Hi all,
So I think people have made some very good points so far. There seems
to be agreement that we need to have explicit batch APIs for the sake of
efficiency, so I added that back.
Contexts seem a little more complex than we thought, so I removed that
from the proposal.
I removed the Impl class. Instead, we now have a KafkaAdminClient
interface and an AdminClient implementation. I think this matches our
other code better, as Jay commented.
Each call now has an "Options" object that is passed in. This will
allow us to easily add new parameters to the calls without having tons
of function overloads. Similarly, each call now has a Results object,
which will let us easily extend the results we are returning if needed.
Many people made the point that Java 7 Futures are not that useful, but
Java 8 CompletableFutures are. With CompletableFutures, you can chain
calls, adapt them, join them-- basically all the stuff people are doing
in Node.js and Twisted Python. Java 7 Futures don't really let you do
anything but poll for a value or block. So I felt that it was better to
just go with a CompletableFuture-based API.
People also made the point that they would like an easy API for waiting
on complete success of a batch call. For example, an API that would
fail if even one topic wasn't created in createTopics. So I came up
with Result objects that provide multiple futures that you can wait on.
You can wait on a future that fires when everything is complete, or you
can wait on futures for individual topics being created.
I updated the wiki, so please take a look. Note that this new API
requires JDK8. It seems like JDK8 is coming soon, though, and the
disadvantages of sticking to Java 7 are pretty big here, I think.
best,
Colin
On Mon, Feb 13, 2017, at 11:51, Colin McCabe wrote:
> On Sun, Feb 12, 2017, at 09:21, Jay Kreps wrote:
> > Hey Colin,
> >
> > Thanks for the hard work on this. I know going back and forth on APIs is
> > kind of frustrating but we're at the point where these things live long
> > enough and are used by enough people that it is worth the pain. I'm sure
> > it'll come down in the right place eventually. A couple things I've found
> > helped in the past:
> >
> > 1. The burden of evidence needs to fall on the complicator. i.e. if
> > person X thinks the api should be async they need to produce a set of
> > common use cases that require this. Otherwise you are perpetually
> > having to
> > think "we might need x". I think it is good to have a rule of "simple
> > until
> > proven insufficient".
> > 2. Make sure we frame things for the intended audience. At this point
> > our apis get used by a very broad set of Java engineers. This is a
> > very
> > different audience from our developer mailing list. These people code
> > for a
> > living not necessarily as a passion, and may not understand details of
> > the
> > internals of our system or even basic things like multi-threaded
> > programming. I don't think this means we want to dumb things down, but
> > rather try really hard to make things truly simple when possible.
> >
> > Okay here were a couple of comments:
> >
> > 1. Conceptually what is a TopicContext? I think it means something
> > like
> > TopicAdmin? It is not literally context about Topics right? What is
> > the
> > relationship of Contexts to clients? Is there a threadsafety
> > difference?
> > Would be nice to not have to think about this, this is what I mean by
> > "conceptual weight". We introduce a new concept that is a bit nebulous
> > that
> > I have to figure out to use what could be a simple api. I'm sure
> > you've
> > been through this experience before where you have these various
> > objects
> > and you're trying to figure out what they represent (the connection to
> > the
> > server? the information to create a connection? a request session?).
>
> The intention was to provide some grouping of methods, and also a place
> to put request parameters which were often set to defaults rather than
> being explicitly set. If it seems complex, we can certainly get rid of
> it.
>
> > 2. We've tried to avoid the Impl naming convention. In general the
> > rule
> > has been if there is only going to be one implementation you don't
> > need an
> > interface. If there will be multiple, distinguish it from the others.
> > The
> > other clients follow this pattern: Producer, KafkaProducer,
> > MockProducer;
> > Consumer, KafkaConsumer, MockConsumer.
>
> Good point. Let's change the interface to KafkaAdminClient, and the
> implementation to AdminClient.
