Sacha/Bill: Suspected bug

[derek]

Hi guys,

I posted a message earlier about JBoss HA-JNDi seemingly not to replicate as a instance of JBoss coming up failed doing a HA-JNDI lookup on an object already bound into the HA-JNDI by another instance of JBoss.

From the clustering docs, the sequence of lookups is described as follows:
1- local HA-JNDI lookup
2- local JNDI lookup
3- remote HA-JNDI lookup
4- remote JNDI lookup

The problem was that the remote HA-JNDI was never checked. When we looked in the HAJNDI source we found that step 3 was never performed, so we modifed the source to include it and now we get the expected result.

Please check the attached file with the changes to see if you agree with what we have done.

Regards,

Derek.

[slaboure]

Hello,

I don't understand the problem. the steps are:
1) look up in the global jndi tree
2) lookup in the local jndi tree
3) ask other distant jndi tree to lookup locally.

We don't have to ask other distant ha-jndi tree to lookup in their global tree because all nodes share the same information => step 1 already checked that!

[derek]

Hi

I understand what you are saying, and that is how we expected it to function. We then saw in the Clustering docs (Figure 6 HA-JNDI lookup process) that this did not seem to be the case. The diagram seems to imply the lookup process as I explained it before.

The problem is that when a new node joins the cluster, it's does not seem to be initialised with the current HA-JNDI data. We found that a lookup on joining a cluster fails until there is a change in the HA-JNDI at which point the cluster-wide JNDI is updated. It looks like it could actually be a problem in the initialisation on the HA-JNDI. I will take a look.

Derek.