[infinispan-dev] Multi get API

Radim Vansa rvansa at redhat.com
Wed Apr 1 03:42:00 EDT 2015 

On 03/31/2015 08:41 PM, Sanne Grinovero wrote:
> On 31 March 2015 at 16:25, Galder Zamarreño <galder at redhat.com> wrote:
>>> On 26 Mar 2015, at 15:46, Radim Vansa <rvansa at redhat.com> wrote:
>>>
>>> On 03/26/2015 03:00 PM, Sanne Grinovero wrote:
>>>> On 26 March 2015 at 13:43, William Burns <mudokonman at gmail.com> wrote:
>>>>> On Thu, Mar 26, 2015 at 8:50 AM, Sanne Grinovero <sanne at infinispan.org> wrote:
>>>>>> In terms of naming, I'd prefer _getMulti_.
>>>>>> My reasoning is it's nice if has a common prefix with _get_ as that
>>>>>> helps someone exploring the API from an IDE, giving better assistance
>>>>>> to people learning.
>>>>>>
>>>>>> getAll is misleading, as is getMany or get getAllOf as they imply I'm
>>>>>> searching for something based on a pattern but you're expecting the
>>>>>> keys and returning tuples key/value.
>>>>> I can understand your point, however since JCache already defined this
>>>>> as getAll I can't see but using that name.
>>>> Ok that makes it win a bit of points, and at least it starts with _get_.
>>> Was there any discussion regarding this during JSR-107 development? I
>>> wonder why have they chosen such misleading name. Galder?
>> Dunno, never saw it discussed, but I supposed they chose that name cos of Map.putAll
>>
>> I think it's the wrong name, getMany, getMulti are better IMO, regardless of what JCache/java.util.Map do.
>>
>> Also, see what they did with CacheWriter and CacheLoader, they're no symmetrical... and we followed the same pattern in our persistence :|
> My initial assumption was that we'd use a varargs parameter to list
> the keys to be fetched; in that case getAll() could have been invoked
> with zero parameters, and it would look like as it should return all
> entries of the grid.

Varargs could look nice, I wonder how much would the API bloat if we had 
both options. Though, it would be even better if we could force at least 
two arguments in the varargs API to prevent obvious mistakes as 
argument-less version or even trying to call it with Set<Object> but 
(e.g. due to refactoring) falling back to the vararg version. However, 
if we had getAll(Object o1, Object o2, Object... otherObjects), it would 
prohibit us from simply wrapping the varargs array into Set (for further 
processing), we would have to allocate another array.
In the end, just Set is probably better :) For varargs version, we can 
provide CollectionFactory.makeSet(T... objects) and recommend this in 
examples in documentation, since there's no appropriate method in Arrays 
nor in Collections.

I am still reluctant to the getAll() name. addAll and putAll insert all 
elements I am passing, while getAll suggests retrieval of all obejcts in 
the collection.  On the other hand, after looking into competitors' API, 
Coherence and EhCache use getAll(Collection), Hazelcast uses 
getAll(Set), so users can live with getAll.


>
> I guess having now a Set for the keys makes the getAll() not too bad.
>
> @William +1 to not implement the future-like behaviour (lazy loaded
> HashMap) in the first step. I wasn't asking for that, just thinking
> ahead about the merits of returning such a type.

I forgot to react on this - making the Map<K, V> lazy underneath is 
maybe obscuring the perf consequences from the user; he could expect 
that everything is local=fast after the getAll request returns. If we're 
going to have a lazy-loading map, it should rather look like Map<K, 
xxxFuture<V> >.

