I agree with Pavel. The suggested approach is indeed utilized quite
frequently, but it's inherently error-prone.
The main issue is that it creates implicit assumptions about the behavior
of both the service and the user's code. For example, if the user's code
must provide a username, what if it doesn't? I assume it will get an error,
which is very counterintuitive. Even more importantly, how should one learn
about this requirement in the first place? It is not reflected in the API
in any way - and that's a big problem.
The fact that the service implementor needs to update the API methods when
such requirements are introduced is actually a good thing, in my opinion.
This forces the developer to stop and think about how the updated API
should look like and how to make sure it's backward-compatible (or not, in
case the new requirements are mandatory). Doing this through an external
context is basically the equivalent of saying "let the end user deal with
this". Not a good practice, in my view.
Conversely, passing everything exclusively via method arguments guarantees
that:
- The user's code is always compliant with the service contract. You
can't "forget" to pass something to the service.
- Any changes in the service contract (backward-compatible or otherwise)
are explicitly reflected in the API.
-Val
On Sun, Oct 10, 2021 at 6:21 AM Pavel Tupitsyn <ptupit...@apache.org> wrote:
> Ivan,
>
> Yes, this approach is used by some other systems, and still, I don't like
> it very much.
> Let's hear more opinions.
>
> On Sat, Oct 9, 2021 at 9:00 PM Ivan Daschinsky <ivanda...@gmail.com>
> wrote:
>
> > Hi.
> > Pavel T., Ok, http rest dosn't have the clean design, in your opinion.
> >
> > But what about grpc? The same?
> >
> > As for me, it is ok to pass additional parameters as list of key-value
> > pairs with keys as strings and values as bytearrays or strings. It is ok
> to
> > allow user to set up middlewares for services and allow to enrich request
> > context in this middlewares. It is very common approach everywhere and is
> > very useful in distributed systems. The use cases are so obvious, aren't
> > they?
> >
> >
> >
> > сб, 9 окт. 2021 г., 20:14 Pavel Tupitsyn <ptupit...@apache.org>:
> >
> > > Pavel,
> > >
> > > Thanks for the explanation, I understand the use cases.
> > >
> > > > in REST service, he can set such parameters in request headers
> > >
> > > I don't consider HTTP-based services as a good example of a
> > > clean architecture.
> > > Data can be passed in URL parameters, in headers, and in body, and each
> > of
> > > those ways has its own limitations.
> > > There is no obvious correct way to do things.
> > >
> > > >> Ambient state is not obvious and the API looks confusing even
> though I
> > > understand our services stack quite well both in Java and .NET
> > > > Can you clarify please?
> > >
> > > The proposed API adds a "side channel" for the data.
> > > Some is passed as arguments, which is obvious, and some becomes
> magically
> > > available on the server side through some external context.
> > > - You have to know about the context
> > > - You have to understand that the context is only available during the
> > > method call (can't use it in some background logic)
> > >
> > > In my opinion, this is a bit too clever. I'm a fan of the functional
> > > programming approach where everything you need is passed as arguments.
> > >
> > >
> > >
> > > On Fri, Oct 8, 2021 at 4:29 PM Pavel Pereslegin <xxt...@gmail.com>
> > wrote:
> > >
> > > > Igor, Pavel.
> > > >
> > > > > Why can not a user implement such context on application level? I
> > > > believe Ignite provides all necessary tools for that.
> > > > The user wants to trace the source of the service call. For example,
> a
> > > > service must log the name of the user who made the calls of the
> > > > service. For now, there's no possibility to do that without modifying
> > > > the service interface and implementation. Moreover, the user must
> > > > modify all methods of service to pass this parameter. For example, in
> > > > REST service, he can set such parameters in request headers, why we
> > > > can't provide such usability in Ignite.
> > > >
> > > > > This will reduce the performance of all calls
> > > > This feature is optional, if the context is not passed - then there's
> > > > shouldn't be any performance difference.
> > > >
> > > > > Ambient state is not obvious and the API looks confusing even
> though
> > I
> > > > understand our services stack quite well both in Java and .NET
> > > > Can you clarify please?
> > > >
> > > > пт, 8 окт. 2021 г. в 15:46, Pavel Tupitsyn <ptupit...@apache.org>:
> > > > >
> > > > > Agree with Igor.
> > > > >
> > > > > I'm not sure this feature is a good fit for Ignite.
> > > > > Ignite should not be responsible for such a high-level concept,
> this
> > > > should
> > > > > be on the application side instead.
> > > > >
> > > > > - As Eduard noted, it is hard to make this type-safe
> > > > > - Ambient state is not obvious and the API looks confusing even
> > though
> > > I
> > > > > understand our services stack quite well both in Java and .NET
> > > > > - This will reduce the performance of all calls
> > > > >
> > > > > On Fri, Oct 8, 2021 at 3:44 PM Igor Sapego <isap...@apache.org>
> > wrote:
> > > > >
> > > > > > Hi guys,
> > > > > >
> > > > > > Why can not a user implement such context on application level?
> > > > > > I believe Ignite provides all necessary tools for that. User can
> > just
> > > > > > implement such a context as user type and pass it to services
> they
> > > > > > need. Are the arguments why would Ignite need a separate feature
> > > > > > for such a use case?
> > > > > >
> > > > > > Best Regards,
> > > > > > Igor
> > > > > >
> > > > > >
> > > > > > On Fri, Oct 8, 2021 at 2:17 PM Eduard Rakhmankulov <
> > > > erixon...@gmail.com>
> > > > > > wrote:
> > > > > >
> > > > > > > I am not aware .NET capabilities, but as I can see service must
> > be
> > > > > > > implemented in *java* and even if can't serialize other that
> Map
> > on
> > > > .NET
> > > > > > > side, on java side we can wrap this map with provided
> > TypedContext
> > > > > > (context
> > > > > > > should be convertible from map in this case).
