[infinispan-dev] Cache Store Does'nt Work Async Whenever Modification is Inside Transaction

Amin Abbaspour a.abbaspour at gmail.com
Sat Jan 30 03:24:10 EST 2010


I did some tests. Unfortunately my results are not much satisfactory.

I do confirm that now changes are synced in a separate
CoalescedAsyncStore thread, but the aggregate process we discussed
before here ( https://jira.jboss.org/jira/browse/ISPN-116 )  does not
work anymore. I can see that not the only final value but all
modifications are flushed. (i.e. sync at then end of a 100
modifications period results in 100 stores)

The other issue is that only cache.put() worked for me this way, and
it ignored all modification via replace for example.

Let me know if you have done similar tests and your results plz.

Thanks
Amin


On Fri, Jan 29, 2010 at 7:56 PM, Amin Abbaspour <a.abbaspour at gmail.com> wrote:
> Thanks Galder for such a quickly fix.
>
> like Philippe I was not expecting this major issue to be fixed so
> fast. I have not tested it myself yet, but will check and inform my
> test result tomorrow.
>
> Amin
>
> On Fri, Jan 29, 2010 at 4:15 PM, Manik Surtani <manik at jboss.org> wrote:
>>
>> On 29 Jan 2010, at 12:21, Philippe Van Dyck wrote:
>>
>> Thanks Galder, I could have.
>> As you noticed in the thread, this bug is a blocker for me and since qi4j
>> 1.0 was released two days ago, I am now only waiting for the latest release
>> of JClouds (beta3) to integrate all of them.
>> So having it solved so quickly (!!! :-) ) did change a lot of my schedule
>> these last days and when I asked the previous question I was far away from
>> any fully fledged browser...
>> Until Jira fully integrates with mailing lists, there will be a lag between
>> a "this is done now" message and the "here is the svn log" explanation for
>> anybody with a crappy phone like me ;-)
>> Actually, having Jira manage (or observe) the mailing list could be a very
>> good idea... linking code changes to specific parts of discussions and
>> having Jira to automagically tag it as 'bug' 'task' or 'improvement' would
>> be very nice (Jira could also add links to code...)!
>>
>> Only once JIRA actually fixes and commits bugs (and makes me coffee) will I
>> be truly impressed!  :)
>>
>> phil
>>
>> On Fri, Jan 29, 2010 at 12:19 PM, Galder Zamarreno <galder at jboss.org> wrote:
>>>
>>>
>>> On 01/28/2010 06:15 PM, Philippe Van Dyck wrote:
>>> > Could you please be a bit more specific ?
>>> > Are there any test to confirm the closing of the bug ?
>>> > Is it in trunk ?
>>>
>>> Dude, you could at least check the jira and its svn tab!
>>>
>>>
>>> https://jira.jboss.org/jira/browse/ISPN-340?page=com.atlassian.jira.plugin.ext.subversion:subversion-commits-tabpanel
>>>
>>> >
>>> > That sounds like a really good news ;-)
>>> >
>>> > phil
>>> >
>>> > On Thu, Jan 28, 2010 at 6:10 PM, Galder Zamarreno<galder at jboss.org>
>>> >  wrote:
>>> >
>>> >> This is done now.
>>> >>
>>> >> On 01/26/2010 01:10 PM, Amin Abbaspour wrote:
>>> >>> Created https://jira.jboss.org/jira/browse/ISPN-340
>>> >>> You can vote.
>>> >>>
>>> >>> On Tue, Jan 26, 2010 at 3:21 PM, Sanne Grinovero
>>> >>> <sanne.grinovero at gmail.com>   wrote:
>>> >>>> +1
>>> >>>> I don't know if it's correct to consider it a bug, but choices on
>>> >>>> JIRA
>>> >>>> are limited.
>>> >>>> I would really need this - performance impact is high - and hope they
>>> >>>> could fix it before 4.0
>>> >>>>
>>> >>>> Regards,
>>> >>>> Sanne
>>> >>>>
>>> >>>> 2010/1/26 Amin Abbaspour<a.abbaspour at gmail.com>:
>>> >>>>> Yes Philippe, I think it is a bug.
>>> >>>>>
>>> >>>>> This totally ruins all attempts to create a write-behind async
>>> >>>>> store.
>>> >>>>> I don't know if JBoss Cache had the same issue or not, but for many
>>> >>>>> use-cases (including ours) one will need both TX and write-behind
>>> >>>>> simultaneously.
>>> >>>>>
>>> >>>>> OK, I will open a bug in Jira for 4.0.0-GA.
>>> >>>>>
>>> >>>>> Amin
>>> >>>>>
>>> >>>>> On Tue, Jan 26, 2010 at 2:12 PM, Philippe Van Dyck<pvdyck at gmail.com>
