Railway Oriented Programming: a functional approach to error handling

Many examples in functional programming assume that you are always on the "happy path". But to create a robust real world application you must deal with validation, logging, network and service errors, and other annoyances.

So, how do you handle all this in a clean functional way? This talk will provide a brief introduction to this topic, using a fun and easy-to-understand railway analogy.

Code, links to video, etc., at http://fsharpforfunandprofit.com/rop

Transcript

1. Railway Oriented Programming A functional approach to error handling Scott Wlaschin @ScottWlaschin fsharpforfunandprofit.com FPbridge.co.uk ...but OCaml and Haskell are very similar. Examples will be in F#...
2. Overview Topics covered: • Happy path programming • Straying from the happy path • Introducing "Railway Oriented Programming" • Using the model in practice • Extending and improving the design
3. Happy path programming Implementing a simple use case
4. A simple use case Receive request Validate and canonicalize request Update existing user record Send verification email Return result to user type Request = { userId: int; name: string; email: string } "As a user I want to update my name and email address"
5. Imperative code string ExecuteUseCase() { var request = receiveRequest(); validateRequest(request); canonicalizeEmail(request); db.updateDbFromRequest(request); smtpServer.sendEmail(request.Email); return "Success"; }
6. Functional flow let executeUseCase = receiveRequest >> validateRequest >> canonicalizeEmail >> updateDbFromRequest >> sendEmail >> returnMessage F# left-to-right composition operator
7. Straying from the happy path... What do you do when something goes wrong?
8. Straying from the happy path
9. “A program is a spell cast over a computer, turning input into error messages”
10. Straying from the happy path Name is blank Email not valid Receive request Validate and canonicalize request Update existing user record Send verification email Return result to user User not found Db error Authorization error Timeout "As a user I want to update my name and email address" type Request = { userId: int; name: string; email: string } - and see sensible error messages when something goes wrong!
11. Imperative code with error cases string ExecuteUseCase() { var request = receiveRequest(); var isValidated = validateRequest(request); if (!isValidated) { return "Request is not valid" } canonicalizeEmail(request); try { var result = db.updateDbFromRequest(request); if (!result) { return "Customer record not found" } } catch { return "DB error: Customer record not updated" } if (!smtpServer.sendEmail(request.Email)) { log.Error "Customer email not sent" } return "OK"; }
12. Request/response (non-functional) design Request Response Validate Update Send Request handling service Request Response Validate Update Send Request handling service Request Errors Response Validate Update Send Request handling service Imperative code can return early
13. Data flow (functional) design Response Validate Update Send A single function representing the use case Request Request Response Validate Update Send A single function representing the use case Request Errors Success Response Validate Update Send Error Response A single function representing the use case Q: How can you bypass downstream functions when an error happens?
14. Functional design How can a function have more than one output? type Result = | Success | ValidationError | UpdateError | SmtpError
15. Functional design How can a function have more than one output? type Result = | Success | Failure
16. Functional design How can a function have more than one output? type Result<'TEntity> = | Success of 'TEntity | Failure of string
17. Functional design Request Errors Success Validate Update Send Failure A single function representing the use case • Each use case will be equivalent to a single function • The function will return a sum type with two cases: "Success" and "Failure". • The use case function will be built from a series of smaller functions, each representing one step in a data flow. • The errors from each step will be combined into a single "failure" path.
18. How do I work with errors in a functional way?
19. Monad dialog
38. v
54. Monads are confusing
55. Railway oriented programming This has absolutely nothing to do with monads.
56. A railway track analogy The Tunnel of Transformation Function pineapple -> apple
57. A railway track analogy Function 1 pineapple -> apple Function 2 apple -> banana
58. A railway track analogy Function 1 pineapple -> apple Function 2 apple -> banana >>
59. A railway track analogy New Function 3 pineapple -> banana Can't tell it was built from smaller functions!
60. An error generating function Request Success Validate Failure let validateInput input = if input.name = "" then Failure "Name must not be blank" else if input.email = "" then Failure "Email must not be blank" else Success input // happy path
