Upload
anirudh-bhatnagar
View
292
Download
1
Tags:
Embed Size (px)
Citation preview
Modelling RESTful applications – Why should I not use verbs in REST urlAnirudh BhatnagarXebia India IT Architects Pvt Ltd
REST
Representational state transfer ??
Specification ???
Guidelines ???
Architecture style??
Its Just a STYLE!
The REST architectural style was developed by W3C Technical Architecture Group (TAG) in parallel with HTTP/1.1, based on the existing design of HTTP/1.0.
Roy Fielding’s paper on REST architectural style
Architectural Styles and the Design of Network-based Software Architectures
DISSERTATIONsubmitted in partial satisfaction of the requirements for the degree ofDOCTOR OF PHILOSOPHYin Information and Computer SciencebyRoy Thomas Fielding2000
REST, HTTP and Web
HTTP Specification
GET ../foods/1 would get you the food with id 1.
PUT ../foods/2 would update the food with id 2.
POST ../foods will add a new food.
DELETE ../foods/1 will delete the food resource with id 1.
But what about other verbs???
--- approve-- reject-- cancel-- search -- increase-- decrease…..…..
Can I just put them in my URL??what would go wrong?
Lets see a video
http://www.youtube.com/watch?v=65ilZ8esAUs
Improper handling of method and no safeguard
Result : Disaster!!!
How do we solve this?
HTTP Specifications
1. Safe and Unsafe Methodsexcerpt from w3 HTTP Specification…
…..GET and HEAD methods SHOULD NOT have the significance of taking an action other than retrieval. These methods ought to be considered "safe". This allows user agents to represent other methods, such as POST, PUT and DELETE, in a special way, so that the user is made aware of the fact that a possibly unsafe action is being requested....
GET and HEAD should have no side-effects : They should not change the state.
Clearly we can not use GET in our API..
2.IdempotencyAn idempotent HTTP method can be called many times without different outcomes.
a = 4 ->idempotenta++ -> non idempotent
...Methods can also have the property of "idempotence" in that (aside from error or expiration issues) the side-effects of N > 0 identical requests is the same as for a single request. The methods GET, HEAD, PUT and DELETE share this property…..POST is non-idempotent..
Fault Tolerant - Idempotent Methods - if request timed out - can you safely retry? - no need to worry if idempotent
Browser SupportConfirm Form submission On refresh of Unsafe Method (POST form)
Caching
Non-safe and non-idempotent methods will never be cached by any middleware proxies.
Safe Methods like GET and HEAD are candidate for caching.
caching with GETEvery GET call is a candidate for caching..
If you have method :
HTTP/1.1 GET …./users/1/update
This might not actually update and return you the cached result.
Bad Urls hamper caching!HTTP 1.1 GET
http://myTestApp/page4.do?dau22.oid=5199&UserCtxParam=0&GroupCtxParam=0&dctx1=25&ctx1=US&crc=712082047
HTTP Caching
Browser Caches
Proxy Cacheexample : Squid
Gateway Cache : Reverse Proxy
Benefits of HTTP Caching
- Server side caching is expensive..
- Reduce latency
- Reduce network traffic
-CDNs can leverage proxy caches.
Leverage Caching effectivelyWith great power comes great responsibility...
How to control caching effectively?Invalidations?Cache expiry?Stale cache?Volatile data?
- expires
- cache control -Etags -last modified
- validation headers
HTTP headers
Expires Header
● HTTP 1.0
So, if we made an API call to retrieve data
……….. GET /users/1
The response header would be:
HTTP/1.1 200 OK
Content-Type: application/xml
Expires: Tue, 25 Aug 2013 16:00 GMT
-----
<user id="1">...</users>
JAX-RS support for expires..@Path("{id}") @GET @Produces(MediaType.APPLICATION_XML) public Response getUserXML(@PathParam("id") Long id){ User user = userDB.get(id); ResponseBuilder builder = Response.ok(user,MediaType.APPLICATION_XML); //Putting expires header for HTTP browser caching. Calendar cal = Calendar.getInstance(); cal.set(2013,7,25,16,0); builder.expires(cal.getTime()); return builder.build(); }
HTTP 1.1
support CDNs, proxy caches and revalidations there was a need for more enhanced headers with richer set of features, having more explicit controls.
Cache-ControlCache-Control has a variable set of comma-delimited directives that define who,how and for how long it can be cached. Lets explore few of them:
-private/public : these are accessibility directives, private means a browser can cache the object but the proxies or CDNs can not and public makes it cachable by all.-no-cache,no-store,max-age are few others where name tells the story.
