Skip to content

AthenaPoolCnvSvc: Reject late SharedWriter clients

A significant portion of the latest DAOD production jobs is observed to fail on certain sites (e.g. about 30% at MWT2 while it is <1.5% at CERN for 23.0.4). During local investigations, we noticed that a process (not necessarily a proper SharedWriter client) sending an interrupt signal to the port that the SharedWriter server listens to may cause a similar behavior. This might be due to a site specific security check or something else. Regardless, we now reject client requests that are more than what we expect (based on the number of clients and streams). Although not bullet-proof, hopefully this'll address the majority of the issues that are observed from a practical point of view.

Relates to ATEAM-857

Edited by Alaettin Serhan Mete

Merge request reports