First of all, I know one way to catch uncaught errors is use process.on('uncaughtException', err, function() {});.
I want to know how I can pass more details, or more context to the error. I want to be able to get the variables used. I'm not trying to recover from the error, only get more details of my environment when the error happened before shutting down the process. Yeah, the stack trace is nice, but I'd like to know how to replicate the error.
For example, I have this function:
function doTheBatman(var1) {
var var2 = 'whatever';
// this causes an uncaught exception later in the code
}
On process.on, I want to be able to access var1 and var2.
process.on('uncaughtException', function(err) {
// process.whatever doesn't contain any active variables
});
A synchronous exception in an Express route handler will be caught by Express itself and the exception will be passed off the default Express error handler where you can catch it yourself and the exception context is passed to that default express error handler.
You can see this code inside of Express where a route handler gets called:
Layer.prototype.handle_request = function handle(req, res, next) {
var fn = this.handle;
if (fn.length > 3) {
// not a standard request handler
return next();
}
try {
fn(req, res, next);
} catch (err) {
next(err);
}
};
The call to next(err) will pass the exception object off to the default Express error handler (which you can install a handler for).
If your code is throwing an exception inside of an asynchronous callback, then that is more complicated to catch in action. If you're using regular async callbacks (not promises), then the only way I know of to catch those at a meaningful spot is to put a try/catch inside of every async callback so you can capture the local stack info.
If you use promises at the lowest level and only program your logic and asynchronous code flow use promise functionality, then an exception in a promise handler will automatically turn into a rejected promise and certain promise libraries (like the Bluebird library can be configured to give you a pretty full stack trace of where things went wrong). Promises have this advantage because every promise .then() or .catch() handler is automatically wrapped in a try/catch handler and those are turned into promise rejections which propagate up the chain to the first .catch() handler they find.
For the non-Express example you just added to your question, you will just have to put a try/catch somehwere in the local neighborhood to catch a local synchronous exception:
function doTheBatman(var1) {
try {
var var2 = 'whatever';
// this causes an uncaught exception later in the code
} catch(e) {
console.log(e);
}
}
You can even set a debugger breakpoint in the catch handler and then examine variables in that scope when it is hit. I don't know of any way to examine the local variables at the point of the exception from the uncaughtException handler. err.stack should give you the stack trace, but not variable values.
Express/frameworks may offer more elegant/robust solutions, but if you're truly after what your question asks for... why not just capture variables outside of the functions scope? This is typically nasty and not considered best practice, but if you have a function that you know could be susceptive to failing often, perhaps the quick and dirty solution is what you need. You could always refine this later, but hopefully this demonstrates the approach...
var transactionVars = {};
function clearTransaction() {
transactionVars = {};
}
function doTheBatman(var1) {
transactionVars['var2'] = 'whatever';
// [...] bunch of stuff, possibly blowing up...
clearTransaction(); // we made it this far? cool, reset...
}
process.on('uncaughtException', function(err) {
console.log(transactionVars['var2']); // whatever
});
Furthermore, if you want to really dirty (in case these two functions are in two files) you can always tack transactionVars on the global object.
This is essentially a poor mans event emitter pattern, which I would highly recommend refactoring into once you grasp the general flow of how this works...
Related
Assume an Express route that makes a call to Mongoose and has to be async so it can await on the mongoose.find(). Also assume we are receiving XML but we have to change it to JSON, and that also needs to be async so I can call await inside of it.
If I do this:
app.post('/ams', async (req, res) => {
try {
xml2js.parseString(xml, async (err, json) => {
if (err) {
throw new XMLException();
}
// assume many more clauses here that can throw exceptions
res.status(200);
res.send("Data saved")
});
} catch(err) {
if (err instanceof XML2JSException) {
res.status(400);
message = "Malformed XML error: " + err;
res.send(message);
}
}
}
The server hangs forever. I'm assuming the async/await means that the server hits a timeout before something concludes.
If I put this:
res.status(200);
res.send("Data saved")
on the line before the catch(), then that is returned, but it is the only thing every returned. The client gets a 200, even if an XMLException is thrown.
I can see the XMLException throw in the console, but I cannot get a 400 to send back. I cannot get anything I that catch block to execute in a way that communicates the response to the client.
Is there a way to do this?
In a nutshell, there is no way to propagate an error from the xml2js.parseString() callback up to the higher code because that parent function has already exited and returned. This is how plain callbacks work with asynchronous code.
