> -----Original Message----- > From: Łukasz Tasz [mailto:lukasz@xxxxxxx] > Sent: Thursday, November 29, 2012 9:00 AM > To: Myklebust, Trond > Cc: Adrien Kunysz; linux-nfs@xxxxxxxxxxxxxxx > Subject: Re: nfs performance - idea. > > 2012/11/29 Myklebust, Trond <Trond.Myklebust@xxxxxxxxxx>: > >> -----Original Message----- > >> From: Łukasz Tasz [mailto:lukasz@xxxxxxx] > >> Sent: Thursday, November 29, 2012 8:47 AM > >> To: Myklebust, Trond > >> Cc: Adrien Kunysz; linux-nfs@xxxxxxxxxxxxxxx > >> Subject: Re: nfs performance - idea. > >> > >> Hi, thanks a lot for answer, > >> > >> looks then even better! so to be 100% sure it good to traverse whole > >> path and open directory content. > > > > No. All you really need is the opendir(). If you want to make your own > > hack, all you need is a program that does something along the lines of > > > > DIR* p = opendir("."); > > if (p) > > closedir(p); > > > I think I got what you mean, but in case of ./a/b/c/d/e just > opendir('./a/b/c/d/e') could end with "No such file or directory" > > that's why I'm thinking about: > for d in "./a/b/c/d/e".split('/') > opendir(d) > > > >> One remark, since I did not mentioned it, I'm using NFSv3, is this > >> behavior same in NFSv3 and NFSv4? > > > > Yes. It is a consequence of the NFS close-to-open caching model, which is > common to all versions of NFS. > great, > could you please point me to document/manual which describe this model? > http://nfs.sourceforge.net/#faq_a8 > thanks a lot for your help! > Łukasz > > > > >> thanks & regards > >> Lukasz Tasz > >> Łukasz Tasz > >> > >> > >> 2012/11/29 Myklebust, Trond <Trond.Myklebust@xxxxxxxxxx>: > >> >> -----Original Message----- > >> >> From: linux-nfs-owner@xxxxxxxxxxxxxxx [mailto:linux-nfs- > >> >> owner@xxxxxxxxxxxxxxx] On Behalf Of Lukasz Tasz > >> >> Sent: Thursday, November 29, 2012 4:41 AM > >> >> To: Adrien Kunysz > >> >> Cc: linux-nfs@xxxxxxxxxxxxxxx > >> >> Subject: Re: nfs performance - idea. > >> >> > >> >> Any idea on this topic? > >> >> > >> >> Can I assume that "touch" could be interface for triggering > >> >> synchronising client state with server state? > >> >> > >> > > >> > 'ls' is better. All opendir() calls will trigger a revalidation of > >> > the directory > >> cache. > >> > > >> > > >> >> many thanks for help! > >> >> regards > >> >> Lukasz Tasz > >> >> > >> >> > >> >> 2012/11/14 Adrien Kunysz <adk@xxxxxxxxx>: > >> >> > On Wed, Nov 14, 2012 at 1:38 PM, Łukasz Tasz <lukasz@xxxxxxx> > wrote: > >> >> >> Hi all, > >> >> >> > >> >> >> I would like to consult some idea with you, > >> >> >> > >> >> >> Problem: > >> >> >> I have two processes which are doing some actions and one of > >> >> >> the action is done on a shared file system. > >> >> >> Issue is that this thing could be done only by one process, and > >> >> >> for this issue, locking mechanism is implemented. > >> >> >> Problem is that while one process is releasing lock, second one > >> >> >> is informed that file processing is finished, but unfortunately > >> >> >> files does not exists in context of second process. > >> >> >> Two processes are executed on two different hosts. NFS share is > >> >> >> mounted in a standard way, no special flag. > >> >> >> Problem I guess is with caches, lookupcache=none solves the > >> >> >> problem, but also causes others :) - performance. > >> >> >> > >> >> >> I know, it not possible to have all things at once - no complains. > >> >> >> But simple idea is that inside second process after > >> >> >> notification that files are generated execute touch function on > >> >> >> directory which holds files, This will cause unnecessary update > >> >> >> of modification date, but as a side effect I noticed that also > >> >> >> file gets visible immediately on client hosts. > >> >> >> > >> >> >> That's why my question is if this is expected and reasonable > >> behaviour? > >> >> >> At the end I'm looking for kind of 'sync' command which will > >> >> >> cause synchronization of directories content inside client and > >> >> >> server something like flush() - but in NFS it's more complex. > >> >> > > >> >> > Doesn't fsync(2) do what you want? If not, can you explain why? > >> >> > > >> >> >> thanks in advance for help, > >> >> >> > >> >> >> regards > >> >> >> Lukasz > >> >> -- > >> >> To unsubscribe from this list: send the line "unsubscribe linux-nfs" > >> >> in the body of a message to majordomo@xxxxxxxxxxxxxxx More > >> majordomo > >> >> info at http://vger.kernel.org/majordomo-info.html ��.n��������+%������w��{.n�����{��w��� jg��� �����ݢj����G���� ���j:+v���w�m������w�������h�����٥
