how to handle, so we just rethrow them.
What happens when an error is not handled? For instance, after the rethrow as in the example above. Or if we forget to append an error handler to the end of the chain, like here:
newPromise(function(){noSuchFunction();// Error here (no such function)});// no .catch attached
Or here:
// a chain of promises without .catch at the endnewPromise(function(){thrownewError("Whoops!");}).then(function(){// ...something...}).then(function(){// ...something else...}).then(function(){// ...but no catch after it!});
In case of an error, the promise state becomes “rejected”, and the execution should jump to the closest rejection handler. But there is no such handler in the examples above. So the error gets “stuck”.
In practice, that’s usually because of the bad code. Indeed, how come there’s no error handling?
Most JavaScript engines track such situations and generate a global error in that case. We can see it in the console.
In the browser we can catch it using the event unhandledrejection:
window.addEventListener('unhandledrejection',function(event){// the event object has two special properties:alert(event.promise);// [object Promise] - the promise that generated the erroralert(event.reason);// Error: Whoops! - the unhandled error object});newPromise(function(){thrownewError("Whoops!");});// no catch to handle the error
The event is the part of the HTML standard. Now if an error occurs, and there’s no .catch, the unhandledrejection handler triggers: the event object has the information about the error, so we can do something with it.
Usually such errors are unrecoverable, so our best way out is to inform the user about the problem and probably report about the incident to the server.
In non-browser environments like Node.JS there are other similar ways to track unhandled errors.
To summarize, .then/catch(handler) returns a new promise that changes depending on what handler does:
If it returns a value or finishes without a return (same as return undefined), then the new promise becomes resolved, and the closest resolve handler (the first argument of .then) is called with that value.
If it throws an error, then the new promise becomes rejected, and the closest rejection handler (second argument of .thenor .catch) is called with it.
If it returns a promise, then JavaScript waits until it settles and then acts on its outcome the same way.
The picture of how the promise returned by .then/catch changes:
The smaller picture of how handlers are called:
In the examples of error handling above the .catch was always the last in the chain. In practice though, not every promise chain has a .catch. Just like regular code is not always wrapped in try..catch.
We should place .catch exactly in the places where we want to handle errors and know how to handle them. Using custom error classes can help to analyze errors and rethrow those that we can’t handle.
For errors that fall outside of our scope we should have the unhandledrejection event handler (for browsers, and analogs for other environments). Such unknown errors are usually unrecoverable, so all we should do is to inform the user and probably report to our server about the incident.