To understand the problem here, you have to follow the code flow for xml2js.parseString() in your function. If you instrumented it like this:
app.post('/ams', async (req, res) => {
try {
console.log("1");
xml2js.parseString(xml, async (err, json) => {
console.log("2");
if (err) {
throw new XMLException();
}
// assume many more clauses here that can throw exceptions
res.status(200);
res.send("Data saved")
});
console.log("3");
} catch (err) {
if (err instanceof XML2JSException) {
res.status(400);
message = "Malformed XML error: " + err;
res.send(message);
}
}
console.log("4");
});
Then, you would see this in the logs:
1 // about to call xml2js.parseString()
3 // after the call to xml2js.parseString()
4 // function about to exit
2 // callback called last after function returned
The outer function has finished and returned BEFORE your callback has been called. This is because xml2js.parseString() is asynchronous and non-blocking. That means that calling it just initiates the operation and then it immediately returns and the rest of your function continues to execute. It works in the background and some time later, it posts an event to the Javascript event queue and when the interpreter is done with whatever else it was doing, it will pick up that event and call the callback.
The callback will get called with an almost empty call stack. So, you can't use traditional try/catch exceptions with these plain, asynchronous callbacks. Instead, you must either handle the error inside the callback or call some function from within the callback to handle the error for you.
When you try to throw inside that plain, asynchronous callback, the exception just goes back into the event handler that triggered the completion of the asynchronous operation and no further because there's nothing else on the call stack. Your try/catch you show in your code cannot catch that exception. In fact, no other code can catch that exception - only code within the exception.
This is not a great way to write code, but nodejs survived with it for many years (by not using throw in these circumstances). However, this is why promises were invented and when used with the newer language features async/await, they provide a cleaner way to do things.
And, fortunately in this circumstance xml2js.parseString() has a promise interface already.
So, you can do this:
app.post('/ams', async (req, res) => {
try {
// get the xml data from somewhere
const json = await xml2js.parseString(xml);
// do something with json here
res.send("Data saved");
} catch (err) {
console.log(err);
res.status(400).send("Malformed XML error: " + err.message);
}
});
With the xml2js.parseString() interface, if you do NOT pass it a callback, it will return a promise instead that resolves to the final value or rejects with an error. This is not something all asynchronous interfaces can do, but is fairly common these days if the interface had the older style callback originally and then they want to now support promises. Newer interfaces are generally just built with only promise-based interfaces. Anyway, per the doc, this interface will return a promise if you don't pass a callback.
You can then use await with that promise that the function returns. If the promise resolves, the await will retrieve the resolved value of the promise. If the promise rejects, because you awaiting the rejection will be caught by the try/catch. FYI, you can also use .then() and .catch() with the promise, but in many cases, async and await are simpler so that's what I've shown here.
So, in this code, if there is invalid XML, then the promise that xml2js.parseString() returns will reject and control flow will go to the catch block where you can handle the error.
If you want to capture only the xml2js.parseString() error separately from other exceptions that could occur elsewhere in your code, you can put a try/catch around just it (though this code didn't show anything else that would likely throw an exception so I didn't add another try/catch). In fact, this form of try/catch can be used pretty much like you would normally use it with synchronous code. You can throw up to a higher level of try/catch too.
A few other notes, many people who first start programming with asynchronous operations try to just put await in front of anything asynchronous and hope that it solves their problem. await only does anything useful when you await a promise so your asynchronous function must return a promise that resolves/rejects when the asynchronous operation is complete for the await to do anything useful.
It is also possible to take a plain callback asynchronous function that does not have a promise interface and wrap a promise interface around it. You pretty much never want to mix promise interface functions with plain callback asynchronous operations because error handling and propagation is a nightmare with a mixed model. So, sometimes you have to "promisify" an older interface so you can use promises with it. In most cases, you can do that with util.promisify() built into the util library in nodejs. Fortunately, since promises and async/await are the modern and easier way to do asynchronous things, most newer asynchronous interfaces in the nodejs world come with promise interfaces already.
You are throwing exceptions inside the callback function. So you cant expect the catch block of the router to receive it.
One way to handle this is by using util.promisify.
try{
const util = require('util');
const parseString = util.promisify(xml2js.parseString);
let json = await parsestring(xml);
}catch(err)
{
...
}
When the Bluebird module detects a non-Error reject, it reports Warning: a promise was rejected with a non-error.
So when we are writing a reusable promise library, how should we handle the situation when it is the client's code that throws simple values?
Simple my-library example:
var Promise = require('bluebird');
module.exports = function (cb) {
return new Promise(function (resolve, reject) {
try {
cb();
resolve('some data');
} catch (e) {
// should we do this?
reject(e);
// or should we do this?
// reject(e instanceof Error ? e : new Error(e));
}
});
};
A client that consumes it:
var lib = require('my-library');
lib(function () {
throw 'Regular text';
})
.catch(function (error) {
// - should we expect exactly 'Regular text' here?
// - or should it be wrapped into Error by the promise library?
});
In the example above we are going to see the same Warning: a promise was rejected with a non-error again, which in this case is obviously the client's fault, but the problem is - it will be reported against the promise library that did the reject.
