On 23.10.2012 00:55, Matej Lazar wrote: > > Hi, > > to simplify a scenario lets say that I have a distributed cache with no > copies on a cluster of two nodes. > > Is stop operation blocked and waiting for cache data to be transferred to > the other node, when one node is stopped (normal stop, not a failure)? > > My use case is using Infinispan in JBoss AS (CapeDwarf). > How is node stopping behaving in Infinispan standalone and how in AS ? > In standalone the data from the node which shuts down is not loaded to other nodes. IMHO this behaviour is correct one, otherwise depending on amount of data the shutdown could take some time. Useful could be an feature request that "lost" entries are loaded from store, if store exists. (This feature I miss too ;-)

Attached simple example as test case. _______________________________________________ infinispan-dev mailing list infinispan-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/infinispan-dev

On 23.10.2012 15:36, Sanne Grinovero wrote: > Hi Thomas, what do you mean by "lost" entries? I'm pretty sure that if > an entry can't be find in any node Infinispan will attempt to look it > up from a CacheLoader. > Yes, if you know about the key. But size(), keyset() and so on are not aware that this particular key is not loaded. So you'll never know if you have all entries in cache... so you need to dig into the cache loader by yourself :-/ This is a bit unsatisfying, rhauch wrote also ISPN ticket for this problem :-) This problem appears, when you start seeing ISPN not only as cache in classic way, but as storage.

On 23 Oct 2012, at 14:53, Thomas Fromm <tf(a)tfromm.com&gt; wrote: > On 23.10.2012 15:36, Sanne Grinovero wrote: >> Hi Thomas, what do you mean by "lost" entries? I'm pretty sure that if >> an entry can't be find in any node Infinispan will attempt to look it >> up from a CacheLoader. >> > > Yes, if you know about the key. But size(), keyset() and so on are not > aware that this particular key is not loaded. > So you'll never know if you have all entries in cache... so you need to > dig into the cache loader by yourself :-/ > This is a bit unsatisfying, rhauch wrote also ISPN ticket for this > problem :-) > > This problem appears, when you start seeing ISPN not only as cache in > classic way, but as storage. I think the problem is in the keySet() and entrySet() methods. These were originally included as a part of extending ConcurrentMap, but IMO these methods are not practical for a data store. -- 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

On 25.10.2012 17:03, Matej Lazar wrote: > Yes, CapeDwarf uses Infinispan as persistent store. > > IMO Infinispan should provide setting to enable lossless shutdown, if we want to use it for storage or distributed file system etc. As I understand external storage is the only option. > > In a cloud environment, where nodes (server instances) come and go, relaying on external storage will force a user to use centralized store. Which is not the best solution if a user has requirements to work with distributed storage in the first place. Yes, nodes come and go... and some of them maybe not leave gracefully... Increase number of owners inside distributed caches to reduce risk. :-)

_______________________________________________ infinispan-dev mailing list infinispan-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/infinispan-dev