Radim

>
> Sanne
>
>>>>
>>>>>> I like the idea of returning a Map, as you can actually make the map
>>>>>> behave like a Future: the Map instance could be returned right away,
>>>>>> but retrieving of some entries would be blocking until they are
>>>>>> retrieved. You wouldn't need to block on having all entries, just the
>>>>>> ones being actively asked for. In other words, it's better than
>>>>>> returning a Future<Map>.
>>>>>> This should invalidate some of the previous concerns about using Map ?
>>>>> I had thought about this briefly, but didn't give it much though since
>>>>> it makes things quite complicated (interceptor stack and tx context I
>>>>> am looking at you).  I would say to do this later if we wanted, since
>>>>> the API doesn't have to be different (other than the fact that people
>>>>> should start possibly expecting CacheExceptions from calling Map.get
>>>>> and other methods).
>>>>>
>>>>>> @William I didn't understand your concern on upper and lower bounds.
>>>>>> The returned types should be coupled to the Cache types; being able to
>>>>> The return type is coupled to the Cache types, this was about the
>>>>> input type defined on the Set.
>>>>>
>>>>>> pass Object might be nice but let's make sure no casting is required
>>>>>> on the client? Could you make some examples of your alternative
>>>>>> proposal?
>>>>> With Set<?> the client would never have to do any casting.
>>>> I'm confused. Not sure which Set you're talking about. And remember
>>>> that if you're being very liberal with the promise of returned types,
>>>> how is the implementation going to handle the guarantee?
>>>>
>>>> For example, what's going to happen if the requested objects don't
>>>> match the user expected type? The API shouldn't make promises it can't
>>>> maintain.
>>> The cache.get(k1) should return object stored under key k2 such that
>>>
>>> k1.hashCode() == k2.hashCode() && k1.equals(k2)
>>>
>>> Nothing is said about k1's type (though, Object.equals() states that
>>> equals should always form symmetric relation).
>>> The contract was once set for all java collections, let's stick to that.
>>>
>>> Radim
>>>
>>>> Thanks,
>>>> Sanne
>>>>
>>>>>> On 26 March 2015 at 10:04, Radim Vansa <rvansa at redhat.com> wrote:
>>>>>>> On 03/26/2015 10:28 AM, Galder Zamarreño wrote:
>>>>>>>> Comments inline:
>>>>>>>>
>>>>>>>>> On 24 Mar 2015, at 14:22, William Burns <mudokonman at gmail.com> wrote:
>>>>>>>>>
>>>>>>>>> I am nearing completion of the new multi get command [1], allowing for
>>>>>>>>> more efficient retrieval of multiple keys at the same time.  Thanks
>>>>>>>>> again to Radim for a lot of the initial work.
>>>>>>>>>
>>>>>>>>> In doing so though I want to make sure I get feedback on how we want
>>>>>>>>> the API to actually look, since this is much more difficult to change
>>>>>>>>> down the line.
>>>>>>>>>
>>>>>>>>> There are a few things in particular I wanted to discuss.
>>>>>>>>>
>>>>>>>>> 1. The actual name of the method.  The main suggestions I have seen
>>>>>>>>> are getAll or getMany.  I do like the naming of the former, however it
>>>>>>>>> seems a bit misleading (remember API is getAll(Set) since we are
>>>>>>>>> really getting a subset.  So I am thinking the possibilities at this
>>>>>>>>> point are getAllOf, getAllFor or getMany.  I am leaning maybe towards
>>>>>>>>> the last one (getMany).  I am open to any suggestions though.
>>>>>>>> JSR-107 has getAll(Set), which I think it's a bit confusing too naming wise.
>>>>>>>>
>>>>>>>> getMany or getAllOf sound better to me.
>>>>>>>>
>>>>>>>>> 2. What should the return value be for this method.  Currently this
>>>>>>>>> returns a Map<K, V> which makes sense when we retain these values in
>>>>>>>>> the local transactional context and is pretty nice and clean for end
>>>>>>>>> users.
>>>>>>>>>
>>>>>>>>> The other idea is to use a streaming API possibly returning an
>>>>>>>>> Iterable<CacheEntry<K, V>>.  The streaming return value doesn't seem
>>>>>>>>> as intuitive to me, but opens us up for more efficient usage
>>>>>>>>> especially when there may be a lot of keys passed in (also results can
>>>>>>>>> be processed concurrently while retrieval is still occurring).
>>>>>>>>>
>>>>>>>>> I would lean towards returning the Map<K, V> value, however the next
>>>>>>>>> point could change that.
>>>>>>>>>
>>>>>>>>> 3. Where this method should live.  Right now this method is on the
>>>>>>>>> BasicCache interface which is a parent of both Cache (embedded) and
>>>>>>>>> RemoteCache (remote).  Unfortunately for remote clients to be as
>>>>>>>>> efficient as possible we really need a Streaming API so that we don't
>>>>>>>>> have multiple copies of data in memory at multiple places at the same
>>>>>>>>> time.  For that reason I suggest we use the streaminig API for both or
>>>>>>>>> have a different API for remote vs embedded.
