520 likes | 1.03k Views
REST. INF 123 – Software architecture tdebeauv@uci.edu. Outline. SOAP and REST REST constraints and gains REST guidelines. Simple Object Access Protocol Representational State Transfer. SOAP and REST. From RPC to SOAP: RPC. Pass remote procedure name and arguments
E N D
REST INF 123 – Software architecture tdebeauv@uci.edu
Outline • SOAP and REST • REST constraints and gains • REST guidelines
Simple Object Access Protocol Representational State Transfer SOAP and REST
From RPC to SOAP: RPC • Pass remote procedure name and arguments • Expect a return value • Procedure signature is implicit • If you change the signature, people have no way to know why it’s not working anymore • It’s about (function) names • Whereas REST is about standard verbs like GET, POST, etc.
From RPC to SOAP: SOAP • 1998: Simple Object Access Protocol • 2000:Web Service Description Language (WSDL) uses SOAP as underlying protocol, cf SOA slides • SOAP = XML RPC done “right” • Long/verbose structured XML messages • Verbosity != metadata
SOAP message POST /InStock HTTP/1.1 Host: www.example.org Content-Type: application/soap+xml; charset=utf-8 Content-Length: 299 SOAPAction: "http://www.w3.org/2003/05/soap-envelope" <?xml version="1.0"?> <soap:Envelopexmlns:soap="http://www.w3.org/2003/05/soap-envelope"> <soap:Header></soap:Header> <soap:Body> <m:GetStockPricexmlns:m="http://www.example.org/stock"> <m:StockName>IBM</m:StockName> </m:GetStockPrice> </soap:Body> </soap:Envelope>
Metadata vs verbosity There are two kinds of pain. The sort of pain that makes you strong, or useless pain that is only suffering.
Metadata vs verbosity Pay attention to the fine printmetadata. It’s far more important than the selling pricedata itself.
Amazon uses both REST and SOAP • Amazon has both SOAP and REST interfaces to their web services, and 85% of their usage is of the REST interface (2003) • http://oreilly.com/pub/wlg/3005
REST and SOAP • They happened concurrently • SOAP: 1998 • REST: 2000 • SOAP = envelope, REST = postcard • SOAP derived from RPC • REST is not a move against SOAP • REST is very complex • But it looks simpler than SOAP
REST History • Hypertext • 1945 memex (VannevarBush) • 1967 hypertext (Project Xanadu) • Internet • 1969 ARPANET (army) • 1992: WWW = Internet + hypertext • 2000: REST = reverse-engineer/document the WWW architectural style • HTTP is not mandatory for REST, but it helps
SOAP vs REST • SOAP is verbose: large overhead of metadata and boilerplate text • SOAP • REST <?xml version="1.0"?> <soap:Envelopexmlns:soap="http://www.w3.org/2001/12/soap-envelope" soap:encodingStyle="http://www.w3.org/2001/12/soap-encoding"> <soap:bodypb="http://www.acme.com/phonebook"> <pb:GetUserDetails> <pb:UserID>12345</pb:UserID> </pb:GetUserDetails> </soap:Body> </soap:Envelope> GET http://www.acme.com/phonebook/UserDetails/12345
Solutions to SOAP’s verbosity • MTOM: Message Transmission Optimization Mechanism • Encode/compress XML into binary • XOP: XML-binary Optimized Packaging • To encode/decode MTOM • TLDR: Binary-encoded XML over HTTP • But HTTP = hypertext transfer protocol
Representational State Transfer REST constraints
Reminder: Architectural style • Set of constraints • Constraints induce properties • Desirable or undesirable • Design trade-offs • REST = architectural style of the WWW
Resource = information, data We have great resources at our disposal.
REST is made of several styles • Client-Server • Stateless • Cache • Layered • Code on demand • Uniform Interface • These styles are not always inter-compatible • But they are in the case of the WWW
Deriving REST from other styles No style http://roy.gbiv.com/talks/200804_REST_ApacheCon.pdf
1 Client-server • Cf week 2 • Separation of concerns: client display vs server logic • Display is client-side: clients can have different UIs • Each website has a server • Same client (browser) can access multiple servers
2 Stateless interactions • AKA context-free interactions • Stateless interaction does not mean “no data in the server” • The server does not store any client-specific information between two requests • State is client-side or in a database
Stateful example: SMTP tagus: crista$ telnet smtp.ics.uci.edu 25 Trying 128.195.1.219... Connected to smtp.ics.uci.edu. Escape character is '^]'. 220 david-tennant-v0.ics.uci.edu ESMTP mailer ready at Mon, 5 Apr 2010 17:15:01 -0700' HELO smtp.ics.uci.edu 250 david-tennant-v0.ics.uci.edu Hello barbara-wright.ics.uci.edu [128.195.1.137], pleased to meet you MAIL FROM:<lopes@ics.uci.edu> 250 2.1.0 <lopes@ics.uci.edu>... Sender ok RCPT TO:<lopes@ics.uci.edu> 250 2.1.5 <lopes@ics.uci.edu>... Recipient ok DATA 354 Enter mail, end with "." on a line by itself test . 250 2.0.0 o360F1Mo029280 Message accepted for delivery QUIT 221 2.0.0 david-tennant-v0.ics.uci.edu closing connection Connection closed by foreign host.
