duncan@vuwcomp.UUCP (12/04/87)
In article <6759@brl-smoke.ARPA> gwyn@brl.arpa (Doug Gwyn) writes: >Fundamentally, NFS remains stateless. Last I heard, the locking was >being done by arrangement with external daemons. (It is actually record >locking, in support of SVID requirements, not just whole-file locking.) How do Sun make this work? If a locking daemon is running on a machine that crashes then that daemon loses track of what files are locked. If a client crashes, how does the locking daemon know to unlock files that it had locked? As Doug says, with the locking done by external daemons, the NFS protocol itself remains stateless, but I can't see how you still don't lose the benifits of it's statelessness. Duncan "The America's Cup -- why wait until 1991...?" Domain: duncan@comp.vuw.ac.nz Path: ...!uunet!vuwcomp!duncan
ed@mtxinu.UUCP (Ed Gould) (12/05/87)
In article <13107@comp.vuw.ac.nz> duncan@comp.vuw.ac.nz (Duncan McEwan) writes: >In article <6759@brl-smoke.ARPA> gwyn@brl.arpa (Doug Gwyn) writes: >>Fundamentally, NFS remains stateless. Last I heard, the locking was >>being done by arrangement with external daemons. > >How do Sun make this work? If a locking daemon is running on a machine >that crashes then that daemon loses track of what files are locked. >If a client crashes, how does the locking daemon know to unlock files >that it had locked? There are actually two cooperating daemons: the lock manager and the status monitor. They both exist on all machines. The status monitor is responsible for notifying the lock manager of client crashes. There is also a recovery protocol that allows clients to regain locks if the server crashes. Before the rebooted server accepts general lock requests, it accepts lock renewal requests for a period of time. -- Ed Gould mt Xinu, 2560 Ninth St., Berkeley, CA 94710 USA {ucbvax,uunet}!mtxinu!ed +1 415 644 0146 "`She's smart, for a woman, wonder how she got that way'..."