-
1. Cache inserts inside a transaction are not propagated across the cluster
If you want your data/locks to be propagated as your tx progresses use eager locking.
-
2. Re: Cache inserts inside a transaction are not propagated across the cluster
i have used eager locking true. useEagerLocking="true".
Even after the transaction commits without any exceptions , I see the second node doesnt get updated.
Its not only the lock, The cache update itself is not propagated to another node.
-
3. Re: Cache inserts inside a transaction are not propagated across the cluster
how many nodes do you have in the cluster? Distribution won't spread the data to all caches. E.g. if your cluster is {A,B,C} if you put k on A it will get replicated on 2(numOwners) nodes only, e.g. A and B . Now if you look at C data won't be there. Unless you do a C.get, which would take it from a remote node (A or B in this example).
-
4. Re: Cache inserts inside a transaction are not propagated across the cluster
yes Mircea. i am only using two nodes .
I am also verifying by checking cache.get() .
Everything works fine(cache.get() returns updated value) when i do my puts outside teh transaction. .
The only problem is when i do i cache.put() inside a transaction , and i do a cache.get() in another nodes it shows the old value.
cache.get() in the same node gives teh udpated value.
When i remove two lines of code - txnManager.begin() and txnManager.commit() - replication happens fine as expected across cluster.
So i doubt that teh issue could be in transaction commit , which does only local updation and doesnt propagate it across cluster.
-
5. Re: Cache inserts inside a transaction are not propagated across the cluster
Very strange. Can you use the dummy tx manager and see if ti works with that?
Just set the tx manager lookup attribute to org.infinispan.transaction.lookup.DummyTransactionManagerLookup
-
6. Cache inserts inside a transaction are not propagated across the cluster
Yes i tried as per your suggestion and it works fine with DummyTransactionmanager.
So could the issue be with Atomikos transaction manager ?
I am using atomikos because i need a JTA complaint transaction manager because i need to maintain tranasctions between multiple resources namely mysql database and inifinispan cache.
-
7. Cache inserts inside a transaction are not propagated across the cluster
I get a feeling this is not a bug with atomikos , instead something needs to be done from infinispan.
The reason for teh above statement is - Infinispan is able to understand transaction commit ,. The changes are relefected in the local VM .
Only changes fail to propagate across teh cluster.
Is there any kind of debugging i can do on teh infinispan code ? Anycheck points a can check to see if clsuter propagation code is called .
-
8. Cache inserts inside a transaction are not propagated across the cluster
I did try extending to some more usescase using atomikos transaction manager - (all usecases by running a loop)
1. 2 threads within the same Vm - are able to see teh changes in cluster.
2. Two JVMs on the same node - are not able to see changes in cache values inside a transaction.
3. Two jvms on different nodes are also not able to see cache modifications
Another thing I noticed is when i start a JVM on a seperate machine - teh first cache.get() - values is teh updated value - from the lastest commited transaction.
So i though it is caching after the first acces and tried to disable L1 caching. But no luck.
Only first iteration value is teh updated one(commited by teh last transaction) .
My loop is as follows
for (;;) {
Thread.currentThread().sleep(3 * 1000);
System.out.println("Cache content " + cache.get("key2"));
}
The first iteration gives teh updated value (one commited by teh transaction in another node ) . Later terations keep giving the old value . and do not see changes made in another node.
-
9. Cache inserts inside a transaction are not propagated across the cluster
I have figured oen more intesreting thing
When i use mode as replication - everything works fine , and commits within a transactiona re visible to otehr nodes.
the issue is only with mode = distribution , where transaction commits are not visvible to other nodes. Could this be a bug ?
-
10. Cache inserts inside a transaction are not propagated across the cluster
Sudheer, I'd suggest you create a test case that we can run to see the problem that you're having. Once you have that, please create a JIRA with the test case and we'll look into it.
-
11. Cache inserts inside a transaction are not propagated across the cluster
Sure Galder.
I will create a testcase and update the same.
-
12. Cache inserts inside a transaction are not propagated across the cluster
I have created the issue in jira along with sample code attached where the issue can be reproduced.
-
13. Cache inserts inside a transaction are not propagated across the cluster
Issue solved. You should try with 4.2.1.CR2.
-
14. Cache inserts inside a transaction are not propagated across the cluster
BTW, Galder, could you explain me how I can upgrade JBoss AS 6 to make use of this latest version of Infinispan ?