hibernate-search-3.1.1.GA , lucene2.4.1

Using synchronous execution with FSMasterDirectoryProvider, we miss from time to time index updates, due to write locks (another process is doing infrequent batch inserts).
I see that the index should get updated via transaction synchronization (aftercompletion). However, if the write lock acquisition fails, the resulting searchexception goes unnoticed by our application.
It would be very helpfull if there is some way we can be notified about such exception.

right good point.
The locking was designed for just two scenarios:
1) single application: no locking problems
2) cluster of 2 JVMs on same machine sharing a directory (some people like this setup): no application owns the lock for a long time period.

Still in your scenario, or even occasionally in case 2 as Lucene doesn't guarantee fairness (just polls), this is an issue.

As a workaround until this is fixed (patches are always welcome), couldn't you share the IndexWriter? you would need to join the code of the other project in same JVM.
I'll open an issue linked to http://opensource.atlassian.com/projects/hibernate/browse/HSEARCH-382

Here is the specific issue: http://opensource.atlassian.com/projects/hibernate/browse/HSEARCH-421

While I agree this should be solved, sharing the IndexWriter is an optimal solution in terms of performance.