Long running transactions might cause Nuxeo Drive to miss some document updates

Suppose a user updates 3 documents in a row (e.g. using drag and drop) and that writing documents take 2s each (for instance computing some thumbnails synchronously, a S3BinaryManager blob upload or something similarly costly). Let's call the first document creation / update start start at t0.

The CoreSession will fire document created / updated events at t0, t0 + 2, t0 + 4. The event audit logger is an asynchronous event listener that will insert the whole batch of events in the audit table at once at time (t0 + 5).

In the mean time let's assume that the client did two non-overlapping range queries: query 1 for the range [t0 - 2 to t0 + 3[ at time (t0 + 3) and query 2 for the range [t0 + 3, t0 + 8[ at time (t0 + 8).

The audit event logger is an asynchronous listener will only see the last document update that occurred at (t0 + 4) in the query 2. query 1 will return 0 result as the changed are not yet committed in the audit db at that time.

Therefore the range queries made by the client should not rely upon log_event_date but rather use a new log_date generated by the database at insertion time to guarantee a near monotonic behavior.