9 Replies Latest reply on Jul 21, 2011 6:51 AM by pmuir

    About Infinispan instances

    jkiper

      Hi all, I'm new to the Infinispan world. I came from the database world, in fact i'm a dba. How could we "translate" the concept of taking a backup (for restore for example in the development or testing environment) in a point in time of a Infinispan instance? For example in the database world, a backup online including the transactional logs, for then making a restore in another server? How could this be done with InfiniSpan?

      I understand that it's a whole different concept, but I don't realize the way to do this with Infinispan and with no database under it.

       

      Thanks in advance...

        • 1. Re: About Infinispan instances
          pmuir

          I think the best thing to do would be to use a Write-Behind cache (see https://docs.jboss.org/author/display/ISPN/Write-Through+And+Write-Behind+Caching)...

          1 of 1 people found this helpful
          • 2. Re: About Infinispan instances
            jkiper

            So, the only way to have a consistent backup would be with synchronus cache... Am I right? Thinking on multiple nodes, with multiple disks, could this be possible without a database? I think that for certain types of applications where we need to restore the data to a specific point in time a database is practically a must?

            • 3. Re: About Infinispan instances
              sannegrinovero

              Yes about the synchronous, as the APIs don't expose "in flight" operations nor expose access to the transaction logs, so you could use the async only if you would stop writing to Infinispan for "some time", to wait that all async store operations have been applied in proper order.

               

              You don't need to use a database-backed cacheloader to store the data, but anything having some notion of "point in time restore" would do, right? some filesystems support such operations, of course a database seems a good fit and you could make good use of your dba skills.

               

              For your use case I'd consider implementing your own cacheloader: the interface is very simple, and you could intercept the changes by configuring it as a sync cacheloader, do what you need with it, and then delegate to an async store; the hard part to remember is that no node as a complete overview of the global state - that might be the big difference when coming from a database experience, and is also the main reason for scalability

              1 of 1 people found this helpful
              • 4. Re: About Infinispan instances
                jkiper

                Of course.... But you also have for example DB2 pureScale for scalability in the midrange database world, or paralel sysplex when talking about mainframes...

                • 5. Re: About Infinispan instances
                  jkiper

                  Please be aware i'm not against data grids or infinispan, i'm just trying to figure out some issues coming in mind about infinispan implementation.

                  • 6. Re: About Infinispan instances
                    manik

                    What about low-latency data access?  Being in-memory, you get the benefit of fast lookup with Infinispan.

                    • 7. Re: About Infinispan instances
                      jkiper

                      Yes, I know... My question is about persistence and backup / restore operations. I've mentioned pureScale and others in the means of scalability.

                      And you have another options for low-latency data access like in memory databases, solid disks, etc.

                      There are "dark areas" for me about Infinispan... like locking, commit and rollback operations, etc. when you have a cluster and a transaction pass the boundaries of a node...

                      A lot of stuff that is common in the database world I don't see them implemented in Infinispan.

                      • 8. Re: About Infinispan instances
                        ckulenk
                        • 9. Re: About Infinispan instances
                          pmuir

                          I've marked this as assumed answered as this dicussion appears to have run it's course. Julian, if you have further questions, feel free to continue or start a new topic.