> > > > > > > That leads to a situation when Java can use TypedContext but
> > other
> > > > > > clients
> > > > > > > can't. I believe that the majority of services users are using
> > Java
> > > > and
> > > > > > it
> > > > > > > should be taken in accordance.
> > > > > > >
> > > > > > > P.S. I think it is possible to send plain objects from .NET
> > context
> > > > to
> > > > > > > cluster.
> > > > > > >
> > > > > > > Best regards, Ed
> > > > > > >
> > > > > > > On Fri, 8 Oct 2021 at 14:40, Pavel Pereslegin <
> xxt...@gmail.com>
> > > > wrote:
> > > > > > >
> > > > > > > > Hi, Eduard!
> > > > > > > >
> > > > > > > > Thanks for your feedback.
> > > > > > > >
> > > > > > > > The idea sounds very good, but don't forget about the
> platform
> > > > > > services.
> > > > > > > > For example, we may call Java service from .Net and
> vice-versa.
> > > I'm
> > > > > > > > not sure if the context can be implemented as a custom class
> > > > (instead
> > > > > > > > of Map/Dictionary) in this case.
> > > > > > > >
> > > > > > > > пт, 8 окт. 2021 г. в 14:21, Eduard Rakhmankulov <
> > > > erixon...@gmail.com>:
> > > > > > > > >
> > > > > > > > > Hi, Pavel
> > > > > > > > >
> > > > > > > > > Is it possible to provide type-safe API for
> > > ServiceProxyContext ?
> > > > > > > > > I think constructions like int arg1 =
> ctx.attribute("arg1");
> > > are
> > > > > > error
> > > > > > > > > prone.
> > > > > > > > >
> > > > > > > > > Can we make something like this :
> > > > > > > > >
> > > > > > > > > //Signature with two generic params which allow the
> compiler
> > to
> > > > check
> > > > > > > > > if the service will be called with the wrong type context.
> > > > > > > > >
> > > > > > > > > public <T extends ContextedWith<CtxType>, CtxType> T
> > > > > > > > > serviceProxyTyped(ClusterGroup prj, String name, Class<?
> > super
> > > T
> > > > >
> > > > > > > > > srvcCls, CtxType optCtx, boolean sticky, long timeout)
> > > > > > > > >
> > > > > > > > > //new interface which services with scoped context should
> > > > implement
> > > > > > > > >
> > > > > > > > > public interface ContextedWith<T> {
> > > > > > > > > T getCtx();
> > > > > > > > > }
> > > > > > > > >
> > > > > > > > > // implementation can delegate to Map-like context or be
> > POJO.
> > > > > > > > > interface MyServiceContext {
> > > > > > > > > int getArg1();
> > > > > > > > > String getUserId();
> > > > > > > > > }
> > > > > > > > >
> > > > > > > > > class MyService implements ContextedWith<MyServiceContext>
> {
> > > > > > > > > void doThings() {
> > > > > > > > > MyServiceContext ctx = getCtx();
> > > > > > > > >
> > > > > > > > > System.out.println("ctx.getArg1() = " + ctx.getArg1());
> > > > > > > > > }
> > > > > > > > >
> > > > > > > > > @Override public MyServiceContext getCtx() {
> > > > > > > > > return ServiceProxyContext.current();
> > > > > > > > > }
> > > > > > > > > }
> > > > > > > > >
> > > > > > > > > WDYT?
> > > > > > > > >
> > > > > > > > > Best regards, Ed.
> > > > > > > > >
> > > > > > > > > On Fri, 8 Oct 2021 at 13:26, Pavel Pereslegin <
> > > xxt...@gmail.com>
> > > > > > > wrote:
> > > > > > > > >
> > > > > > > > > > Hello Igniters!
> > > > > > > > > >
> > > > > > > > > > I want to implement a feature to support a custom
> "caller"
> > > > context
> > > > > > in
> > > > > > > > > > ignite services (see example in ticket description [1]).
> > > > > > > > > >
> > > > > > > > > > Sometimes, when using Ignite services, it becomes
> necessary
> > > to
> > > > pass
> > > > > > > > > > custom parameters from the "request source" to the
> service.
> > > > This is
> > > > > > > > > > most commonly used to track the origin of a service call
> > > (user
> > > > id,
> > > > > > > > > > request id, session id eg see this user question [2]).
> > > > > > > > > > At the moment, the only way to pass such parameters to a
> > > > service is
> > > > > > > by
> > > > > > > > > > adding argument(s) to all called methods of the service,
> > > which
> > > > > > makes
> > > > > > > > > > the code messy and also complicates development and
> > > > maintenance.
> > > > > > > > > >
> > > > > > > > > > I propose letting the user set a custom context for the
> > > service
> > > > > > proxy
> > > > > > > > > > and implicitly pass that context to the methods being
> > called.
> > > > This
> > > > > > > > > > function should not affect the execution of service
> methods
> > > in
> > > > any
> > > > > > > way
> > > > > > > > > > unless the user has specified a context.
> > > > > > > > > >
> > > > > > > > > > An example of using the proposed API [1].
> > > > > > > > > > PoC (except thin clients) [3].
> > > > > > > > > >
> > > > > > > > > > WDYT?
> > > > > > > > > >
> > > > > > > > > > [1] https://issues.apache.org/jira/browse/IGNITE-15572
> > > > > > > > > > [2]
> > > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > > >
> > > >
> > >
> >
> https://stackoverflow.com/questions/57459071/apache-ignite-service-grid-service-call-context
> > > > > > > > > > [3] https://github.com/apache/ignite/pull/9440
> > > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > > >
> > > >
> > >
> >
>