JAX-RS support for Cache-Control@Path("{id}") @GET @Produces(MediaType.APPLICATION_XML) public Response getUserXMLwithCacheControl(@PathParam("id") Long id){ User user = userDB.get(id); CacheControl cc = new CacheControl(); cc.setMaxAge(300); cc.setNoStore(true); cc.setPrivate(true); ResponseBuilder builder = Response.ok(user,MediaType.APPLICATION_XML); builder.cacheControl(cc); return builder.build(); }
Validation Headers and Conditional GETs
When cache is stale, client can ask server if cache still valid
To be able to revalidate client needs additional headersbeyond Cache-Control from a server response
•Last-Modified - a date when the resource was last modified•ETag - a unique hash-like key that identifies a version of the resource
Client should cache these headers along with response body
To revalidate client sends conditional GETs using values of these header tags.
Last-Modified and If-Modified-Since Server sends in response header
HTTP/1.1 200 OK....Cache-Control: max-age=1000Last-Modified: Mon, 19 aug 2013 16:00 IST
Client revalidates using conditional GET
GET /users/23 HTTP/1.1If-Modified-Since: Mon, 19 aug 2013 16:00 IST
in case it is modified after this date; a response code 200 (OK) with current value of resource will be sent.
And if the data is not modified a response code of “304″
Etag and If-None-Match● an MD5 hash value.● generated from resource is sent by server in
response.● client caches it and uses this to revalidate using If-
None-Match tag in request header. GET /users/23 HTTP/1.1
If-None-Match: "23432423423454654667444"
Server verifies the hash, if it matches sends “304” else sends current value with response code 200 and resets the etag.
JAX-RS support Validation
JAX-RS also provided one injectable helper class Request, which has methods like…
....ResponseBuilder evalutatePostConditions(EntityTag eTag);ResponseBuilder evaluatePreConditions(Date isLastModified);.....
And...JAX-RS provides us with javax.ws.rs.core.EntityTag for the same
The values sent by client (which they have cached) are compared with latest values at the server.
JAX-RS and Validation@Path("{id}")@GET@Produces(MediaType.APPLICATION_XML)public Response getUserWithEtagSupport(@PathParam("id") Long id, @Context Request request){ User user = userDB.get(id); //generating Etag out of hashCode of user EntityTag tag = new EntityTag(Integer.toString(user.hashCode())); CacheControl cc = new CacheControl(); cc.setMaxAge(1000); ResponseBuilder builder = request.evaluatePreconditions(tag); if(builder!=null){ //means the preconditions have been met and the cache is valid //we just need to reset the cachecontrol max age (optional) builder.cacheControl(cc); return builder.build(); } //preconditions are not met and the cache is invalid //need to send new value with response code 200 (OK) builder = Response.ok(user,MediaType.APPLICATION_XML); //reset cache control and eTag (mandatory) builder.cacheControl(cc); builder.tag(tag); return builder.build(); }
HTTP PURGE
HTTP has an unofficial PURGE method that is used for purging caches.
When an API receives a call with an unsafe method on a resource, it should fire a PURGE request on that resource so that the reverse proxy knows that the cached resource should be expired.
We dont need to perform explicit revalidations in this case.
GET /article/1234 HTTP/1.1 - - The resource is not cached yet - Send request to the API - Store response in cache and return
GET /article/1234 HTTP/1.1 - The resource is cached - Return response from cache
PUT /article/1234 HTTP/1.1 - Unsafe method, send to API
PURGE /article/1234 HTTP/1.1 - API sends PURGE method to the cache
- The resources is removed from the cache GET /article/1234 HTTP/1.1 - The resource is not cached yet - Send request to the API - - Store response in cache and return
Let’s complete our “pitaji ki patloon” problem
GET
-No side effects- should not change the state-idempotent
HTTP1.1 GET /pitaji/patloon/12/length?method=decrease&size=1b
Caching will not work!
PUT
- idempotent
- HTTP1.1 PUT /pitaji/patloon/12/length {“decrease” : “1 bilaank” }
This will result in disaster, as the browser can call the PUT multiple times, in case of timeouts/network latency etc.
DELETE
HTTP/1.1 DELETE /pitaji/patloon/12/length {“decrease” : “1 bilaank” }
this API does not make sense, it will confuse the client!moreover again performing unsafe operation with safe method.
POST
Unsafe method
HTTP1.1 POST /pitaji/patloon/length {“decrease” : “1 bilaank” }
Use Case
An example of a social Site :
1.) Add friend2.) Remove Friend3.) Approve Friend Request4.) Reject Friend Request5.) Make a new account6.) Delete account.7.) Search Users.…...