>
> > 3. We generally don't use setters or getters as a naming convention. I
> > personally think mutating the setting in place seems kind of like late
> > 90s
> > Java style. I think it likely has thread-safety issues. i.e. even if
> > it is
> > volatile you may not get the value you just set if there is another
> > thread... I actually really liked what you described as your original
> > idea
> > of having a single parameter object like CreateTopicRequest that holds
> > all
> > these parameters and defaults. This lets you evolve the api with all
> > the
> > various combinations of arguments without overloading insanity. After
> > doing
> > literally tens of thousands of remote APIs at LinkedIn we eventually
> > converged on a rule, which is ultimately every remote api needs a
> > single
> > argument object you can add to over time and it must be batched. Which
> > brings me to my next point...
>
> Just to clarify, volatiles were never a part of the proposal. I think
> that context objects or request objects should be used by a single
> thread at a time.
>
> I'm not opposed to request objects, but I think they raise all the same
> questions as context objects. Basically, the thread-safety issues need
> to be spelled out and understood by the user, and the user needs more
> lines of code to make a request. And there will be people trying to do
> things like re-use request objects when they should not, and so forth.
>
> > 4. I agree batch apis are annoying but I suspect we'll end up adding
> > one. Doing 1000 requests for 1000 operations if creating or deleting
> > will
> > be bad, right? This won't be the common case, but when you do it it
> > will be
> > a deal-breaker problem. I don't think we should try to fix this one
> > behind
> > the scenes.
> > 5. Are we going to do CompletableFuture (which requires java 8) or
> > normal Future? Normal Future is utterly useless for most things other
> > than
> > just calling wait. If we can evolve in place from Future to
> > CompletableFuture that is fantastic (we could do it for the producer
> > too!).
> > My belief was that this was binary incompatible but I actually don't
> > know
> > (obviously it's source compatible).
>
> In my testing, replacing a return type with a subclass of that return
> type did not break binary compatibility. I haven't been able to find
> chapter and verse on this from the Java implementers, though.
>
> best,
> Colin
>
>
> >
> > -Jay
> >
> > On Wed, Feb 8, 2017 at 5:00 PM, Colin McCabe <cmcc...@apache.org> wrote:
> >
> > > Hi all,
> > >
> > > I made some major revisions to the proposal on the wiki, so please check
> > > it out.
> > >
> > > The new API is based on Ismael's suggestion of grouping related APIs.
> > > There is only one layer of grouping. I think that it's actually pretty
> > > intuitive. It's also based on the idea of using Futures, which several
> > > people commented that they'd like to see.
> > >
> > > Here's a simple example:
> > >
> > > > AdminClient client = new AdminClientImpl(myConfig);
> > > > try {
> > > > client.topics().create("foo", 3, (short) 2, false).get();
> > > > Collection<String> topicNames = client.topics().list(false).get();
> > > > log.info("Found topics: {}", Utils.mkString(topicNames, ", "));
> > > > Collection<Node> nodes = client.nodes().list().get();
> > > > log.info("Found cluster nodes: {}", Utils.mkString(nodes, ", "));
> > > > } finally {
> > > > client.close();
> > > > }
> > >
> > > The good thing is, there is no Try, no 'get' prefixes, no messing with
> > > batch APIs. If there is an error, then Future#get() throws an
> > > ExecutionException which wraps the relevant exception in the standard
> > > Java way.
> > >
> > > Here's a slightly less simple example:
> > >
> > > > AdminClient client = new AdminClientImpl(myConfig);
> > > > try {
> > > > List<Future<Void>> futures = new LinkedList<>();
> > > > for (String topicName: myNewTopicNames) {
> > > > creations.add(client.topics().
> > > > setClientTimeout(30000).setCreationConfig(myTopicConfig).
> > > > create(topicName, 3, (short) 2, false));
> > > > }
> > > > Futures.waitForAll(futures);
> > > > } finally {
> > > > client.close();
> > > > }
> > >
> > > I went with Futures because I feel like ought to have some option for
> > > doing async. It's a style of programming that has become a lot more
> > > popular with the rise of Node.js, Twisted python, etc. etc. Also, as
> > > Ismael commented, Java 8 CompletableFuture is going to make Java's
> > > support for fluent async programming a lot stronger by allowing call
> > > chaining and much more.
> > >
> > > If we are going to support async, the simplest thing is just to make
> > > everything return a future and let people call get() if they want to run
> > > synchronously. Having a mix of async and sync APIs is just going to be
> > > confusing and redundant.
