From: Brian Pawlowski (beepy@netapp.com)
Date: 09/25/98-04:17:16 PM Z
From: beepy@netapp.com (Brian Pawlowski) Message-Id: <199809252117.OAA00861@cranford.netapp.com> Subject: Re: locking errors Date: Fri, 25 Sep 1998 14:17:16 -0700 (PDT) > This brings me to the subject of lock failure. I don't think leases > are going to be acceptable in all environments. Could variable duration > leases, with a client facility for requesting "no lease" - the client > holds the lock until it lets it go be made to work?. I'll respond between the two paragraphs. I'm not sure that CIFS/Windows locks have the persistence you desire. Leases are a mechanism to simplify state management in the face of failures (including network partitions). They simply provide the server a final control over access to a file while simplifying the state clean-up (lease expiration) in the face of non-responding clients. My understanding is today, if a CIFS client terminates (surely an infrequent occurrence:-) the server will time out something like an OpLock break to the non-responding client and yield the OpLock to the new requester. Or some such. The end result of this behaviour is similar to lease expiry and non-responding clients. I think when designing V4 and locking, we need to consider the >99% solution that works in most cases and explain the <1% failure modes (result of lease expiry and possible affect on client data). Network partitions and unexpected client failures are unavoidable. > Yes, this implies an administrative intervention scheme. > But it would be better than abandoning some LAN > environments where they are expecting strict CIFS/Windoz-like > behavior and the administrator finds it easier to manually > release locks than to explain to users continually why > NFS lock lease expiration terminated their application > or corrupted their data. > > John > > > -- > John Maddalozzo John speaketh not for his employer. > IBM RS/6000 Division internet: john@austin.ibm.com > 11400 Burnet Road, M.S. 9541 Notes: jmadd@us.ibm.com > Austin, TX 78758-3493 U.S.A phone: +1 (512) 838-2686 >
This archive was generated by hypermail 2.1.2 : 03/04/05-01:46:20 AM Z CST