Re: [infinispan-dev] input regarding ISPN-2103

I had a quick look. The way AtomicMaps are used is that they are retrieved using the AtomicMapLookup and an AtomicMapProxy is passed back. This proxy is then used, spanning multiple transactions. Each caller thread gets its own AtomicMapProxy instance, all referring to a single AtomicHashMap instance which is actually stored in the cache. So, for readers, just checking whether the AtomicHashMap is null isn't enough since the caller's AtomicMapProxy may have a handle onto an AtomicHashMap that has since been removed. I think the 'removed' flag should remain, and if, on the proxy at any given time, removed == true, then: 1. For an existing transaction, it should allow the OL/PL code handle anything to do with write skews 2. For a new transaction starting up, re-establish the reference to the AtomicHashMap from the proxy (if it exists), otherwise throw an ISE (new transaction cannot start since it is referring to an instance that doesn't exist anymore). WDYT? Cheers Manik On 11 Jun 2012, at 11:40, Adrian Nistor wrote: -- Manik Surtani manik(a)jboss.org twitter.com/maniksurtani Lead, Infinispan http://www.infinispan.org