61. Introducing switches Success! Failure Input ->
62. Connecting switches Validate UpdateDb on success bypass
63. Connecting switches Validate UpdateDb
64. Connecting switches Validate UpdateDb SendEmail
65. Connecting switches Validate UpdateDb SendEmail
66. The two-track model in practice
67. Composing switches Validate UpdateDb SendEmail
68. Composing switches Validate UpdateDb SendEmail
69. Composing switches Validate UpdateDb SendEmail >> >> Composing one-track functions is fine...
70. Composing switches Validate UpdateDb SendEmail >> >> ... and composing two-track functions is fine...
71. Composing switches Validate UpdateDb SendEmail   ... but composing switches is not allowed!
72. Composing switches Validate Two-track input Two-track input Validate One-track input Two-track input  
73. Bind as an adapter block Two-track input Slot for switch function Two-track output
74. Bind as an adapter block Two-track input Two-track output Validate Validate
75. Bind as an adapter block Two-track input Two-track output let bind switchFunction = fun twoTrackInput -> match twoTrackInput with | Success s -> switchFunction s | Failure f -> Failure f bind : ('a -> Result<'b>) -> Result<'a> -> Result<'b>
76. Bind as an adapter block Two-track input Two-track output let bind switchFunction twoTrackInput = match twoTrackInput with | Success s -> switchFunction s | Failure f -> Failure f bind : ('a -> Result<'b>) -> Result<'a> -> Result<'b>
77. name50 Bind example let nameNotBlank input = if input.name = "" then Failure "Name must not be blank" else Success input let name50 input = if input.name.Length > 50 then Failure "Name must not be longer than 50 chars" else Success input let emailNotBlank input = if input.email = "" then Failure "Email must not be blank" else Success input nameNotBlank emailNotBlank
78. Bind example nameNotBlank (combined with) name50 (combined with) emailNotBlank nameNotBlank name50 emailNotBlank
79. Bind example bind nameNotBlank bind name50 bind emailNotBlank nameNotBlank name50 emailNotBlank
80. Bind example bind nameNotBlank >> bind name50 >> bind emailNotBlank nameNotBlank name50 emailNotBlank
81. Bind example let validateRequest = bind nameNotBlank >> bind name50 >> bind emailNotBlank // validateRequest : Result<Request> -> Result<Request> validateRequest
82. Bind example let (>>=) twoTrackInput switchFunction = bind switchFunction twoTrackInput let validateRequest twoTrackInput = twoTrackInput >>= nameNotBlank >>= name50 >>= emailNotBlank validateRequest
83. Bind doesn't stop transformations FunctionB type Result<'TEntity> = | Success of 'TEntity | Failure of string FunctionA
84. Composing switches - review Validate UpdateDb SendEmail Validate UpdateDb SendEmail
85. Comic Interlude What do you call a train that eats toffee? I don't know, what do you call a train that eats toffee? A chew, chew train!
86. More fun with railway tracks... ...extending the framework
87. More fun with railway tracks... Fitting other functions into this framework: • Single track functions • Dead-end functions • Functions that throw exceptions • Supervisory functions
88. Converting one-track functions Fitting other functions into this framework: • Single track functions • Dead-end functions • Functions that throw exceptions • Supervisory functions
89. Converting one-track functions // trim spaces and lowercase let canonicalizeEmail input = { input with email = input.email.Trim().ToLower() } canonicalizeEmail
90. Converting one-track functions UpdateDb SendEmail Validate canonicalizeEmail  Won't compose
91. Converting one-track functions Two-track input Slot for one-track function Two-track output
92. Converting one-track functions Two-track input Two-track output Canonicalize Canonicalize
93. Converting one-track functions Two-track input Two-track output let map singleTrackFunction twoTrackInput = match twoTrackInput with | Success s -> Success (singleTrackFunction s) | Failure f -> Failure f map : ('a -> 'b) -> Result<'a> -> Result<'b> Single track function
94. Converting one-track functions Two-track input Two-track output let map singleTrackFunction = bind (singleTrackFunction >> Success) map : ('a -> 'b) -> Result<'a> -> Result<'b> Single track function
95. Converting one-track functions UpdateDb SendEmail Validate canonicalizeEmail  Will compose
96. Converting dead-end functions Fitting other functions into this framework: • Single track functions • Dead-end functions • Functions that throw exceptions • Supervisory functions
97. Converting dead-end functions let updateDb request = // do something // return nothing at all updateDb
98. Converting dead-end functions SendEmail Validate UpdateDb  Won't compose
99. Converting dead-end functions One-track input Slot for dead end function One-track output
100. Converting dead-end functions One-track input One-track output let tee deadEndFunction oneTrackInput = deadEndFunction oneTrackInput oneTrackInput tee : ('a -> unit) -> 'a -> 'a Dead end function
101. Converting dead-end functions SendEmail Validate UpdateDb  Will compose
102. Functions that throw exceptions Fitting other functions into this framework: • Single track functions • Dead-end functions • Functions that throw exceptions • Supervisory functions
103. Functions that throw exceptions One-track input Two-track output SendEmail SendEmail Add try/catch to handle timeouts, say Looks innocent, but might throw an exception
104. Functions that throw exceptions Even Yoda recommends not to use exception handling for control flow: Guideline: Convert exceptions into Failures "Do or do not, there is no try".
