Problem with secondary parent associations in cluster

As we transition to the new Hyland Community, some post links may be temporarily disrupted as we finalize redirects. We apologize for the inconvenience and are working to resolve this as quickly as possible. Details.

francescof

Hi,

I have a problem with parent associations in my clustered setup (Alfresco 2.1 Community on a JBoss 4.0.5 2-node cluster): newly created nodes and their (primary) parent associations are replicated correctly. Secondary parent assocs are not.

HibernateNodeDaoServiceImpl.java look like this in 2.1 Community Final (line s from 990 to 1003):

            for (Long parentAssocId : parentAssocIds)
            {
                ChildAssoc parentAssoc = (ChildAssoc) getSession().get(ChildAssocImpl.class, parentAssocId);
                if (parentAssoc == null)
                {
                    // The cache is out of date, so just repopulate it
                    parentAssoc = null;
                    break;
                }
                else
                {
                    parentAssocs.add(parentAssoc);
                }
            }
‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍

I suppose the assigment "parentAssoc = null" should be "parentAssocs = null" and, if I'm right, that typo prevents assocs that are no longer in the DB from being expired from cache. What I can't figure out is whether this bug also influences correct propagation of new secondary child associations to other cluster nodes.

I also found a JIRA issue related to this part of the code: http://issues.alfresco.com/browse/AR-1743

Any ideas?

(Please note that after a fresh restart of the cluster I can see all the associations from all of the nodes.)

francescof

After deeper investigations I think I have identified the cause of the problem.

I noticed that, after a secondary child assoc creation no transaction is recorded on alf_transaction because newChildAssoc() and deleteChildAssocInternal() in HibernateNodeDaoServiceImpl don't update NodeStatus for the modified node. This way the remote node of the cluster doesn't know anything about the new child association and thus it doesn't update its lucene indexes.

A quick hack that seems to do the job is to add (with a check to execute this code ONLY on non-primary child-assocs) to newChildAssoc() and deleteChildAssocInternal():

NodeStatus nodeStatus = getNodeStatus(nodeRef, true);

nodeStatus.getTransaction().setChangeTxnId(AlfrescoTransactionSupport.getTransactionId());‍‍‍

This way alf_transaction and alf_node_status tables get updated, the remote node updates its lucene indexes and the secondary association is correctly viewable on the remote node web client.

My concern is: is it possible that this change carries dangerous side effects? I'm thinking about possible index corruption. I'm also aware that the hack described is not the best and cleanest solution for the problem. Have you any suggestions for improvements?

Bye.