The reading thread in kill_sessions_common.cpp does acquire the client lock to access the logical session id on opCtx, the writing thread in transaction_oplog_application does not. As a result, we may have a read after write data race while accessing the logical session id.

A potential suggestion would be for the writing thread to also acquire the client lock before setting the logical session id.