> > >
> > > I think we should try to avoid creating single functions that start
> > > multiple requests if we can. It makes things much uglier. It means
> > > that you have to have some kind of request class that wraps up the
> > > request the user is trying to create, so that you can handle an array of
> > > those requests. The return value has to be something like Map<Node,
> > > Try<Value>> to represent which nodes failed and succeeded. This is the
> > > kind of stuff that, in my opinion, makes people scratch their heads.
> > >
> > > If we need to, we can still get some of the efficiency benefits of batch
> > > APIs by waiting for a millisecond or two before sending out a topic
> > > create() request to see if other create() requests arrive. If so, we
> > > can coalesce them. It might be better to figure out if this is an
> > > actual performance issue before implementing it, though.
> > >
> > > I think it would be good to get something out there, annotate it as
> > > @Unstable, and get feedback from people building against trunk and using
> > > it. We have removed or changed @Unstable APIs in streams before, so I
> > > don't think we should worry that it will get set in stone prematurely.
> > > The AdminClient API should get much less developer use than anything in
> > > streams, so changing an unstable API should be much easier.
> > >
> > > best,
> > > Colin
> > >
> > >
> > > On Wed, Feb 8, 2017, at 07:49, Ismael Juma wrote:
> > > > Thanks for elaborating Jay. I totally agree that we have to be very
> > > > careful
> > > > in how we use our complexity budget. Easier said than done when people
> > > > don't agree on what is complex and what is simple. :) For example, I
> > > > think
> > > > batch APIs are a significant source of complexity as you have to do a
> > > > bunch
> > > > of ceremony to group things before sending the request and error
> > > > handling
> > > > becomes more complex due to partial failures (things like `Try` or other
> > > > mechanisms that serve a similar role are then needed).
> > > >
> > > > Maybe a way forward is to write API usage examples to help validate that
> > > > the suggested API is indeed easy to use.
> > > >
> > > > Ismael
> > > >
> > > > On Wed, Feb 8, 2017 at 4:40 AM, Jay Kreps <j...@confluent.io> wrote:
> > > >
> > > > > Totally agree on CompletableFuture. Also agree with some of the rough
> > > edges
> > > > > on the Consumer.
> > > > >
> > > > > I don't have much of a leg to stand on with the splitting vs not
> > > splitting
> > > > > thing, really hard to argue one or the other is better. I guess the
> > > > > one
> > > > > observation in watching us try to make good public apis over the years
> > > is I
> > > > > am kind of in favor of a particular kind of simple. In particular I
> > > think
> > > > > since the bar is sooo high in support and docs and the community of
> > > users
> > > > > so broad in the range of their capabilities, it makes it so there is a
> > > lot
> > > > > of value in dead simple interfaces that don't have a lot of conceptual
> > > > > weight, don't introduce a lot of new classes or concepts or general
> > > > > patterns that must be understood to use them correctly. So things like
> > > > > nesting, or the Try class, or async apis, or even just a complex set
> > > > > of
> > > > > classes representing arguments or return values kind of all stack
> > > > > conceptual burdens on the user to figure out correct usage. So like,
> > > for
> > > > > example, the Try class is very elegant and represents a whole
> > > generalized
> > > > > class of possibly completed actions, but the flip side is maybe I'm
> > > just a
> > > > > working guy who needs to list his kafka topics but doesn't know Rust,
> > > take
> > > > > pity on me! :-)
> > > > >
> > > > > Nit picking aside, super excited to see us progress on this.
> > > > >
> > > > > -Jay
> > > > >
> > > > > On Tue, Feb 7, 2017 at 3:46 PM Ismael Juma <ism...@juma.me.uk> wrote:
> > > > >
> > > > > > Hi Jay,
> > > > > >
> > > > > > Thanks for the feedback. Comments inline.
> > > > > >
> > > > > > On Tue, Feb 7, 2017 at 8:18 PM, Jay Kreps <j...@confluent.io> wrote:
> > > > > > >
> > > > > > > - I think it would be good to not use "get" as the prefix for
> > > things
> > > > > > > making remote calls. We've tried to avoid the java getter
> > > convention
> > > > > > > entirely (see code style guide), but for remote calls in
> > > particular
> > > > > it
> > > > > > > kind
> > > > > > > of blurs the line between field access and remote RPC in a way
> > > that
> > > > > > > leads
> > > > > > > people to trouble. What about, e.g., fetchAllGroups() vs
> > > > > > getAllGroups().
