nerror: rich JavaScript errors

Netflix services uses VError to make operation of Node.js applications easier through meaningful error chains. VError is an amazing library by Joyent and we are glad for all the hard work for the contributors made during the years. In early 2019 Netflix error handling requirements started to broaden enough that we had to find a way to make quick iterations on VError with minimizing the churn on existing VError customers. As a result of this we decided to fork VError as NError. We hope in the future after the initial development period we can seek convergence between the two projects.

This module provides several classes in support of Joyent's Best Practices for Error Handling in Node.js. If you find any of the behavior here confusing or surprising, check out that document first.

API

See API

Classes

The error classes here support:

• printf-style arguments for the message
• chains of causes
• properties to provide extra information about the error
• creating your own subclasses that support all of these

The classes here are:

• VError, for chaining errors while preserving each one's error message. This is useful in servers and command-line utilities when you want to propagate an error up a call stack, but allow various levels to add their own context. See examples below.
• WError, for wrapping errors while hiding the lower-level messages from the top-level error. This is useful for API endpoints where you don't want to expose internal error messages, but you still want to preserve the error chain for logging and debugging.
• PError, which is just like VError but does not interpret printf-style arguments at all.
• SError, which is just like VError but interprets printf-style arguments more strictly.
• MultiError, which is just an Error that encapsulates one or more other errors. (This is used for parallel operations that return several errors.)

For the full list of features see API.

Quick start

First, install the package:

npm install @netflix/nerror

If nothing else, you can use VError as a drop-in replacement for the built-in JavaScript Error class, with the addition of printf-style messages:

const { VError } = require('@netflix/nerror');
const err = new VError('missing file: "%s"', '/etc/passwd');
console.log(err.message);

This prints:

missing file: "/etc/passwd"

You can also pass a cause argument, which is any other Error object:

const fs = require('fs');
const filename = '/nonexistent';
fs.stat(filename, function (err1) {
    const err2 = new VError(err1, 'stat "%s"', filename);
    console.error(err2.message);
});

This prints out:

stat "/nonexistent": ENOENT, stat '/nonexistent'

which resembles how Unix programs typically report errors:

$ sort /nonexistent
sort: open failed: /nonexistent: No such file or directory

To match the Unixy feel, when you print out the error, just prepend the program's name to the VError's message. Or just call node-cmdutil.fail(your_verror), which does this for you.

You can get the next-level Error using err.cause():

console.error(err2.cause().message);

prints:

ENOENT, stat '/nonexistent'

Of course, you can chain these as many times as you want, and it works with any kind of Error:

const err1 = new Error('No such file or directory');
const err2 = new VError(err1, 'failed to stat "%s"', '/junk');
const err3 = new VError(err2, 'request failed');
console.error(err3.message);

This prints:

request failed: failed to stat "/junk": No such file or directory

The idea is that each layer in the stack annotates the error with a description of what it was doing. The end result is a message that explains what happened at each level.

You can also decorate Error objects with additional information so that callers can not only handle each kind of error differently, but also construct their own error messages (e.g., to localize them, format them, group them by type, and so on). See the example below.

Deeper dive

The two main goals for VError are:

• Make it easy to construct clear, complete error messages intended for people. Clear error messages greatly improve both user experience and debuggability, so we wanted to make it easy to build them. That's why the constructor takes printf-style arguments.
• Make it easy to construct objects with programmatically-accessible metadata (which we call informational properties). Instead of just saying "connection refused while connecting to 192.168.1.2:80", you can add properties like "ip": "192.168.1.2" and "tcpPort": 80. This can be used for feeding into monitoring systems, analyzing large numbers of Errors (as from a log file), or localizing error messages.

To really make this useful, it also needs to be easy to compose Errors: higher-level code should be able to augment the Errors reported by lower-level code to provide a more complete description of what happened. Instead of saying "connection refused", you can say "operation X failed: connection refused". That's why VError supports causes.

In order for all this to work, programmers need to know that it's generally safe to wrap lower-level Errors with higher-level ones. If you have existing code that handles Errors produced by a library, you should be able to wrap those Errors with a VError to add information without breaking the error handling code. There are two obvious ways that this could break such consumers:

• The error's name might change. People typically use name to determine what kind of Error they've got. To ensure compatibility, you can create VErrors with custom names, but this approach isn't great because it prevents you from representing complex failures. For this reason, VError provides findCauseByName, which essentially asks: does this Error or any of its causes have this specific type? If error handling code uses findCauseByName, then subsystems can construct very specific causal chains for debuggability and still let people handle simple cases easily. There's an example below.
• The error's properties might change. People often hang additional properties off of Error objects. If we wrap an existing Error in a new Error, those properties would be lost unless we copied them. But there are a variety of both standard and non-standard Error properties that should not be copied in this way: most obviously name, message, and stack, but also fileName, lineNumber, and a few others. Plus, it's useful for some Error subclasses to have their own private properties -- and there'd be no way to know whether these should be copied. For these reasons, VError first-classes these information properties. You have to provide them in the constructor, you can only fetch them with the info() function, and VError takes care of making sure properties from causes wind up in the info() output.

