Cluster & CoreSession.setLock(…) [5.6]

[Léa_____DÉCORNO]

Hi,

I have two back-end servers on a single nuxeo-5.6 repository with concurrent scheduled import jobs running. To avoid concurrent document creation, I use a try {} catch {} block and set a document lock on a common parent document :

try {
 /* … */
 session.setLock(parentDocument.getRef());
 /* … */
 /* do multiple documents creation */
} finally {
 session.removeLock(parentDocument.getRef());
}

It did worked fine… until today, two concurrent document creation occurred.

So is CoreSession#setLock(DocumentRef ref) a good way of handling mutual exclusion between servers ? Is there a good (or more atomic) way of writing cluster-safe process ?

Cheers, Léa Raya DÉCORNOD

[rg1_]

The forthcoming Nuxeo State Manager, described in NXP-14553, will likely provide the ultimate solution to the issue you described. If you can't wait, you might consider adding a uniqueness constraint on specific columns in the hierarchy table as described in NXP-16912 and then writing some fairly low-level Java code to enforce unique pathSegments for children within a single Folderish document.

[Florent_Guillau]

This should work but 5.6 is a quite old Nuxeo version so we can't rule out locking bugs at this stage. Without a reproducible test case there's little hope to get to the bottom of this.