users@jax-rs-spec.java.net

[jax-rs-spec users] Re: High Level API for SSE support

From: Pavel Bucek <pavel.bucek_at_oracle.com>
Date: Fri, 3 Feb 2017 11:16:26 +0100

Hi Marcus, Sergey,

thanks for thinking about this topic and coming with an enhancement (or
maybe extension is a better name).

I believe the idea is a sound one. I also share Sergeys opinion that
JAX-RS should provide SSE support on the level which is currently proposed.

I know you won't like this, but.. can you please formalize your proposal
as a jira ticket?

We need to keep "closing" stuff instead of adding and I we don't even
have any comprehensive review of the proposed SSE yet. If we'd include
this as part of it, it would most likely significantly prolong the
review period, not mentioning that it can be treated as completely
separate feature. We can call it "High level SSE API". Also, the idea
will need some work - currently, there is no mention of the message
format (I believe it will need to be pluggable, we don't want to define
any "standard" there), relation to the existing API (esp. broadcaster), etc.

Having said that, please don't take it as "I don't want to hear your
thoughts" - it's exactly the opposite. But I believe there is a general
conclusion: "we want to have API similar to what has been proposed by
JAX-RS Spec leads". Let's finish that one. We can build on top of it later.

Thanks and regards,
Pavel


On 03/02/2017 10:41, Sergey Beryozkin wrote:
> I think the proposed SSE API should stay (to let users control SSE
> flows in a fine grained manner when needed),
>
> However offering the users an option to cover the basic SSE flows with
> the use of annotations is interesting which
> if accepted should be complementary IMHO
>
> Cheers, Sergey
>
> On 02/02/17 23:22, Markus KARG wrote:
>>
>> Experts,
>>
>> Oracle's current SSE API proposal clearly is a good API for SSE. The
>> question is, whether it really fulfils what application developers
>> expert from JAX-RS in particular.
>>
>> There might be programmers that don't like to explicitly learn a new
>> (and rather complex) set of classes and methods just to get the
>> additional benefit of update notifications, and that the style of the
>> SSE API does not very well fit into the style of JAX-RS. SSE API is
>> completely algorithmic, while server-side JAX-RS is mostly
>> declarative. I could imagine that some programmers would love to get
>> a high-level SSE support in JAX-RS that more looks like "SSE inside
>> JAX-RS" and less like a standalone SSE API.
>>
>> To illustrate, let me give an example. Think of a simple CRUD
>> application that works pretty well in JAX-RS 2.0 already (@POST /
>> @GET / @PUT / @DELETE). Now the developers decide that once data is
>> updated (PUT) or deleted (DELETE), the client shall be notified about
>> that immediately using SSE technology with minimal additional code.
>> But the amount of additional code lines with the new SSE API is
>> really heavy and clutters the previously clean application with lots
>> of SSE special code.
>>
>> So I wonder whether it might be beneficial to provide some high-level
>> API that simplifies this use case?
>>
>> For example, it could look like this (simplified for illustration
>> purposes):
>>
>> @SSE public class MyResource {
>>
>> @GET @SseInit public void notifications() {};
>>
>> @POST public void create() {};
>>
>> @GET @Path("{id}") public MyObject read() {};
>>
>> @PUT @SseNotify public void update() {};
>>
>> @DELETE @SseNotify public void delete() {};
>>
>> }
>>
>> The idea would be the SSE API as outlined by Oracle is used under the
>> hood by the JAX-RS container. Once a requests to get SSE
>> notifications by calling the @SseInit-annotated GETter, the JAX-RS
>> container sets up an implied SSE subscription for that client with
>> this resource. Whenever the JAX-RS container leaves the body of
>> @SseNotify-annotated methods, it pushes notifications to that
>> subscriptions (the messages as synthetically built from the HTTP
>> method and the URL, so the client knows what the message means).
>>
>> This is not as far as flexible as Oracle's full-blown SSE API, but it
>> is only intended as sugar ontop for simple but frequent use cases.
>>
>> -Markus
>>
>