Approach 1 : userFriendMapping table@Entity@Table(name = "userFriendMapping")public class UserFriendMapping { private long id; private User user; private User friend; private String status; @Id @GeneratedValue(strategy = IDENTITY) @Column(name = "id", unique = true, nullable = false) public long getId() { return id; } @ManyToOne(fetch = FetchType.LAZY) @JoinColumn(name="userId") public User getUser() { ……
Add and Approve friend request
1. Add a friend (send friendRequest)POST ../userfriendmapping{userId: 1,friendId : 2,status:pending}
2. Approve friend RequestPOST ../userfriendmapping{userId: 1,friendId : 2,status:approved}
reject friend, get pending requests
3. Reject friend RequestDELETE ../userfriendmapping/1
4.Get pending friendsGET ../userfriendmapping/users/1?status=pending
5. Delete existing FriendDELETE ../userfriendmapping/2
More extensions
List all friend requestsList all pending friends..List all friends..List all rejected requests..Do not allow a user to resend the friend request..BlackList UsersIgnore a friend request
ProblemsSingle domain catering to responsibilty of two states :
1.) FriendRequest2.) UserFriendRelation
Increases complexity, more effort, tightly coupled, separation of concern?
1. Separate domains give more flexibility and ease for extensibility.
2. As we have states and resources as domains, making RESTful urls is easy.
3. Querying is easy.
example :- to find friends need 2 calls to DB, or put a UNION
API : Find all myfriends@Override public List<User> findFriends(Long userId,String status) { List<UserFriendMapping> allFriends = userFriendMappingPersistence.getAllFriends(userId,status); List<UserFriendMapping> friendsWhoAddedMe = userFriendMappingPersistence.getByFriendId(userId,status); List<User> friends = new ArrayList<User>(); for (UserFriendMapping userFriendMapping : allFriends) { friends.add(userFriendMapping.getFriend()); } for (UserFriendMapping userFriendMapping : friendsWhoAddedMe) { friends.add(userFriendMapping.getUser()); } return friends; }
Resource Oriented ArchitectureA resource-oriented architecture is the structural design supporting the internetworking of resources.
A resource, in this context, is any entity that can be identified and assigned a uniform resource identifier (URI).
any states , verbs which acts as a resource can be made model like FriendRequest or BookOrder.
Alternate Approach Model driven Architecture and Resource Driven Architecture.provides intuitive way of designing APIs in RESTful manner.Add 2 domain classes ● FriendRequest● UserFriend or FriendShip or RelationThe RESTful APIs :1. add FriendPOST ../users/1/friendrequests?friendid=2
@Path("/users/{id}/friendrequests") @POST public String createFriendRequest(@PathParam("id") Long userId, @QueryParam(value="friendid")Long friendId){ …...
Approve and Reject friendRequest
2. Approve: POST .. /userfriends/friendrequests/22 -> creating a new friend from friendRequest with id22
3.RejectDELETE ../friendrequests/22
4.Remove a friendDELETE ../userfriends/3
5. GET on ..users/2/friendrequests will give all pending friend requests6. GET on ..users/1/userfriends/ will give all friends of user
Search UsersSearch is GETUSE GET with QUERY PARAMSHTTP1.1 GET ../users?firstname=abc&age=25
Versioning APIs in RESTAdd version in URLGET ../version/users/1
Example twitter: GET https://api.twitter.com/1/statuses/user_timeline.json
Use HTTP Redirection Response codes for versioning
● 301 Moved permanently - point to new URL
● 302 Found indicating that the requested resource temporarily is located at another location, while requested URI may still supported.
Model Driven Design produces RESTful Urls
RAD tools which generate code like Spring ROO or Rails/Grails.
These are made on top of domains and models.
Take business domains from framework to other.More extensibility and portability.
and of course they provide RESTful URLs.
Finally, Is it Just to avoid verbs and have better Urls?
The approach should be the other way :
Better modelling and better design gives way to better URLs and cleaner approach.
ConclusionREST is no specification, its a style which adheres to HTTP specification.
So, in order to make full use of HTTP and REST
--- Better modelling will automatically avoid verbs. --- Take care of idempotent and safe/unsafe methods. --- Use cache-control headers to make best use of caching.
Thanks!!!
Questions and Feedback.
twitter : anirudh_bhblog : http://anirudhbhatnagar.commail : [email protected]: https://github.com/anirudh83
Referenceswww.w3.org/Protocols/rfc2616/rfc2616-sec13.htmlhttp://www.w3.org/Protocols/rfc2616/rfc2616-sec9.htmlhttp://javasymposium.techtarget.com/html/images/BBurke_Scaling_JAX-RS.pdfhttp://restcookbook.com/Basics/caching/http://www.w3.org/Protocols/rfc2616/rfc2616-sec13.htmlhttp://www.squid-cache.org/http://odino.org/rest-better-http-cache/