This issue has been assigned to me for long, but I'm unable to find time to work on it soon. Still, it's quite an important one, quite annoying it's still open. Any volunteer? Requires a bit of coding in the Query module, but is totally unrelated to Lucene so anyone familiar with Infinispan could do it. Cheers, Sanne _______________________________________________ infinispan-dev mailing list infinispan-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/infinispan-dev

That's correct. It does not affect only joiners though, there are several cases: - first node starting but having existing data in some CacheStore - node A does a put with new type, existing node B then runs a query Looks like the easy solution is to store the set of known types as an entry, but even assuming there are no classloader issues there are some not so clear side effects, such as statistics getting skewed and similar. On Apr 11, 2013 8:24 PM, "Manik Surtani" <msurtani(a)redhat.com&gt; wrote: > I presume this is to deal with the case when a node joins the cluster > (shared index), and someone does a SearchManager.getQuery() without passing > in any types? Is that correct? If a type or more than one type is passed > in to that method then you don't have this problem, right? > > - M > > > On 11 Apr 2013, at 18:40, Sanne Grinovero <sanne(a)infinispan.org&gt; wrote: > > > This issue has been assigned to me for long, but I'm unable to find > > time to work on it soon. Still, it's quite an important one, quite > > annoying it's still open. > > > > Any volunteer? > > > > Requires a bit of coding in the Query module, but is totally unrelated > > to Lucene so anyone familiar with Infinispan could do it. > > > > Cheers, > > Sanne > > _______________________________________________ > > infinispan-dev mailing list > > infinispan-dev(a)lists.jboss.org > > https://lists.jboss.org/mailman/listinfo/infinispan-dev > > -- > Manik Surtani > manik(a)jboss.org > twitter.com/maniksurtani > > Platform Architect, JBoss Data Grid > http://red.ht/data-grid > > > _______________________________________________ > infinispan-dev mailing list > infinispan-dev(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/infinispan-dev > _______________________________________________ infinispan-dev mailing list infinispan-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/infinispan-dev

That's correct. It does not affect only joiners though, there are several cases: - first node starting but having existing data in some CacheStore - node A does a put with new type, existing node B then runs a query Looks like the easy solution is to store the set of known types as an entry, but even assuming there are no classloader issues there are some not so clear side effects, such as statistics getting skewed and similar. On Apr 11, 2013 8:24 PM, "Manik Surtani" <msurtani(a)redhat.com&gt; wrote: I presume this is to deal with the case when a node joins the cluster (shared index), and someone does a SearchManager.getQuery() without passing in any types? Is that correct? If a type or more than one type is passed in to that method then you don't have this problem, right? - M On 11 Apr 2013, at 18:40, Sanne Grinovero <sanne(a)infinispan.org&gt; wrote: > This issue has been assigned to me for long, but I'm unable to find > time to work on it soon. Still, it's quite an important one, quite > annoying it's still open. > > Any volunteer? > > Requires a bit of coding in the Query module, but is totally unrelated > to Lucene so anyone familiar with Infinispan could do it. > > Cheers, > Sanne > _______________________________________________ > infinispan-dev mailing list > infinispan-dev(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/infinispan-dev -- Manik Surtani manik(a)jboss.org twitter.com/maniksurtani Platform Architect, JBoss Data Grid http://red.ht/data-grid _______________________________________________ infinispan-dev mailing list infinispan-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/infinispan-dev _______________________________________________ infinispan-dev mailing list infinispan-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/infinispan-dev

Thanks Israel! You're very welcome back. In mandating the types : First problem is you either change the method signature or you need to throw a runtime exception when the varargs array is empty. More importantly, the query targets are supposed to be polymorphic and we actually need the list of class definitions of the subtypes too. The easy solution is to have them all listed in the configuration. Alternatively if you have users list all index names, we can initialize them eagerly and query the index for known class types, but this is problematic as - harder to check the configuration is consistent with runtime usage - such a query would need to be run periodically (before any other query?) to catch new types inserted by other nodes - some parameters of how the index is opened are actually inferred from the indexed types : this would need no be a two phases index initialization. Most user friendly solution is class path scanning, as we do in Hibernate.

As a side note : Hibernate Search had to implement capability to transparently hot-restart itself to support the dynamic type discovery of Infinispan but really that code hurts my eyes and is very tricky to maintain: every new functionality needs to be designed for it and tested for services or new configuration details which could appear at any time. I would love to remove it.

Sanne On Apr 12, 2013 10:42 AM, "Manik Surtani" <msurtani(a)redhat.com&gt; wrote: Mandating passing in the types in SearchManager.getQuery() won't work? On 11 Apr 2013, at 20:50, Sanne Grinovero <sanne(a)infinispan.org&gt; wrote: > That's correct. It does not affect only joiners though, there are several cases: > > - first node starting but having existing data in some CacheStore > - node A does a put with new type, existing node B then runs a query > > Looks like the easy solution is to store the set of known types as an entry, but even assuming there are no classloader issues there are some not so clear side effects, such as statistics getting skewed and similar. > > On Apr 11, 2013 8:24 PM, "Manik Surtani" <msurtani(a)redhat.com&gt; wrote: > I presume this is to deal with the case when a node joins the cluster (shared index), and someone does a SearchManager.getQuery() without passing in any types? Is that correct? If a type or more than one type is passed in to that method then you don't have this problem, right? > > - M > > > On 11 Apr 2013, at 18:40, Sanne Grinovero <sanne(a)infinispan.org&gt; wrote: > > > This issue has been assigned to me for long, but I'm unable to find > > time to work on it soon. Still, it's quite an important one, quite > > annoying it's still open. > > > > Any volunteer? > > > > Requires a bit of coding in the Query module, but is totally unrelated > > to Lucene so anyone familiar with Infinispan could do it. > > > > Cheers, > > Sanne > > _______________________________________________ > > infinispan-dev mailing list > > infinispan-dev(a)lists.jboss.org > > https://lists.jboss.org/mailman/listinfo/infinispan-dev > > -- > Manik Surtani > manik(a)jboss.org > twitter.com/maniksurtani > > Platform Architect, JBoss Data Grid > http://red.ht/data-grid > > > _______________________________________________ > infinispan-dev mailing list > infinispan-dev(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/infinispan-dev > _______________________________________________ > infinispan-dev mailing list > infinispan-dev(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/infinispan-dev -- Manik Surtani manik(a)jboss.org twitter.com/maniksurtani Platform Architect, JBoss Data Grid http://red.ht/data-grid _______________________________________________ infinispan-dev mailing list infinispan-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/infinispan-dev _______________________________________________ infinispan-dev mailing list infinispan-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/infinispan-dev