This brings a question of how a reusable library should handle such cases.
Option 1: do nothing (just use reject(e)), let the client see the warning message Warning: a promise was rejected with a non-error.
Option 2: verify if the error thrown is an instance of Error and if not - do reject(new Error(e)). This enables the client to throw and reject with simple values, which is probably bad.
I am looking for a general recommendation here.
Also consider that sometimes the client can be an old library that you are trying to wrap into promises.
You should definitely and objectively wrap non-errors with errors.
Non errors are terrible for debugging, the bluebird warning is there because it is very hard to figure out where an error came from if it's a primitive. You can't attach properties to primitives and they have no stack traces.
Stack traces mean you don't have to guess where the errors come from - you know.
That said, bluebird already does this automatically with promisify and promisifyAll and consider using them. If you're doing this manually - just make sure you throw or reject things with stacks too.
Node in addition looks for the stack traces of your unhandled rejections - so rejecting with junk will not only give you useless information in Node but it might actually crash your program.
Error messages should be wrapped in error objects, but not automatically.
Automatic wrapping reduces the usefulness of stack traces. For example, lets say the client function is f and your library function is g:
function f(x) { throw 'Hello world' }
function g(f, x) { try { f(x); } catch (e) { throw new Error(e); } }
g(f, 1)
The client function and line numbers where the error originated will be completely missing from the stack:
Error: Hello world
at g (repl:1:52)
at repl:1:1
While if the client function creates the error object:
function f(x) { throw new Error('Hello world'); }
function g(f, x) { try{ f(x); } catch (e) { throw e; } }
g(f, 1)
we get a full stack trace:
Error: Hello world
at f (repl:1:23)
at g (repl:1:25)
at repl:1:1
There is no way to recover the stack up to the point in your library, making it useless for users (they wont be able to debug their own code).
That said, this largely depends on what sort of library you're making and what sort of contract you would like it to present to the users. You should think very carefully before using user-passed callbacks, as they invert the normal chain of responsibility in the stack.
Promises make this easier because they un-invert the inversion: the library is called first to generate a promise, then that promise is passed a callback. As such, creating a new error together with the creation of the promise allows Bluebird to at least partially look into / augment the stack of the client code (the line that created the promise will be present in the stack). The warning is still there because its impossible to provide a complete stack.
No, they simply want you to wrap your JSON error objects into something.
This is not freedom, it's dictature.
There's no point in wrapping simple JSON objects in bulky Error wrappers.
Don't listen to them.
I spent too much time trying to figure out why my express.js controller did not respond to a simple query, and figured out that runtime errors fired from a Mongoose-promise callback were silently interrupting the callback process.
Here is a simplified version of my code:
server.get('/api/test', function (req, res, next) {
User.find({}).exec().then(function success(users){
console.log('SUCCESS');
typo[0] = 1; // throws a runtime error
res.json(users);
}, function error(err){
console.log('ERROR');
res.json({error: err});
});
});
This results in SUCCESS showing up in my console, but nothing happens then. No response is given to the user, the error caused by my typo is not appearing in my console, and the error callback is not called either.
I am aware that one should not throw exceptions from a callback function, but in that case, this was just a typo, and it would make sense to me to be warned (e.g. a stack trace in my standard output) whenever one makes this kind of mistake. (we're humans, after all...)
In your opinion, what's the best way to get feedback whenever this kind of mistakes are made in promise callbacks?
This is Mongoose's fault for using a bad promise implementation. Promises are throw-safe so exceptions are caught (so they can be later handled by future code) - the future code never comes and Mongoose never reports that it did not. Good promise implementations do not suffer from this issue.
Your options are two:
Use a library like Bluebird:
var Promise = require("bluebird");
var mongoose = Promise.promisifyAll(require("mongoose"));
User.findAsync({}).then(function(data){
JSON.prase("dsa"); // not a silent failure, will show up, easy debugging
});
This has the advantage of being faster than mongoose promises so there is no performance penalty. Alternatively, if you're super conservative and don't want the performance and API gains of bluebird - you can use native promises:
// Promise is the native promise
Promise.resolve(User.find({}).exec()).then(function(data){
JSON.prase("dsa");
});
And then, assuming you're running a modern variant of nodejs (read: io.js v 1.4.1 or upper), you can subscribe to promise rejections:
process.on("unhandledRejection", function(p, why){
console.log("FOUND ERROR!!!!", p , why);
});
So exceptions are not silently suppressed.
The exec() has two promises
.then(function)
.then(null , function)
try this, I think it will help
server.get('/api/test', function(req, res, next) {
User.find({}).exec()
.then(function success(users) {
console.log('SUCCESS');
typo[0] = 1; // throws a runtime error
res.json(users);
})
.then(null, function error(err) {
console.log('ERROR');
res.json({
error: err
});
});
});
The Nodejs API states...
