Mailing-List: contact dev-help@ignite.incubator.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@ignite.incubator.apache.org
Received-SPF: pass (nike.apache.org: domain of sboikov@gridgain.com designates
 209.85.212.174 as permitted sender)
MIME-Version: 1.0
In-Reply-To: 
 <CA+0=VoUfL8xouR1zAShRSS3WJeepfEgEYy-SzabxP1R85x45zQ@mail.gmail.com>
References: 
 <CAFx+i-btxbGKEve2ansfWtqdkmiEQpDjDMUj=JAsWb3Q90FYCQ@mail.gmail.com>
	<CALH+G9quUp7XQc8=E2ZP-4ckn5NqjJbdsET-kd_FM=snb+v_Eg@mail.gmail.com>
	<CA+0=VoUfL8xouR1zAShRSS3WJeepfEgEYy-SzabxP1R85x45zQ@mail.gmail.com>
Date: Tue, 20 Jan 2015 18:27:32 +0300
Message-ID: 
 <CAFx+i-ZtZ86AX=AkqBXOeOvL45tpX0PNxp=ynYN_iaBFLGpEzg@mail.gmail.com>
Subject: Re: JCache read-through behavior
From: Semyon Boikov <sboikov@gridgain.com>
To: dev@ignite.incubator.apache.org
Content-Type: multipart/alternative; boundary=089e01493b40df8204050d1713ca

--089e01493b40df8204050d1713ca
Content-Type: text/plain; charset=UTF-8

This behavior is clearly stated in JCache spec, also there are
corresponding TCK tests (see
CacheLoaderTest.shouldNotLoadUsingPutIfAbsent,
CacheLoaderTest.shouldNotLoadUsingGetAndPut).

On Tue, Jan 20, 2015 at 6:19 PM, Dmitriy Setrakyan <dsetrakyan@gridgain.com>
wrote:

> Are we absolutely certain that this is indeed the JCache behavior and not
> our misconfiguration? Do we have CacheLoader and CacheWriter configured and
> isWriteThrough() and isReadThrough() flags enabled?
>
> If yes, then we should probably have a configuration property to enable and
> disable such behavior. How about isReadThroughPreviousValue()?
>
> D.
>
> On Tue, Jan 20, 2015 at 6:44 AM, Alexey Kuznetsov <akuznetsov@gridgain.com
> >
> wrote:
>
> > But what will happen if user select Ignite to replace some JCache
> > implementation in his production application?
> > He may get unexpected behaviour. Or not?
> >
> > I think we should change current implementation to conform to JCache for
> > compatibility with other implementations and have configuration option
> for
> > our current implementation.
> >
> > On Tue, Jan 20, 2015 at 4:49 PM, Semyon Boikov <sboikov@gridgain.com>
> > wrote:
> >
> > > Hi all,
> > >
> > > I'm implementing support for JCache CacheLoader and found that
> according
> > to
> > > JCache specification read-through behavior differs from current Ignite
> > > implementation for some cases:
> > >     - in current Ignite implementation for 'putIfAbsent(K key, V
> value)'
> > if
> > > value is not in cache then it is loaded from store, but in JCache
> > > CacheLoader should not be called in this case (the same for replace(K
> > key,
> > > V oldValue, V newValue), replace(K key, V oldValue), remove(K key, V
> > > oldValue))
> > >     - in current Ignite implementation if cache method returns previous
> > > value (getAndPut, getAndRemove) and value is not in cache then it is
> > loaded
> > > from store, but in JCache CacheLoader should not be called in this case
> > >
> > > In my opinion current Ignite behavior is more correct than in JCache
> > spec.
> > > Do you think we should change current implementation to conform to
> > JCache?
> > >
> >
> >
> >
> > --
> > Alexey Kuznetsov
> > GridGain Systems
> > www.gridgain.com
> >
>

--089e01493b40df8204050d1713ca--