Gains from statelessness • Immune to server restart/migration • Server restart = lose all data • But stateless = no data to lose! • No server affinity • Client requests can be processed by ANY server, not just one particular server • Scalability • Server never knows if/when client sends its next request • So stateful servers timeout the sessions of clients with long inter-request times • Stateless servers don’t have any memory management issues
Losses from statelessness • Client is less efficient • Server needs to pull data for every request • Pulling data is straightforward when this data is a static web page (most of the WWW in the 90s) • How do you authenticate users? • Cookies (not good?) • External auth and directory services
3 Caching • Optional • Store data locally so I don’t have to retrieve it • In clients, in servers, or in intermediaries (cf layered constraint) • Reduces latency • Improves efficiency and scalability • But degrades reliability (stale data)
4 Layered • Intermediaries between client and server • Proxies, such as nginx • Caches • Content Delivery Network, such as Akamai • Web accelerator, such as CloudFlare • Pros: ability to balance load (improves scalability), can reduce latency (when cache hits) • Cons: can add latency (when cache misses)
Do you know what I like about peopleintermediaries? They stack so neatly.
Intermezzo: ISPs • Internet Service Providers (ISP): COX, Comcast Verizon, UCI • When a client requests content, it goes through: • The client’s ISP • Intermediary ISP 1 • Intermediary ISP 2 • The server’s ISP • The server delivers the content • Many hops = lots of delay
Content Delivery Network: Akamai • Akamai pays ISPs to host their servers within a few hops of many clients • Many clients = urban areas • I pay Akamai to deliver my content • Now, when a client requests my content: • Client ISP • Akamai server delivers my content!
5 Code on demand • Optional • Fetch JavaScript when a web page asks for it <html><head><script src=‘blabla.js’></script> • Pros: thinner clients, improves extensibility • Cons: reduces visibility
I won't be a slave to anybody or anything you can order with a toll free numberany static code.
6 Uniform interface • The hardest constraint to get right • Uniform identification of resources • Manipulation of resources via representations • Hypermedia as the engine of app state (HATEOAS)
Uniform Interfaces REST Data Elements (The following slides are from Crista Lopes)
Uniform Interfaces Resources and their identifiers • Resource = Abstraction of information, any information that can be named • A document, a temporal service (“today’s weather”), collection of other resources • Some are static, some are dynamic • Identifiers: Universal Resource Identifiers (URIs)
Uniform Interfaces Representations • Server returns representations of resources, not the resources themselves. • E.g. HTML, XML • Server response contains all metadata for client to interpret the representation
Uniform Interfaces HATEOAS Create Account Question is: Where is the clients’ state stored? LoggedOut Logged In User Change Account Logged In Admin Search Users … Hypermedia As The Engine Of Application State Idea: the application is a state machine
HATEOAS LoggedOut Logged In User Change Account • Non-REST • Clients’ state kept on the server • Server is both state machine and holder of state • REST • State machine on the server • At any step, client is sent a complete “picture” of where it can go next, ie its state and transitions
HATEOAS LoggedOut Logged In User Change Account • Server sends representation of the client’s state back to the client • Hence, REpresentionalState Transfer • Server does not “hold on” to client’s state • Possible next state transitions of the client are encoded in Hypermedia • Anchors, forms, scripted actions, …
HTTP Operations Idempotent methods: the side effects of many invocations are exactly the same as the side effects of one invocation PS: remember main and subroutines? GET PUT DELETE HEAD OPTIONS TRACE POST CONNECT
Example: Paypal’s API • https://developer.paypal.com/docs/api/
RESTful Design Guidelines • Embrace hypermedia • Name your resources/features with URIs • Design your namespace carefully • Hide mechanisms • Bad: http://example.com/cgi-bin/users.pl?name=John • Good: http://example.com/users/John • Serve POST, GET, PUT, DELETE on those resources • Nearly equivalent to CRUD (Create, Retrieve, Update, Delete) • Don’t hold on to state • Serve and forget (functional programming-y) • Consider serving multiple representations • HTML, XML, JSON
RESTful Design Guidelines https://eee.uci.edu/toolbox/dropbox/index.php?op=createdropboxform http://us.mc510.mail.yahoo.com/mc/welcome?.gx=1&.tm=1271634041& .rand=9anflcttvlh7n#_pg=showFolder&fid=Inbox&order=down&tt=237&pSize=100& .rand=952814729&.jsrand=4316826 URIs are nouns The 8 HTTP operations are verbs Very different from CGI-inspired web programming: Many/most web frameworks promote URIs as verbs and query data as nouns – old CGI model.
Choosing moneyCGI over powerREST is a mistake almost everyone makes. They just don’t know …
RESTful Design Guidelines Canonical example
REST vs Linked Data • Linked data • A data model • Proposed by Berners-Lee • REST • An interaction model • Proposed by Fielding
Taylor’s REST principles • Any information is a resource, named by an URL. (uniform interface) • Resource representation is accompanied by metadata about the representation. (uniform interface, code on demand) • Interactions are context-free. (stateless) • Small set of methods. Each method can be applied to any resource. The result of a method is a representation. • Idempotent operations help caching. (cache) • Intermediaries use metadata from requests or responses to filter, redirect, or modify representations. This is transparent to client and server. (layered, cache)