105. Supervisory functions Fitting other functions into this framework: • Single track functions • Dead-end functions • Functions that throw exceptions • Supervisory functions
106. Supervisory functions Two-track input Two-track output Slot for one-track function for Success case Slot for one-track function for Failure case
107. Putting it all together
108. Putting it all together Validate UpdateDb SendEmail Canonicalize Input Output??
109. Putting it all together Validate UpdateDb SendEmail Canonicalize returnMessage Input Output let returnMessage result = match result with | Success _ -> "Success" | Failure msg -> msg
110. Putting it all together - review The "two-track" framework is a useful approach for most use-cases. You can fit most functions into this model.
111. Putting it all together - review The "two-track" framework is a useful approach for most use-cases. let executeUseCase = receiveRequest >> validateRequest >> updateDbFromRequest >> sendEmail >> returnMessage let executeUseCase = receiveRequest >> validateRequest >> updateDbFromRequest >> sendEmail >> returnMessage Let's look at the code -- before and after adding error handling
112. Comic Interlude Why can't a steam locomotive sit down? I don't know, why can't a steam locomotive sit down? Because it has a tender behind!
113. Designing for errors Unhappy paths are requirements too
114. Designing for errors let validateInput input = if input.name = "" then Failure "Name must not be blank" else if input.email = "" then Failure "Email must not be blank" else Success input // happy path type Result<'TEntity> = | Success of 'TEntity | Failure of string Using strings is not good
115. Designing for errors let validateInput input = if input.name = "" then Failure NameMustNotBeBlank else if input.email = "" then Failure EmailMustNotBeBlank else Success input // happy path type Result<'TEntity> = | Success of 'TEntity | Failure of ErrorMessage type ErrorMessage = | NameMustNotBeBlank | EmailMustNotBeBlank Special type rather than string
116. Designing for errors let validateInput input = if input.name = "" then Failure NameMustNotBeBlank else if input.email = "" then Failure EmailMustNotBeBlank else if (input.email doesn't match regex) then Failure EmailNotValid input.email else Success input // happy path type ErrorMessage = | NameMustNotBeBlank | EmailMustNotBeBlank | EmailNotValid of EmailAddress Add invalid email as data
117. Designing for errors type ErrorMessage = | NameMustNotBeBlank | EmailMustNotBeBlank | EmailNotValid of EmailAddress // database errors | UserIdNotValid of UserId | DbUserNotFoundError of UserId | DbTimeout of ConnectionString | DbConcurrencyError | DbAuthorizationError of ConnectionString * Credentials // SMTP errors | SmtpTimeout of SmtpConnection | SmtpBadRecipient of EmailAddress Documentation of everything that can go wrong -- And it's type-safe documentation that can't go out of date!
118. Designing for errors – service boundaries Translation function needed at a service boundary type DbErrorMessage<'PK> = | PrimaryKeyNotValid of 'PK | RecordNotFoundError of 'PK | DbTimeout of ConnectionString * TimeoutMs | DbConcurrencyError | DbAuthorizationError of Credentials type MyUseCaseError = | NameMustNotBeBlank | EmailMustNotBeBlank | EmailNotValid of EmailAddress // database errors | UserIdNotValid of UserId | DbUserNotFoundError of UserId | DbTimeout of ConnectionString | DbConcurrencyError | DbAuthorizationError of Credentials // SMTP errors | SmtpTimeout of SmtpConnection | SmtpBadRecipient of EmailAddress let dbResultToMyResult dbError = match dbError with | DbErrorMessage.PrimaryKeyNotValid id -> MyUseCaseError.UserIdNotValid id | DbErrorMessage.RecordNotFoundError id -> MyUseCaseError.DbUserNotFoundError id | _ -> // etc
119. Designing for errors – converting to strings No longer works – each case must now be explicitly converted to a string returnMessage let returnMessage result = match result with | Success _ -> "Success" | Failure msg -> msg
120. Designing for errors – converting to strings let returnMessage result = match result with | Success _ -> "Success" | Failure err -> match err with | NameMustNotBeBlank -> "Name must not be blank" | EmailMustNotBeBlank -> "Email must not be blank" | EmailNotValid (EmailAddress email) -> sprintf "Email %s is not valid" email // database errors | UserIdNotValid (UserId id) -> sprintf "User id %i is not a valid user id" id | DbUserNotFoundError (UserId id) -> sprintf "User id %i was not found in the database" id | DbTimeout (_,TimeoutMs ms) -> sprintf "Could not connect to database within %i ms" ms | DbConcurrencyError -> sprintf "Another user has modified the record. Please resubmit" | DbAuthorizationError _ -> sprintf "You do not have permission to access the database" // SMTP errors | SmtpTimeout (_,TimeoutMs ms) -> sprintf "Could not connect to SMTP server within %i ms" ms | SmtpBadRecipient (EmailAddress email) -> sprintf "The email %s is not a valid recipient" email Each case must be converted to a string – but this is only needed once, and only at the last step. All strings are in one place, so translations are easier. returnMessage (or use resource file)
121. Parallel tracks
122. Parallel validation nameNotBlank name50 emailNotBlank Problem: Validation done in series. So only one error at a time is returned
123. Parallel validation nameNotBlank name50 emailNotBlank Split input Combine output Now we do get all errors at once! ... But how to combine?
