rhartman@thestepchild.sgi.com (Robert Hartman) (04/20/91)
In article <784@necssd.NEC.COM> harrison@necssd.NEC.COM (Mark Harrison) writes: >In article <1991Apr17.201931.14536@odin.corp.sgi.com>, >rhartman@thestepchild.sgi.com (Robert Hartman) writes: > >> Aside from the fact that it isn't guaranteed over NFS, you can get some >> measure of protection this way: > > [example using lockfile] > >Could you elaborate on why it's not guaranteed over NFS? > >Mark. ("Yikes!") >-- >Mark Harrison harrison@ssd.dl.nec.com >(214)518-5050 {necntc, cs.utexas.edu}!necssd!harrison >standard disclaimers apply... I was under the impression that NFS guaranteed idempotency, but when I posted my csh example, someone followed up with an explanation of why this is not so. The upshot is that unless you use the NFS lock manager, lock files over NFS are not guaranteed. To use the NFS lock manager you need to write a C program. As of yet I don't think there's a command-level implementation. My examples made it "highly unlikely" that you'd have a race condition on the lock file. Depending on your application, that might be good enough. BTW, does this imply that SCCS and RFS can also fail over NFS? I'm not aware of them using the NFS lock manager! -r