2026.3.1

What's Changed

• DM-52017: Race conditions in the implementation the Qserv query processor for some queries (bugs)

Notes:

• A significant (many orders of magnitude) reduced probability of the query lock-ups.
• Lockups of the Individual queries no longer result in locking up the corresponding frontend (Czar). An exception is a case explained in the Known Issues section.
• A "leak" of the TCP sockets open by the frontends during the internal result processing has been eliminated. The previous version of Qserv demonstrated a steady growth of the open (TCP state CONNECTED) sockets when processing results of queries constrained by LIMIT <N>. Eventually, the number of open file descriptors (sockets) would hit the limit configured for the frontend's process. After that, the frontend will misbehave (fail to process queries, accept user requests, etc.).

Known issues

• DM-53046: Sub-optimal performance of the ANTLR4-based query parser when the query text size is large (many kilobytes). The problem is still under investigation. It may be addressed in a later release (performance
• DM-54321: A possible race condition in the XRootD/SSI software stack exists in the release. It may cause occasional deadlocks of queries where the result set is constrained byLIMIT <N> (TOP <N> in ADQL). The problem is typically seen in 1 out of 10,000 such queries in which the constraint is actually enforced. Normally, the locked-up queries do not prevent Qserv from processing other queries. However, at some point, the total number of the internally locked threads will reach a configuration limit for the number of the XRootD/SSI CB (call-back) threads. After that, the query processing stalls. Cancelling the locked-up queries doesn't help to address the problem, as the cancellation doesn't release the locked threads. The frontend affected by the problem needs to be restarted. (bugs)

Full Changelog: 2026.2.2-rc1...2026.3.1-rc1