Let's put this all together with an example from the node-fast RPC library. node-fast implements a simple RPC protocol for Node programs. There's a server and client interface, and clients make RPC requests to servers. Let's say the server fails with an UnauthorizedError with message "user 'bob' is not authorized". The client wraps all server errors with a FastServerError. The client also wraps all request errors with a FastRequestError that includes the name of the RPC call being made. The result of this failed RPC might look like this:

name: FastRequestError
message: "request failed: server error: user 'bob' is not authorized"
rpcMsgid: <unique identifier for this request>
rpcMethod: GetObject
cause:
    name: FastServerError
    message: "server error: user 'bob' is not authorized"
    cause:
        name: UnauthorizedError
        message: "user 'bob' is not authorized"
        rpcUser: "bob"

When the caller uses VError.info(), the information properties are collapsed so that it looks like this:

message: "request failed: server error: user 'bob' is not authorized"
rpcMsgid: <unique identifier for this request>
rpcMethod: GetObject
rpcUser: "bob"

Taking this apart:

• The error's message is a complete description of the problem. The caller can report this directly to its caller, which can potentially make its way back to an end user (if appropriate). It can also be logged.
• The caller can tell that the request failed on the server, rather than as a result of a client problem (e.g., failure to serialize the request), a transport problem (e.g., failure to connect to the server), or something else (e.g., a timeout). They do this using findCauseByName('FastServerError') rather than checking the name field directly.
• If the caller logs this error, the logs can be analyzed to aggregate errors by cause, by RPC method name, by user, or whatever. Or the error can be correlated with other events for the same rpcMsgid.
• It wasn't very hard for any part of the code to contribute to this Error. Each part of the stack has just a few lines to provide exactly what it knows, with very little boilerplate.

It's not expected that you'd use these complex forms all the time. Despite supporting the complex case above, you can still just do:

new VError("my service isn't working");

for the simple cases.

Examples

The "Demo" section above covers several basic cases. Here's a more advanced case:

const err1 = new VError('something bad happened');
/* ... */
const err2 = new VError({
    'name': 'ConnectionError',
    'cause': err1,
    'info': {
        'errno': 'ECONNREFUSED',
        'remote_ip': '127.0.0.1',
        'port': 215
    }
}, 'failed to connect to "%s:%d"', '127.0.0.1', 215);

console.log(err2.message);
console.log(err2.name);
console.log(VError.info(err2));
console.log(err2.stack);

This outputs:

failed to connect to "127.0.0.1:215": something bad happened
ConnectionError
{ errno: 'ECONNREFUSED', remote_ip: '127.0.0.1', port: 215 }
ConnectionError: failed to connect to "127.0.0.1:215": something bad happened
    at Object.<anonymous> (/home/dap/node-verror/examples/info.js:5:12)
    at Module._compile (module.js:456:26)
    at Object.Module._extensions..js (module.js:474:10)
    at Module.load (module.js:356:32)
    at Function.Module._load (module.js:312:12)
    at Function.Module.runMain (module.js:497:10)
    at startup (node.js:119:16)
    at node.js:935:3

Information properties are inherited up the cause chain, with values at the top of the chain overriding same-named values lower in the chain. To continue that example:

const err3 = new VError({
    'name': 'RequestError',
    'cause': err2,
    'info': {
        'errno': 'EBADREQUEST'
    }
}, 'request failed');

console.log(err3.message);
console.log(err3.name);
console.log(VError.info(err3));
console.log(err3.stack);

This outputs:

request failed: failed to connect to "127.0.0.1:215": something bad happened
RequestError
{ errno: 'EBADREQUEST', remote_ip: '127.0.0.1', port: 215 }
RequestError: request failed: failed to connect to "127.0.0.1:215": something bad happened
    at Object.<anonymous> (/home/dap/node-verror/examples/info.js:20:12)
    at Module._compile (module.js:456:26)
    at Object.Module._extensions..js (module.js:474:10)
    at Module.load (module.js:356:32)
    at Function.Module._load (module.js:312:12)
    at Function.Module.runMain (module.js:497:10)
    at startup (node.js:119:16)
    at node.js:935:3

You can also print the complete stack trace of combined Errors by using VError.fullStack(err).

const err1 = new VError('something bad happened');
/* ... */
const err2 = new VError(err1, 'something really bad happened here');

console.log(VError.fullStack(err2));

This outputs:

VError: something really bad happened here: something bad happened
    at Object.<anonymous> (/home/dap/node-verror/examples/fullStack.js:5:12)
    at Module._compile (module.js:409:26)
    at Object.Module._extensions..js (module.js:416:10)
    at Module.load (module.js:343:32)
    at Function.Module._load (module.js:300:12)
    at Function.Module.runMain (module.js:441:10)
    at startup (node.js:139:18)
    at node.js:968:3
caused by: VError: something bad happened
    at Object.<anonymous> (/home/dap/node-verror/examples/fullStack.js:3:12)
    at Module._compile (module.js:409:26)
    at Object.Module._extensions..js (module.js:416:10)
    at Module.load (module.js:343:32)
    at Function.Module._load (module.js:300:12)
    at Function.Module.runMain (module.js:441:10)
    at startup (node.js:139:18)
    at node.js:968:3

VError.fullStack is also safe to use on regular Errors, so feel free to use it whenever you need to extract the stack trace from an Error, regardless if it's a VError or not.