Hello,

I'm currently encountering an issue with keys getting locked forever while using distributed cache.

Here's my scenario :

1. client make a request to the server.

2. request go trough a proxy to encapsulate pessimistic transaction over the default cache :

// begin transaction

try{

  defaultCache.getAdvancedCache().getTransactionManager().begin();

} catch (Exception e){

   throw e;

}

try{

   // do stuff

  Object o = method.invoke(...);

  // commit

   try{

     defaultCache.getAdvancedCache().getTransactionManager().commit();

   } catch (Exception e) {

     throw e;

   }

} catch (Exception e){

   // rollback

   try{

     defaultCache.getAdvancedCache().getTransactionManager().rollback();

   } catch (Exception e){

     // log as error

    // do not throw

   }

}

return o;

default cache is only used for it's transactionnal aspect (no data is ever stored in it) and configured like this :

// transaction

transactionManagerLookupClass="org.infinispan.transaction.lookup.DummyTransactionManagerLookup"

autoCommit="false"

useSynchronization="true"

transactionMode="TRANSACTIONAL"

syncCommitPhase="true"

useEagerLocking="true"

eagerLockSingleNode="true"

syncRollbackPhase="false"

lockingMode="PESSIMISTIC"

// clustering

mode="local"

3. in the "do stuff process" i'll have potentially many interactions with differents sync cache (either local or distributed) which locking is managed per cache and configured as follow :

// clustering

mode="local"

// l1

enabled="false"

// locking

writeSkewCheck="false"

lockAcquisitionTimeout="10000"

isolationLevel="READ_COMMITTED"

useLockStriping="false"

concurrencyLevel="32"

4. under high load, if a certain cache key get locked it will forever, and each subsequent attempt to lock the same key will result in a lock acquisition timeout exception.

Could my issue related to : https://issues.jboss.org/browse/ISPN-445

Do I manage properly exception in my proxy ? I haven't found any documentation on how exception should be managed.

Do the presented use case seems viable (default cache for transaction, multiple sync cache for data)  ?

Is the configuration appears alright for the presented use case ?

Can actually rollback fail ? What can we do to actually try to prevent it ? Is it any way to fix broken transaction / stalled keys without using recovery service ; to force unlock key ?

Thanks a lot for your time,