Skip to content

Latest commit

 

History

History

errors

Errors

To avoid burden of mapping errors returned from 3rd party libraries you can gather all error mappings in one place and put an interceptor provided by atlas-app-toolkit package in a middleware chain as following:

interceptor := errors.UnaryServerInterceptor(
	// List of mappings

	// Base case: simply map error to an error container.
	errors.NewMapping(fmt.Errorf("Some Error"), errors.NewContainer(/* ... */).WithDetail(/* ... */)),
)

Contents

  1. Background
  2. Error Mappers
  3. Usage
  4. Validation Errors
  5. PQ Errors

Error Handling

Background

This document is a brief overview of facilities provided by error handling package. The rationale for implementing it are four noble reasons:

  1. Provide ability to add specific details and field information to an error.
  2. Provide ability to handle multiple errors without returning control to a callee.
  3. Ability to map errors from 3-rd party libraries (gorm, to name one).
  4. Mapping from error to container should be performed automatically in gRPC interceptor.

Error Mappers

Error mapper performs conditional mapping from one error message to another. Error mapping functions are passed to a gRPC Error interceptor and called against error returned from handler.

Currently there are two mappers available:

  • ValidationErrors: error mapper and interceptor for protoc-gen-validate validation errors
  • PQErrors: error mapper for go postgres driver

Error Container

Error container is a data structure that implements Error interface and GRPCStatus method, enabling passing it around as a conventional error from one side and as a protobuf Status to gRPC gateway from the other side.

There are several approaches exist to work with it:

  1. Single error mode
  2. Multiple errors mode

Usage

Single Error Return

This code snippet demonstrates the usage of error container as conventional error:

func validateNameLength(name string) error {
	if len(name) > 255 {
		return errors.NewContainer(
			codes.InvalidArgument, "Name validation error."
		).WithDetail(
			codes.InvalidArgument, "object", "Invalid name length."
		).WithField(
			"name", "Specify name with length less than 255.")
	}

	return nil
}

Gather Multiple Errors

func (svc *Service) validateName(name string) error {
	err := errors.InitContainer()
	if len(name) > 255 {
		err = err.WithDetail(codes.InvalidArgument, "object", "Invalid name length.")
		err = err.WithField("name", "Specify name with length less than 255.")
	}

	if strings.HasPrefix(name, "_") {
		err = err.WithDetail(codes.InvalidArgument, "object", "Invalid name.").WithField(
			"name", "Name cannot start with an underscore")
	}

	return err.IfSet(codes.InvalidArgument, "Invalid name.")
}

To gather multiple errors across several procedures use context functions:

func (svc *Service) globalValidate(ctx context.Context, input *pb.Input) error {
	svc.validateName(ctx, input.Name)
	svc.validateIP(ctx, input.IP)

	// in some particular cases we expect that something really bad
	// should happen, so we can analyse it and throw instead of validation errors.
	if err := validateNamePermExternal(svc.AuthInfo); err != nil {
		return errors.New(ctx, codes.Unauthorized, "Client is not authorized.").
			WithDetails(/* ... */)
	}

	return errors.IfSet(ctx, codes.InvalidArgument, "Overall validation failed.")
	// Alternatively if we want to return the latest errCode/errMessage set instead
	// of overwriting it:
	// return errors.Error(ctx)
}

func (svc *Service) validateName(ctx context.Context, name string) {
	if len(name) > 255 {
		errors.Detail(ctx, codes.InvalidArgument, "object", "Invalid name length.")
		errors.Field(ctx, "name", "Specify name with length less than 255.")
	}
}

func (svc *Service) validateIP(ctx context.Context, ip string) { /* ip validation */ }

Error Mapper

Error mapper performs conditional mapping from one error message to another. Error mapping functions are passed to a gRPC Error interceptor and called against error returned from handler.

Below we demonstrate a cases and customization techniques for mapping functions:

interceptor := errors.UnaryServerInterceptor(
	// List of mappings
	
	// Base case: simply map error to an error container.
	errors.NewMapping(fmt.Errorf("Some Error"), errors.NewContainer(/* ... */).WithDetail(/* ... */)),

	// Extended Condition, mapped if error message contains "fk_contraint" or starts with "pg_sql:"
	// MapFunc calls logger and returns Internal Error, depending on debug mode it could add details.
	errors.NewMapping(
		errors.CondOr(
			errors.CondReMatch("fk_constraint"),
			errors.CondHasPrefix("pg_sql:"),
		),
		errors.MapFunc(func (ctx context.Context, err error) (error, bool) {
			logger.FromContext(ctx).Debug(fmt.Sprintf("DB Error: %v", err))
			err := errors.NewContainer(codes.Internal, "database error")

			if InDebugMode(ctx) {
				err.WithDetail(/* ... */)
			}

			// boolean flag indicates whether the mapping succeeded
			// it can be used to emulate fallthrough behavior (setting false)
			return err, true
		})
	),