> > > > > > >
> > > > > >
> > > > > > Agreed that we should avoid the `get` prefix for remote calls. There
> > > are
> > > > > a
> > > > > > few possible prefixes for the read operations: list, fetch,
> > > > > > describe.
> > > > > >
> > > > > >
> > > > > > > - I think futures and callbacks are a bit of a pain to use. I'd
> > > > > second
> > > > > > > Becket's comment: let's ensure there a common use case
> > > motivating
> > > > > > these
> > > > > > > that wouldn't be just as easily satisfied with batch operations
> > > > > (which
> > > > > > > we
> > > > > > > seem to have at least for some things). In terms of flexibility
> > > > > > > Callbacks >
> > > > > > > Futures > Batch Ops but I think in terms of usability it is the
> > > > > exact
> > > > > > > opposite so let's make sure we have worked out how the API will
> > > be
> > > > > > used
> > > > > > > before deciding. In particular I think java Futures are often
> > > > > > > an
> > > > > > > uncomfortable half-way point since calling get() and blocking
> > > the
> > > > > > > thread is
> > > > > > > often not what you want for chaining sequences of operations in
> > > a
> > > > > > truly
> > > > > > > async way, so 99% of people just use the future as a way to
> > > batch
> > > > > > calls.
> > > > > > >
> > > > > >
> > > > > > We should definitely figure out how the APIs are going to be used
> > > before
> > > > > > deciding. I agree that callbacks are definitely painful and there's
> > > > > little
> > > > > > reason to expose them in a modern API unless it's meant to be very
> > > low
> > > > > > level. When it comes to Futures, I think it's important to
> > > distinguish
> > > > > what
> > > > > > is available in Java 7 and below versus what is available from Java
> > > > > > 8
> > > > > > onwards. CompletableFuture makes it much easier to compose/chain
> > > > > operations
> > > > > > (in a similar vein to java.util.Stream, our own Streams API, etc.)
> > > and it
> > > > > > gives you the ability to register callbacks if you really want to
> > > > > (avoiding
> > > > > > the somewhat odd situation in the Producer where we return a Future
> > > _and_
> > > > > > allow you to pass a callback).
> > > > > >
> > > > > >
> > > > > > > - Personally I don't think splitting the admin methods up
> > > actually
> > > > > > makes
> > > > > > > things more usable. It just makes you have to dig through our
> > > > > > > hierarchy. I
> > > > > > > think a flat class with a bunch of operations (like the
> > > > > > > consumer
> > > > > api)
> > > > > > is
> > > > > > > probably the easiest for people to grok and find things on. I
> > > > > > > am
> > > > > kind
> > > > > > > of a
> > > > > > > dumb PHP programmer at heart, though.
> > > > > > >
> > > > > >
> > > > > > I am not sure it's fair to compare the AdminClient with the
> > > Consumer. The
> > > > > > former has APIs for a bunch of unrelated APIs (topics, ACLs,
> > > > > > configs,
> > > > > > consumer groups, delegation tokens, preferred leader election,
> > > partition
> > > > > > reassignment, etc.) where the latter is pretty specialised. For each
> > > of
> > > > > the
> > > > > > resources, you may have 3-4 operations, it will get confusing fast.
> > > Also,
> > > > > > do you really think an API that has one level of grouping will mean
> > > that
> > > > > > users have to "dig through our hierarchy"? Or are you concerned that
> > > once
> > > > > > we go in that direction, there is a danger of making the hierarchy
> > > more
> > > > > > complicated?
> > > > > >
> > > > > > Finally, I am not sure I would use the consumer as an example of
> > > > > something
> > > > > > that is easy to grok. :) The fact that methods behave pretty
> > > differently
> > > > > > (some are blocking while others only have an effect after poll) with
> > > no
> > > > > > indication from the type signature or naming convention makes it
> > > harder,
> > > > > > not easier, to understand.
> > > > > >
> > > > > > Ismael
> > > > > >
> > > > >
> > >