>>>>>>>> Hmmm, I've been thinking about the same thing for Java 8 API (see the proposal wiki), and I think returning a Map limits things. On Java 8, I'm considering switching `evalAll` from returning a Map of CompletableFutures to a Stream... More to come next week.
>>>>>>>>
>>>>>>>> Now, for the current version, I think Map is limiting for both remote and embedded, because even for embedded, retrieving an entry might be a remote operation, so that forces you to wait for all of the entries to be available before you can return the map. Same thing for Remote.
>>>>>>>>
>>>>>>>> What about returning an Iterator? That could be lazy and more suited for the use case? If you need to return a Map, e.g. to implement JCache.getAll(), you can wait for all and construct a map.
>>>>>>> What's the expected usecase? I would expect that users want to just
>>>>>>> retrieve few keys at once, without any fancy lazy
>>>>>>> wait-for-completion-and-do-after stuff. Just imagine that you want to
>>>>>>> pass the two values into another method, Iterator/Stream API wouldn't
>>>>>>> help you with that much.
>>>>>>>
>>>>>>> It seems to me that it is very useful to have multi get with Map<K, V>
>>>>>>> as return type in the API - we don't want to force constructs like:
>>>>>>>
>>>>>>> Iterator<Map.Entry<K, V>> it = cache.multiGet(key1, key2, key3);
>>>>>>> for (; it.hasNext();) {
>>>>>>>       Map.Entry<K, V> entry = it.next();
>>>>>>>       if (entry.getKey().equals(key1)) value1 = entry.getValue();
>>>>>>>       else if (entry.getKey().equals(key2)) value2 = entry.getValue();
>>>>>>>       else if (entry.getKey().equals(key3)) value3 = entry.getValue();
>>>>>>> }
>>>>>>>
>>>>>>> or let them flip it into the map manually (HashMap does not have
>>>>>>> constructor accepting iterator)
>>>>>>>
>>>>>>> Iterator<Map.Entry<K, V>> it = cache.multiGet(key1, key2, key3);
>>>>>>> Map<K, V> map = new HashMap<>();
>>>>>>> for (; it.hasNext();) {
>>>>>>>       Map.Entry<K, V> entry = it.next();
>>>>>>>       map.put(entry.getKey(), entry.getValue());
>>>>>>> }
>>>>>>>
>>>>>>>
>>>>>>>> Now, whether to have a separate API for this, since we have Java 8 API coming up, I'd not worry too much at this point.
>>>>>>> It seems to me that with 8, we're going to have complicated lazy
>>>>>>> omnipotent API, and several facades simplifying that into Map, JCache or
>>>>>>> whatever. So, let's provide the simplistic API with Map now, and the
>>>>>>> lazy versions can get into the lazy API as soon as we'll see how the
>>>>>>> other methods will look like.
>>>>>>>
>>>>>>> Radim
>>>>>>>
>>>>>>>> Cheers,
>>>>>>>>
>>>>>>>>> Let me know what you guys think.
>>>>>>>>>
>>>>>>>>> Cheers,
>>>>>>>>>
>>>>>>>>> - Will
>>>>>>>>>
>>>>>>>>> [1] https://issues.jboss.org/browse/ISPN-2183
>>>>>>>>> _______________________________________________
>>>>>>>>> infinispan-dev mailing list
>>>>>>>>> infinispan-dev at lists.jboss.org
>>>>>>>>> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>>>>>>>> --
>>>>>>>> Galder Zamarreño
>>>>>>>> galder at redhat.com
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> infinispan-dev mailing list
>>>>>>>> infinispan-dev at lists.jboss.org
>>>>>>>> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>>>>>>> --
>>>>>>> Radim Vansa <rvansa at redhat.com>
>>>>>>> JBoss Performance Team
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> infinispan-dev mailing list
>>>>>>> infinispan-dev at lists.jboss.org
>>>>>>> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>>>>>> _______________________________________________
>>>>>> infinispan-dev mailing list
>>>>>> infinispan-dev at lists.jboss.org
>>>>>> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>>>>> _______________________________________________
>>>>> infinispan-dev mailing list
>>>>> infinispan-dev at lists.jboss.org
>>>>> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>>>> _______________________________________________
>>>> infinispan-dev mailing list
>>>> infinispan-dev at lists.jboss.org
>>>> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>>>
>>> --
>>> Radim Vansa <rvansa at redhat.com>
>>> JBoss Performance Team
>>>
>>> _______________________________________________
>>> infinispan-dev mailing list
>>> infinispan-dev at lists.jboss.org
>>> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>>
>> --
>> Galder Zamarreño
>> galder at redhat.com
>>
>>
>>
>>
>>
>> _______________________________________________
>> infinispan-dev mailing list
>> infinispan-dev at lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/infinispan-dev
> _______________________________________________
> infinispan-dev mailing list
> infinispan-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/infinispan-dev


-- 
Radim Vansa <rvansa at redhat.com>
JBoss Performance Team

More information about the infinispan-dev mailing list