	// Skip error
	errors.NewMapping(fmt.Errorf("Error to Skip"), nil)
)

Such model allows us to define our own error classes and map them appropriately as in example below:

// service validation code.

type RequiredFieldErr string
(e RequiredFieldErr) Error() { return string(e) }

func RequiredFieldCond() errors.MapCond {
	return errors.MapCond(func(err error) bool {
		_, ok := err.(RequiredFieldErr)
		return ok
	})
}

func validateReqArgs(in *pb.Input) error {
	if in.Name == "" {
		return RequiredFieldErr("name")
	}

	return nil
}
// interceptor init code
interceptor := errors.UnaryServerInterceptor(
	errors.NewMapping(
		RequiredFieldCond(),
		errors.MapFunc(func(ctx context.Context, err error) (error, bool) {
			return errors.NewContainer(
				codes.InvalidArgument, "Required field missing: %v", err
			).WithField(string(err), "%q argument is required.", string(err)
			), true
		}),
	)
)

Validation Errors

import "github.com/infobloxopen/atlas-app-toolkit/errors/mappers/validationerrors"

validationerrors is a request contents validator server-side middleware for gRPC.

Request Validator Middleware

This middleware checks for the existence of a Validate method on each of the messages of a gRPC request. In case of a validation failure, an InvalidArgument gRPC status is returned, along with the error that caused the validation failure.

It is intended to be used with plugins like https://github.com/lyft/protoc-gen-validate, Go protocol buffers codegen plugins that create the Validate methods (including nested messages) based on declarative options in the .proto files themselves.

func UnaryServerInterceptor

func UnaryServerInterceptor() grpc.UnaryServerInterceptor

UnaryServerInterceptor returns a new unary server interceptor that validates incoming messages and returns a ValidationError.

Invalid messages will be rejected with InvalidArgument and the error before reaching any userspace handlers.

func DefaultMapping

func DefaultMapping() errors.MapFunc

DefaultMapping returns a mapper that parses through the lyft protoc-gen-validate errors and only returns a user friendly error.

Example Usage:

  1. Add validationerrors and errors interceptors to your application:

    errors.UnaryServerInterceptor(ErrorMappings...),
    validationerrors.UnaryServerInterceptor(),
  2. Create an ErrorMapping variable with all your mappings.

  3. Add DefaultMapping as part of your ErrorMapping variable

    var ErrorMappings = []errors.MapFunc{
       // Adding Default Validations Mapping
       validationerrors.DefaultMapping(), 
    
    }

    Example return after DefaultMapping on a invalid email:

    {
        "error": {
            "status": 400,
            "code": "INVALID_ARGUMENT",
            "message": "Invalid primary_email: value must be a valid email address"
        },
        "fields": {
            "primary_email": [
                "value must be a valid email address"
            ]
        }
    }
  4. You can also add custom validation mappings:

    var ErrorMappings = []errors.MapFunc{
        // Adding custom Validation Mapping based on the known field and reason from lyft
       errors.NewMapping(
    		errors.CondAnd(
    			validationerrors.CondValidation(),
                validationerrors.CondFieldEq("primary_email"),
    			validationerrors.CondReasonEq("value must be a valid email address"),
    		),
    		errors.MapFunc(func(ctx context.Context, err error) (error, bool) {
    			vErr, _ := err.(validationerrors.ValidationError)
    			return errors.NewContainer(codes.InvalidArgument, "Custom error message for field: %v reason: %v", vErr.Field, vErr.Reason), true
            }),
       ),
    }

PQ Errors

import "github.com/infobloxopen/atlas-app-toolkit/errors/mappers/pqerrors"

pqerrors is a error mapper for postgres.

Dedicated error mapper for go postgres driver (lib/pq.Error) package is included under the path of github.com/atlas-app-toolkit/errors/mappers/pqerrors. This package includes following components:

  • Condition function CondPQ, CondConstraintEq, CondCodeEq for conditions involved in *pq.Error detection, specific constraint name and specific status code of postgres error respectively.

  • ToMapFunc function that converts mapping function that deals with pq.Error to avoid burden of casting errors back and forth.

  • Default mapping function that can be included into errors interceptor for FK contraints (NewForeignKeyMapping), RESTRICT (NewRestrictMapping), NOT NULL (NewNotNullMapping), PK/UNIQUE (NewUniqueMapping)

Example Usage:

import (
	...
	"github.com/atlas-app-toolkit/errors/mappers/pqerrors"
)

interceptor := errors.UnaryServerInterceptor(
	...
	pqerrors.NewUniqueMapping("emails_address_key", "Contacts", "Primary Email Address"),
	...
)

Any violation of UNIQUE constraint "email_address_key" will result in following error:

{
  "error": {
    "status": 409,
    "code": "ALREADY_EXISTS",
    "message": "There is already an existing 'Contacts' object with the same 'Primary Email Address'."
  }
}