Skip to main content

How to handle errors

What we want

When users of Milmove make a mistake, we need a way to guide them back to the happy path. This means returning not only coherent response codes for our front-end to consume, but more helpful detail around what went wrong. This will become increasingly important as we need to support users that will be using Milmove programmatically as opposed to using our own front-end.

How we're currently doing it

We currently have two ways of generating error responses in our APIs:

  1. Using the generated Swagger response methods that conform to our API contract as defined in our Swagger yaml file.
  2. Using handlers.ResponseForError and handlers.ResponseForVErrors, convenience functions which allow us to pass in any errors that are returned to us from model or service code. They take on the responsibility of deciding which error code to return based on the error values we pass in.

The approach in #2 is ergonomic, but makes it easy for our error handling to drift from the API spec since the convenience functions write their own response headers.

Defining error models in Swagger

We typically define error responses in our Swagger definition like this:

400:
description: invalid request
401:
description: request requires user authentication
404:
description: office not found
500:
description: server error

This will tell go-swagger to generate error response methods for each respective status code, but we can take it further in order to reach our goal of giving more robust, detailed responses to API consumers.

Both Swagger 2.0 and 3.0 allow us to define error models like the following:

definitions:
ClientError:
type: object
properties:
title:
type: string
detail:
type: string
instance:
type: string
format: uuid
required:
- title
- detail
- instance
ValidationError:
allOf:
- $ref: '#/definitions/ClientError'
- type: object
properties:
invalid_fields:
type: object
additionalProperties:
type: string
required:
- invalid_fields

With this approach, we can use generated code to add any data we'd like to an error response.

RFC #7807: Problem details for HTTP APIs

This RFC proposes some interesting ways of standardizing a way of providing better descriptions for API errors. We will use the concepts of an error title, detail, and extension fields in the rest of this document.

422 vs. 400

We currently return a 400 Bad Request for validation errors. It is recommended that we use 422 Unprocessable Entitity instead.

Note: If we choose to specify a format validation in our yaml documentation, Swagger will return its own 422's with a different format than the errors we return with title, detail, and instance fields. In these cases, we need to make sure the client can handle both cases.

Error Titles

The title property for the ClientError should be a constant string that is used for all errors of that type. All custom messaging should be in the detail property, or within the messages for the fields in invalidFields. This helps increase consistency with our error messaging and ensures that the client always knows where to look for the pertinent information about the error.

Example setups

Validation errors

# ...
post:
summary: create an office user
description: creates and returns an office user record
operationId: createOfficeUser
tags:
- office
parameters:
- in: body
name: officeUser
description: Office user information
schema:
$ref: '#/definitions/OfficeUserCreatePayload'
responses:
201:
description: Successfully created Office User
schema:
$ref: '#/definitions/OfficeUser'
422:
description: validation error
schema:
$ref: '#/definitions/ValidationError' #=> the interesting part
500:
description: internal server error
func (h CreateOfficeUserHandler) Handle(params officeuserop.CreateOfficeUserParams) middleware.Responder {
// ...

createdOfficeUser, verrs, err := h.OfficeUserCreator.CreateOfficeUser(&officeUser, transportationIDFilter)
if verrs != nil {
payload := &adminmessages.ValidationError{
InvalidFields: handlers.NewValidationErrorsResponse(verrs).Errors,
}

payload.Title = handlers.FmtString(handlers.ValidationErrMessage)
payload.Detail = handlers.FmtString("The information you provided is invalid.")
payload.Instance = handlers.FmtUUID(h.GetTraceID())

return officeuserop.NewCreateOfficeUserUnprocessableEntity().WithPayload(payload)
}

if err != nil {
return officeuserop.NewCreateOfficeUserInternalServerError()
}

returnPayload := payloadForOfficeUserModel(*createdOfficeUser)
return officeuserop.NewCreateOfficeUserCreated().WithPayload(returnPayload)
}

Move is not in a state to be approved

/moves/{moveId}/submit:
post:
summary: Submits a move for approval
description: Submits a move for approval by the office. The status of the move will be updated to SUBMITTED
operationId: submitMoveForApproval
tags:
- moves
parameters:
- name: moveId
in: path
type: string
format: uuid
required: true
description: UUID of the move
- name: submitMoveForApprovalPayload
in: body
required: true
schema:
$ref: '#/definitions/SubmitMoveForApprovalPayload'
responses:
200:
description: returns updated (submitted) move object
schema:
$ref: '#/definitions/MovePayload'
400:
description: invalid request
401:
description: must be authenticated to use this endpoint
403:
description: not authorized to approve this move
409:
description: the move is not in a state to be approved
schema:
$ref: '#/definitions/ClientError' #=> the interesting part
500:
description: server error
func (h SubmitMoveHandler) Handle(params moveop.SubmitMoveForApprovalParams) middleware.Responder {
// ...

submitDate := time.Time(*params.SubmitMoveForApprovalPayload.PpmSubmitDate)
err = move.Submit(submitDate)
if err != nil {
payload := &internalmessages.ClientError{
Title: handlers.FmtString(handlers.ConflictErrMessage),
Detail: handlers.FmtString("This move is not in a state to be approved - make sure the move is in state x before attempting to approve."),
Instance: handlers.FmtUUID(h.GetTraceID()),
}

return moveop.NewSubmitMoveForApprovalConflict().WithPayload(payload)
}

// ...
}