Hi Udo, I am not quite sure exactly why the index behavior was decided to be this way. It probably had to do with specific use cases at the time (a guess would be executing replicated region queries on specific nodes only and leaving the others as replication/failover backups?), and now not wanting to change the existing behavior. I think we'd have to put to vote whether we want to make this change and assess the impact of doing so...
Thanks, -Jason On Sun, Nov 8, 2015 at 3:15 PM, Udo Kohlmeyer <[email protected]> wrote: > @Jason/Anil, > > Why are the index creation semantics for partitioned and replicate regions > different? > > This might just be my thinking, but the implicit creation of indexes > across the cluster might fly in the face of the "shared nothing" > architecture. > > But I can understand from a usability and user experience perspective, if > I (as a user) define an index, I'd want it to be defined across all nodes. > Now, with the implicit creation of and the index on NodeB, why would the > functionality be different for Replicate regions? > > Surely consistency would be the best approach here? A user defines an > index (AX) on region (X). Which means, all nodes that contain region (X) > should contain index (AX). > > That said, I might have missed this, but are we moving towards a per > region configuration paradigm. Where we can define a region, with > indexes,cache*Adapters. All that a user has to decide, is what distributed > system the region needs to be deployed on. Which I admit will be a better > fit into a global cluster configuration. > > Kindest Regards > ----------------------------- > *Udo Kohlmeyer * | *Snr Solutions Architect * | *Pivotal* > Mobile: +61 409-279-160 | [email protected] > www.pivotal.io > On 07/11/15 17:30, agingade wrote: > > Yes...that's right... > > -------- Original message -------- > From: John Blum <[email protected]> <[email protected]> > Date: 11/06/2015 6:38 PM (GMT-08:00) > To: [email protected] > Subject: Re: Indexes with WAN replication > > Ok, 1 more scenario... > > What happens when Node A defines PR1, but does not define an Index. > > Then Node B comes online, defined PR1 and also defines an Index. I am > assuming here that Node B will exchange PR information and also send A an > index create message in this case, similar to scenario 3, but in the > opposite order? > > Thanks, > John > > > On Fri, Nov 6, 2015 at 12:26 PM, Jason Huynh <[email protected]> wrote: > >> Scenario 1: NodeA and NodeB are up and running with a Partitioned Region >> (let's call this PR1) >> Now if an index is created on PR1 on NodeA, the message will be >> distributed to NodeB. >> >> Scenario 2: Startup both nodes with index defined in cache.xml: >> Both will start and create the PR and indexes defined. Both will send a >> create message to the other, where an IndexExistsException will be logged >> on at least one side but the end result should be both have the index. >> >> Scenario 3: Startup NodeA with index defined in cache.xml and some time >> later start up NodeB >> NodeB on creation of PR1 I think we exchange information, part of this >> processing will trigger an index create message to be sent to the newly >> created member if the region created is a partitioned region. >> >> Attempting to answer your questions now: >> >> So, in my scenario, suppose the cluster with Nodes A and B are not using >> Cluster Config, but both define a PARTITION Region (X) using cache.xml >> where Node A defines Index AX. Your saying Node B will "implicitly" define >> the same Index (i.e. AX) even though it was not "explicitly" defined in >> cache.xml for Node B on PARTITION Region X? >> I think this fits into Scenario 3 >> >> What happens if Node B goes down? Where does Node B get the Index >> information for PARTITION Region X? (primary?) >> Also Scenario 3 >> >> What happens if Node A (primary for Region X) and B both define the same >> Index (X) but with different definitions? (IndexExistsException?) >> Scenario 2 >> >> On Fri, Nov 6, 2015 at 11:31 AM, John Blum < <[email protected]> >> [email protected]> wrote: >> >>> Jason, is that true even without Cluster Config? I thought this applied >>> to both REPLICATE and PARTITION Regions (well any type of Region for that >>> matter... Local-only, NORMAL, etc). >>> >>> So, in my scenario, suppose the cluster with Nodes A and B are not using >>> Cluster Config, but both define a PARTITION Region (X) using cache.xml >>> where Node A defines Index AX. Your saying Node B will "implicitly" define >>> the same Index (i.e. AX) even though it was not "explicitly" defined in >>> cache.xml for Node B on PARTITION Region X? >>> >>> What happens if Node B goes down? Where does Node B get the Index >>> information for PARTITION Region X? (primary?) >>> >>> What happens if Node A (primary for Region X) and B both define the same >>> Index (X) but with different definitions? (IndexExistsException?) >>> >>> Thanks, >>> John >>> >>> >>> >>> On Fri, Nov 6, 2015 at 10:40 AM, Jason Huynh < <[email protected]> >>> [email protected]> wrote: >>> >>>> I think what John said is true for an index on a replicated region. If >>>> the index is created on a partitioned region, it will be distributed to the >>>> other nodes. >>>> >>>> On Fri, Nov 6, 2015 at 10:28 AM, John Blum < <[email protected]> >>>> [email protected]> wrote: >>>> >>>>> I think it is also worth nothing that that behavior is the same within >>>>> the same peer/distributed system as well. If Node A and B in the same >>>>> cluster define the same Region (X), but only Node A defines Index AX, that >>>>> index will not exist in Region X on Node B.\ >>>>> >>>>> -j >>>>> >>>>> On Fri, Nov 6, 2015 at 9:55 AM, Anilkumar Gingade < >>>>> <[email protected]>[email protected]> wrote: >>>>> >>>>>> Nikhil, >>>>>> >>>>>> Indexes are on regions; they are not replicated on their own...When >>>>>> data is changed in the region; that will be applied to indexes on that >>>>>> region. >>>>>> >>>>>> -Anil. >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> On Fri, Nov 6, 2015 at 7:23 AM, Nikhil Chandrappa < >>>>>> <[email protected]>[email protected]> wrote: >>>>>> >>>>>>> Hi All, >>>>>>> >>>>>>> I am working on configuring WAN replication between two cluster, I >>>>>>> had a question on indexes. >>>>>>> >>>>>>> Changes made to Indexes in one Gem cluster, does it get replicated >>>>>>> in remote Gem cluster? >>>>>>> >>>>>>> Thanks, >>>>>>> Nikhil >>>>>>> >>>>>> >>>>>> >>>>> >>>>> >>>>> -- >>>>> -John >>>>> 503-504-8657 >>>>> john.blum10101 (skype) >>>>> >>>> >>>> >>> >>> >>> -- >>> -John >>> 503-504-8657 >>> john.blum10101 (skype) >>> >> >> > > > -- > -John > 503-504-8657 > john.blum10101 (skype) > > >