124. Combining switches + Trick: if we create an operation that combines pairs into a new switch, we can repeat to combine as many switches as we like.
125. Combining switches + + Trick: if we create an operation that combines pairs into a new switch, we can repeat to combine as many switches as we like.
126. Combining switches + Success (S2) Failure (F2) Success (S1) S1 or S2 F2 Failure (F1) F1 [F1; F2]
127. Combining switches + Success (S2) Failure (F2) Success (S1) S1 or S2 F2 Failure (F1) F1 [F1; F2] Either input is OK, they are both the same value
128. Combining switches + Success (S2) Failure (F2) Success (S1) S1 or S2 F2 Failure (F1) F1 [F1; F2] type Result<'TEntity> = | Success of 'TEntity | Failure of ErrorMessage list
129. Combining switches + Success (S2) Failure (F2) Success (S1) S1 or S2 [F2] Failure (F1) [F1] [F1; F2] type Result<'TEntity> = | Success of 'TEntity | Failure of ErrorMessage list
130. Handling lists of errors let errToString err = match err with | NameMustNotBeBlank -> "Name must not be blank" | EmailMustNotBeBlank -> "Email must not be blank" // etc returnMessage Collapse a list of strings into a single string Convert all messages to strings let returnMessage result = match result with | Success _ -> "Success" | Failure errs -> errs |> List.map errToString |> List.reduce (fun s1 s2 -> s1 + ";" + s2)
131. Domain events Communicating information to downstream functions
132. Events are not errors Validate UpdateDb SendEmail Tell CRM that email was sent
133. Events are not errors Validate UpdateDb SendEmail Tell CRM that email was sent type MyUseCaseMessage = | NameMustNotBeBlank | EmailMustNotBeBlank | EmailNotValid of EmailAddress // database errors | UserIdNotValid of UserId // SMTP errors | SmtpTimeout of SmtpConnection // Domain events | UserSaved of AuditInfo | EmailSent of EmailAddress * MsgId
134. Events are not errors Validate UpdateDb SendEmail Tell CRM that email was sent type MyUseCaseMessage = | NameMustNotBeBlank | EmailMustNotBeBlank | EmailNotValid of EmailAddress // database errors | UserIdNotValid of UserId // SMTP errors | SmtpTimeout of SmtpConnection // Domain events | UserSaved of AuditInfo | EmailSent of EmailAddress * MsgId type Result<'TEntity> = | Success of 'TEntity * Message list | Failure of Message list
135. Comic Interlude Why can't a train driver be electrocuted? I don't know, why can't a train driver be electrocuted? Because he's not a conductor!
136. Summary A recipe for handling errors in a functional way
137. Recipe for handling errors in a functional way type Result<'TEntity> = | Success of 'TEntity * Message list | Failure of Message list Validate UpdateDb SendEmail type Message = | NameMustNotBeBlank | EmailMustNotBeBlank | EmailNotValid of EmailAddress
138. Some topics not covered... ... but could be handled in an obvious way.
139. Topics not covered • Async on success path (instead of sync) • Compensating transactions (instead of two phase commit) • Logging (tracing, app events, etc.)
140. I don’t always have errors...
141. I don’t always have errors... Railway Oriented Programming @ScottWlaschin fsharpforfunandprofit.com FPbridge.co.uk