@dotcom-reliability-kit/errors

A suite of error classes which help you throw the most appropriate error in any situation, and identify when errors are known vs unknown. This module is part of FT.com Reliability Kit.

• Usage
  • OperationalError
    • .relatesToSystems
    • .cause
    • .isErrorMarkedAsOperational
  • HttpError
    • Why use this over http-errors?
  • DataStoreError
  • UpstreamServiceError
  • UserInputError
  • BaseError
• Migrating
• Contributing
• License

Usage

Install @dotcom-reliability-kit/errors as a dependency:

1npm install --save @dotcom-reliability-kit/errors

Include in your code:

1import {OperationalError} from '@dotcom-reliability-kit/errors';
2// or
3const {OperationalError} = require('@dotcom-reliability-kit/errors');

This module exports different Error classes which have different jobs. All can be imported in the same way as the example above.

OperationalError

The OperationalError class is the base class for most other error types. "Operational" in this context means "we understand why this error has occurred", so by using this error type you're helping your team to understand when a thrown error is unexpected.

Joyent's Error Handling docs have a good explanation of Operational Errors.

It's always best to use a more specific error, e.g. UpstreamServiceError, if one exists that suits your needs. So review the docs here to find the most suitable error.

OperationalError can work in the same way as a normal error, expecting a message:

1throw new OperationalError('example message');

You can alternatively construct an operational error with a data object. This accepts a code property, which must be set to a unique identifier for the type of error which is occurring, and a message property which contains a human-readable message:

1throw new OperationalError({
2    message: 'example message',
3    code: 'EXAMPLE_CODE'
4});

Error codes are normalized to be uppercase, alphanumeric, and underscore-delimited. Error properties can be accessed like any other property:

1error.message // example message
2error.code // EXAMPLE_CODE

You can also combine these two ways of constructing errors, passing in both a message as well as additional options. This applies to all error types:

1throw new OperationalError('example message', {
2    code: 'EXAMPLE_CODE'
3});

You may also pass additional properties into an error object, these will be collected and stored on a data property on the error:

1const error = new OperationalError({
2    message: 'example message',
3    code: 'EXAMPLE_CODE',
4    article: 'd92acacb-ac53-4505-aa88-eae4b42de994'
5});
6
7error.data.article // d92acacb-ac53-4505-aa88-eae4b42de994

OperationalError.relatesToSystems

The relatesToSystems property of an operational error stores a list of FT systems which are related to the error that you're throwing.

This array could include:

• dependencies which have returned an HTTP error status code
• data stores which haven't provided the expected data

OperationalError.cause

The cause property of an operational error stores the root cause error instance, e.g. an error that has been caught as part of a try/catch block. It allows the operational error to include the diagnostic information captured by the root cause error.

OperationalError.isErrorMarkedAsOperational()

You can test whether an error is operational (known about) either by using the isErrorMarkedAsOperational method. It accepts an error object of any kind and will return true if that error has a truthy isOperational property and false otherwise:

1OperationalError.isErrorMarkedAsOperational(new OperationalError('example message')); // true
2OperationalError.isErrorMarkedAsOperational(new Error('example message')); // false

HttpError

The HttpError class extends OperationalError and represents an HTTP error status. It can work in the same way as a normal error, expecting a message. In this case it will represent an HTTP 500:

1throw new HttpError('example message');

You can alternatively construct an HTTP error with a data object. This accepts a statusCode property, which is a valid HTTP status code number, as well as all of the properties you can set in OperationalError:

1throw new HttpError({
2    message: 'your thing was not found',
3    statusCode: 404
4});

It's also possible to create an HTTP error with a status code alone, which will default the message to the corresponding HTTP status message:

1throw new HttpError(404);

Error properties can be accessed like any other property:

1error.message // your thing was not found
2error.statusCode // 404
3error.status // 404
4error.statusMessage // Not Found
5error.code // HTTP_404

Why use this over http-errors?

The benefit of using this error rather than the excellent http-errors library is that we extend OperationalError by default. This means that all HTTP errors you throw are considered "known errors" by the rest of our tooling. We also set a code property by default which results in less code in our monitoring dashboards – we don't need to check both code and statusCode properties to determine the type of error thrown.

DataStoreError

The DataStoreError class extends OperationalError and represents an error which occurred while accessing a data store, e.g. MongoDB, PostgreSQL, or Redis. It has all of the features of operational errors, you can construct with just a message:

1throw new DataStoreError('Could not connect to Redis');

You can alternatively construct a data store error with a data object. You can use any of the properties defined in OperationalError:

1throw new DataStoreError({
2    code: 'REDIS_CONNECTION_FAILED',
3    message: 'Could not connect to Redis',
4});

UpstreamServiceError

The UpstreamServiceError class extends HttpError and represents an error which occurred while connecting to an upstream service, e.g. an FT or third-party API. It has all of the features of operational and HTTP errors, you can construct with just a message:

1throw new UpstreamServiceError('Content could not be fetched');

The HTTP status code defaults to a 502. This indicates that while connecting to an upstream service, your system has received a response that it cannot serve to an end user.

You can alternatively construct an upstream service error with a data object. You can use any of the properties defined in HttpError and OperationalError:

1throw new UpstreamServiceError({
2    code: 'CONTENT_PIPELINE_FAILED',
3    message: 'Content could not be fetched, the content pipeline is responding with a 503 status',
4    statusCode: 503,
5    relatesToSystems: ['cp-content-pipeline-graphql']
6});

UserInputError

The UserInputError class extends HttpError and represents an error which occurred based on invalid user input, e.g. they inputted a malformed email address into a form. It has all of the features of operational and HTTP errors but defaults to a 400 status code. You can construct with just a message:

1throw new UserInputError('An invalid email address was input');

You can alternatively construct a user input error with a data object. You can use any of the properties defined in HttpError and OperationalError:

1throw new UserInputError({
2    code: 'REGISTRATION_INFO_INVALID',
3    message: 'An invalid email address was input'
4});

BaseError

Our getting started guide talks about Operational vs Programmer errors. The rest of the errors in this library provide operational errors, but they all extend a base class named BaseError. This class offers up some of the conveniences of OperationalError but it's marked as non-operational:

1const error = new BaseError('This is an error');
2error.isOperational === false; // true;

This allows you to use Reliability Kit errors in library code, where it doesn't make sense to throw an operational error. The code, message, and cause properties work in the same way as an OperationalError:

1throw new BaseError({
2    message: 'example message',
3    code: 'EXAMPLE_CODE',
4    cause: new TypeError('example cause')
5});

Migrating

Consult the Migration Guide if you're trying to migrate to a later major version of this package.

Contributing

See the central contributing guide for Reliability Kit.

License

Licensed under the MIT license.
Copyright © 2022, The Financial Times Ltd.