>>> >>   wrote:
>>> >>>>>> Sounds like a good idea.
>>> >>>>>> How do you qualify it ? Bug ? (perf)  Improvement ?
>>> >>>>>> It has a major impact on performance when using the S3 CacheStore,
>>> >>>>>> so
>>> >> I
>>> >>>>>> would like it to be a "performance bug".
>>> >>>>>> WDYT ... especially regarding the planning of the next release ?
>>> >>>>>> Phil
>>> >>>>>> On Tue, Jan 26, 2010 at 11:28 AM, Amin Abbaspour<
>>> >> a.abbaspour at gmail.com>
>>> >>>>>> wrote:
>>> >>>>>>>
>>> >>>>>>> Hi,
>>> >>>>>>>
>>> >>>>>>> Should I open a jira task for this?
>>> >>>>>>>
>>> >>>>>>> Amin
>>> >>>>>>>
>>> >>>>>>> On Sat, Jan 23, 2010 at 4:59 PM, Amin Abbaspour<
>>> >> a.abbaspour at gmail.com>
>>> >>>>>>> wrote:
>>> >>>>>>>> I guess this is a somehow a bug or typo.
>>> >>>>>>>>
>>> >>>>>>>> Transactions and stores are two separate concepts. While it
>>> >>>>>>>> sounds
>>> >> to
>>> >>>>>>>> flush to normal store at the TX commit but when one knowingly
>>> >>>>>>>> sets
>>> >>>>>>>> store as async, he accepts the risk of consistence in favor of
>>> >> speed.
>>> >>>>>>>>
>>> >>>>>>>> I expect this to be fixed (or at least discussed) before GA.
>>> >>>>>>>>
>>> >>>>>>>> Amin
>>> >>>>>>>>
>>> >>>>>>>> On Sat, Jan 23, 2010 at 3:48 PM, Philippe Van
>>> >>>>>>>> Dyck<pvdyck at gmail.com
>>> >>>
>>> >>>>>>>> wrote:
>>> >>>>>>>>> I do confirm this behavior.
>>> >>>>>>>>> Transactions and asynchronous updates may seem antagonistic, but
>>> >> they
>>> >>>>>>>>> are
>>> >>>>>>>>> not on a multi-level cache.
>>> >>>>>>>>> I plan to use a synchronous transactional disk cache and a
>>> >>>>>>>>> second
>>> >> non
>>> >>>>>>>>> transactional asynchronous s3 cache, to get the best of both
>>> >> worlds.
>>> >>>>>>>>> Since the asynchronous aspect of the s3 cache will not allow me
>>> >>>>>>>>> to
>>> >> get
>>> >>>>>>>>> any
>>> >>>>>>>>> feedback (aka Future), I want it to 'write-behind' in the
>>> >> background
>>> >>>>>>>>> (and
>>> >>>>>>>>> hope everything goes well).
>>> >>>>>>>>> I will wait for the qi4j v1 release (probably Monday), adapt
>>> >> infinispan
>>> >>>>>>>>> to
>>> >>>>>>>>> the blobstore of jClouds and take a closer look at this in a
>>> >>>>>>>>> couple
>>> >> of
>>> >>>>>>>>> weeks.
>>> >>>>>>>>> Phil
>>> >>>>>>>>>
>>> >>>>>>>>> On Sat, Jan 23, 2010 at 12:58 PM, Amin Abbaspour
>>> >>>>>>>>> <a.abbaspour at gmail.com>
>>> >>>>>>>>> wrote:
>>> >>>>>>>>>>
>>> >>>>>>>>>> Hi All,
>>> >>>>>>>>>>
>>> >>>>>>>>>> Regarding
>>> >>>>>>>>>>
>>> >> http://community.jboss.org/wiki/Write-ThroughAndWrite-BehindCaching
>>> >>>>>>>>>> , Even if Syncer is in Async mode, but if we modify a key
>>> >>>>>>>>>> inside a
>>> >>>>>>>>>> transaction boundary it is always stored synchronously! Why is
>>> >>>>>>>>>> it
>>> >> so?
>>> >>>>>>>>>> How can I have both locking and async store. FYI I use JBossTM
>>> >>>>>>>>>> Standalone.
>>> >>>>>>>>>>
>>> >>>>>>>>>> Regards,
>>> >>>>>>>>>> Amin Abbaspour
>>> >>>>>>>>>> _______________________________________________
>>> >>>>>>>>>> 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
>>> >>>>>>
>>> >>>>>
>>> >>>>> _______________________________________________
>>> >>>>> 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
>>> >>
>>> >> --
>>> >> Galder Zamarreño
>>> >> Sr. Software Engineer
>>> >> Infinispan, JBoss Cache
>>> >> _______________________________________________
>>> >> 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
>>>
>>> --
>>> Galder Zamarreño
>>> Sr. Software Engineer
>>> Infinispan, JBoss Cache
>>> _______________________________________________
>>> 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
>>
>> --
>> Manik Surtani
>> manik at jboss.org
>> Lead, Infinispan
>> Lead, JBoss Cache
>> http://www.infinispan.org
>> http://www.jbosscache.org
>>
>>
>>
>>
>> _______________________________________________
>> infinispan-dev mailing list
>> infinispan-dev at lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>>
>




More information about the infinispan-dev mailing list