6 Replies Latest reply on Apr 26, 2013 4:55 AM by paulpa63

    Question related to ISPN-1440 index rebuilding

    paulpa63

      Hi,

       

      I noticed that the latest Infinispan (5.2) includes the index rebuild feature of ISPN-1440.  I have attempted to hook this into our application but discovered that it only works for "distributed" mode caches.  What is the rationale for this, will it be extended to support indexes on replicated and local caches?  If there is a server crash and the index update is missed then it would become inconsistent with the cache, in this scenario it would be necessary to perform a manual index rebuild.

       

      Paul

        • 1. Re: Question related to ISPN-1440 index rebuilding
          galder.zamarreno

          What problems do you have with replicated or local caches?

          • 2. Re: Question related to ISPN-1440 index rebuilding
            sannegrinovero

            I guess it's related to

            https://issues.jboss.org/browse/ISPN-2812

             

            The MassIndexer is using Map/Reduce.

            • 3. Re: Question related to ISPN-1440 index rebuilding
              paulpa63

              Hi Sanne,

               

              So what is the upshot, can we expect to see index rebuilding supported for replicated (and local mode) caches or is it not technically feasible or is it just not required by sufficient numbers of other Infinispan users?  To me it seems like a fairly fundamental facility and I would feel much happier if I knew that our application could perform this action on the command of a (sufficiently privileged) administrative user.

               

              Paul

               

              PS I have noticed that in the future the Infinispan internals will be refactored so that replicated/local caches become special cases of the distributed cache (in which case all cache classes would presumbly support a common set of core / associated operations) but this development will likely come much too late for our application.

              • 4. Re: Question related to ISPN-1440 index rebuilding
                sannegrinovero

                Hi Paul,

                I don't know I just asked information myself about ISPN-2812 too. I guess te problem you mention is related to it, but that conflicts with what I see reported on ISPN-2192; I'd need to check what others did but need to postpone to next week. Maybe you could have a look?

                 

                I based the MassIndexer implementation on Map/Reduce as it seemed a good fit for the problem and I wasn't aware of the limitation, but if the Infinispan core team can't provide Map/Reduce on local/repl mode in short time it's certainly not hard to adapt the MassIndexer implementation to use a simpler approach when running on such a cache mode. Annoying though, sorry I thought this worked whatever the mode.

                • 5. Re: Question related to ISPN-1440 index rebuilding
                  paulpa63

                  Hi Sanne,

                   

                  I have been thinking a bit more about this area and on reading another discussion article "Cache Store really supports transactional updates" I have been prompted to continue this discussion.  As you may recall, we are using indexes extensively in our application and have encountered a few issues which you have fixed.  My outstanding fundamental question is: are Infinispan index updates transactional? i.e. is the Hibernate Search / Lucene indexing engine enlisted in the current transaction, when XA transactions have been enabled - so that there is a guarantee that cache updates and index content remain consistent?

                   

                  TIA

                   

                  Paul

                  • 6. Re: Question related to ISPN-1440 index rebuilding
                    paulpa63

                    This outstanding question will be moved to a new discussion article.