By the very nature of how throw works in JavaScript, there is almost
never any way to safely "pick up where you left off", without leaking
references, or creating some other sort of undefined brittle state.
However Koa traps errors and avoids exiting the nodejs process. What enables Koa to safely flout this advice?
There are primarily two cases where koa can not safely handle errors.
Thrown errors on different ticks:
app.use(function* () {
setImmediate(function () {
throw new Error('boom')
})
})
Emitter errors that are not set as response.body=:
app.use(function* () {
this.response.body = stream.pipe(zlib.createGzip())
})
Any function or library that does the first case is malformed and should not be used. If the function/library uses promises and/or callbacks correctly, it will never happen.
For emitters, simply always set each stream as the body (or use middleware):
app.use(function* () {
this.response.body = stream
this.response.body = this.response.body.pipe(zlib.createGzip())
})
Koa does this by allowing you to use try/catch on "async" stuff, specifically callbacks and promises. You can't, however, try/catch an error thrown on a different tick or an emitter.
I want to use Node.js Domains to catch exceptions. It is working so far, but there is one place I can't get domains to catch the exception. exception2 in the callback is caught and handled in the domain.on('error') handler, but exception1 is not caught. The odd thing is that when exception1 is thrown, it doesn't shutdown Node like I would expect. Here is my example app:
var domain = require('domain');
var request = require('request');
var express = require('express');
var serverDomain = domain.create();
serverDomain.on('error', function(err) {
console.log("Server Domain Error: " + err);
});
var app;
serverDomain.run(function() {
app = express();
app.listen(3000);
});
app.use(function(req, res, next) {
var reqDomain = domain.create();
reqDomain.add(req);
reqDomain.add(res);
reqDomain.on('error', function(err) {
console.log("Req Domain Error: " + err);
reqDomain.dispose();
next(err);
});
next();
});
app.get('/', function(req, res) {
var uri = "http://google.com";
exception1.go();
request.get({url:uri, json: {}},
function (error, response, body) {
if(response.statusCode === 200) {
exception2.go();
res.send('Success getting google response');
}
});
});
To get exception2 to execute, I comment out exception 1.
The problem is that the exception happens during Connect's routing, which has both a try/catch block around its execution, as well as a default error handler which prints out stack trace details when running in a non-production mode. Since the exception is handled inside of Express, it never reaches your outer layer for the domains to handle.
How it differs from exception2 is that the handler function for the '/' route is executed directly by that Connect block, in the same stack as the original call that went through Express. The second exception occurs in a callback, after some I/O operation has returned, and therefore is executed by a stack originating from Node's event loop I/O handler, and so the try/catch of Express isn't available to snag that exception and save the app server. In fact, if you comment out all the domain stuff, and trip exception2 it crashes Node.
Since only unhandled exceptions are routed to the domain mechanism, and since exception1 has a try/catch visible in it's call stack above it, the exception is handled, and not forwarded to the domain.
Connect-domain allows you to catch all errors for connect modules including express.
Connect-domain
https://github.com/baryshev/connect-domain
The example for express3:
http://masashi-k.blogspot.com/2012/12/express3-global-error-handling-domain.html
#user1007983
No, in production, the try/catch handling still exists in Connect/Express. The way to solve it is to add your own try/catch block in the "root" which you can then use to emit an "error" event to the domain before connect sends the error response.
try {
// .. code here ..
} catch (err) {
domain.emit('error', err);
}
Another way to get around it is to just disconnect from the handler, like wrapping your code in a setImmediate block
I've tried try/catch blocks (which may not work the way you think with async code). I've tried node domains in several different ways. But I was unable to handle an exception thrown in a 3rd party lib (sequelize). Why did I get the exception? Well, it was because the SQL that was generated was not well formed. (My fault).
Anywho, the solution that worked best for me and my (small) project was to use forever. Let the exceptions happen, but fire up node again if they do. I doubt it is the most elegant solution, but it works for me and my small project.
With a larger project, domains combined with the clustering API might be a good choice.
Winston is another choice. It might be cool to combine forever with winston so that if you DO get an exception, you can have winston email you, so you can fix the code. Meanwhile, forever will happily restart the app for you.
I came across this problem while testing my domain-based error handling.
I went with the approach suggested here by Issac, with a couple of minor changes: use 'domain.active' to get the currently active domain and emit the error event on that, and I used a wrapper function to avoid having to modify all of my handler functions:
domainWrapper = function(func) {
return function(req, res) {
try {
return func(req, res);
} catch (err) {
domain.active.emit('error', err);
}
}
}
Then changed this sort of thing:
app.get('/jobs', handlerFunc);
to this:
app.get('/jobs', domainWrapper(handlerFunc));