Menu
▾
▴
Re: [fuse-devel] fuse-devel Digest, Vol 47, Issue 7
|
From: neutron <neu...@gm...> - 2010-03-29 16:01:07
|
On Mon, Mar 29, 2010 at 2:41 AM, <fus...@li...> wrote: > Send fuse-devel mailing list submissions to > fus...@li... > > To subscribe or unsubscribe via the World Wide Web, visit > https://lists.sourceforge.net/lists/listinfo/fuse-devel > or, via email, send a message with subject or body 'help' to > fus...@li... > > You can reach the person managing the list at > fus...@li... > > When replying, please edit your Subject line so it is more specific > than "Re: Contents of fuse-devel digest..." > > > Today's Topics: > > 1. Re: Node not found assertions (Michael Raskin) > 2. fuse 2.8.1 kernel module not installed? (neutron) > 3. fuse read/write sizes (neutron) > 4. Re: fuse 2.8.1 kernel module not installed? (Kevin Krieser) > 5. Re: fuse 2.8.1 kernel module not installed? (neutron) > 6. ceph mounting using fuse (Amarnath BS) > 7. Re: Echoing poll()/select() calls to underlying filesystems > (Goswin von Brederlow) > 8. fuse single/multi thread (neutron) > 9. Re: fuse single/multi thread (Miklos Szeredi) > > > ---------------------------------------------------------------------- > > Message: 1 > Date: Thu, 25 Mar 2010 10:14:29 +0300 > From: Michael Raskin <fb0...@ra...> > Subject: Re: [fuse-devel] Node not found assertions > To: fus...@li... > Message-ID: <4BA...@ra...> > Content-Type: text/plain; charset=ISO-8859-1 > > On 03/24/2010 04:57 PM, Michael Raskin wrote: >> fuse internal error: node 42 not found >> fatal error encountered in SBCL pid 19315(tid 3064134512): >> SIGABRT received. >> >> .sbcl-wrapped: fuse.c:807: forget_node: Assertion `node->nlookup >= >> nlookup' failed. >> fatal error encountered in SBCL pid 27730(tid 3050625904): >> SIGABRT received. > > By the way, is this all related to libfuse doing some caching? I already > have to do my own caching (some SQL queries I need are quite exepnsive), > so I would be totally OK with it not doing any, if there is a way to > tell it. > > > > ------------------------------ > > Message: 2 > Date: Thu, 25 Mar 2010 12:30:01 -0400 > From: neutron <neu...@gm...> > Subject: [fuse-devel] fuse 2.8.1 kernel module not installed? > To: k_k...@sb... > Cc: fus...@li... > Message-ID: > <7d5...@ma...> > Content-Type: text/plain; charset=ISO-8859-1 > > Hi Kevin, thanks! > > I loaded the fuse.ko from fuse-2.7.4, and compile my program with > fuse-2.8.1. It seems my program can run. Is this a right config ( > compile user program with 2.8.1, but use kernel module 2.7.4 )? > > > -Neutron > > >> Hi all, >> >> Sorry for t this silly question. I installed fuse 2.8.1 on my RHEL5.4 >> 2.6.18-128.7.1.el5. I used all default: >> >> ./configure >> make >> sudo make install >> >> I thought the fuse.ko should be installed to >> "/lib/modules/2.6.18-128.7.1.el5/kernel/fs/fuse/fuse.ko". But >> "modprobe" fail to find this file. Actually I couldn't see "fuse.ko" >> inside the fuse source dir after compiling. What's wrong with my >> config? >> >> >> -Neutron > > The 2.8 series uses the fuse.ko that is built as part of the kernel. > And fuse.ko is not included yet in 2.6.18-128... You would need the > latest 2.7 source to work with that kernel. Or upgrade to a later > redhat kernel. > > > > ------------------------------ > > Message: 3 > Date: Thu, 25 Mar 2010 16:34:34 -0400 > From: neutron <neu...@gm...> > Subject: [fuse-devel] fuse read/write sizes > To: fus...@li... > Message-ID: > <7d5...@ma...> > Content-Type: text/plain; charset=ISO-8859-1 > > Hi all, > > In my test with fuse-2.7.4, I noticed that a fuse_write request is at > most 4KB. So a user write of 1MB is translated into multiple > fuse_write each of which is 4KB. is there a tunable that determines > the size of a fuse-write? Or, how can I change fuse module to > increase this size? > > Also, I noticed that each fuse_read() is at least 4KB, at most 128KB, > with 128KB read-ahead if possible. > Can I change the size of a fuse_read? In my application I want the > size to be bigger. > > Thanks! > > -Neutron > > > > ------------------------------ > > Message: 4 > Date: Thu, 25 Mar 2010 16:56:17 -0500 > From: Kevin Krieser <k_k...@sb...> > Subject: Re: [fuse-devel] fuse 2.8.1 kernel module not installed? > To: neutron <neu...@gm...> > Cc: fus...@li... > Message-ID: <ACE...@sb...> > Content-Type: text/plain; charset=us-ascii > > > On Mar 25, 2010, at 11:30 AM, neutron wrote: > >> Hi Kevin, thanks! >> >> I loaded the fuse.ko from fuse-2.7.4, and compile my program with >> fuse-2.8.1. It seems my program can run. Is this a right config ( >> compile user program with 2.8.1, but use kernel module 2.7.4 )? >> >> >> -Neutron >> >> >>> Hi all, >>> >>> Sorry for t this silly question. I installed fuse 2.8.1 on my RHEL5.4 >>> 2.6.18-128.7.1.el5. I used all default: >>> >>> ./configure >>> make >>> sudo make install >>> >>> I thought the fuse.ko should be installed to >>> "/lib/modules/2.6.18-128.7.1.el5/kernel/fs/fuse/fuse.ko". But >>> "modprobe" fail to find this file. Actually I couldn't see "fuse.ko" >>> inside the fuse source dir after compiling. What's wrong with my >>> config? >>> >>> >>> -Neutron >> >> The 2.8 series uses the fuse.ko that is built as part of the kernel. >> And fuse.ko is not included yet in 2.6.18-128... You would need the >> latest 2.7 source to work with that kernel. Or upgrade to a later >> redhat kernel. > > I don't know if that is a correct combination or not. At work, I've used 2.7.4 for everything (to support ntfs-3g), and just did the same thing at home. If I reinstall at home, it will be to a version with the kernel module, so I'll just go to 2.8 > > > ------------------------------ > > Message: 5 > Date: Thu, 25 Mar 2010 18:00:53 -0400 > From: neutron <neu...@gm...> > Subject: Re: [fuse-devel] fuse 2.8.1 kernel module not installed? > To: Kevin Krieser <k_k...@sb...> > Cc: fus...@li... > Message-ID: > <7d5...@ma...> > Content-Type: text/plain; charset=ISO-8859-1 > > thanks Kevin! > > > > > On Thu, Mar 25, 2010 at 5:56 PM, Kevin Krieser <k_k...@sb...> wrote: >> >> On Mar 25, 2010, at 11:30 AM, neutron wrote: >> >>> Hi Kevin, thanks! >>> >>> I loaded the fuse.ko from fuse-2.7.4, ?and compile my program with >>> fuse-2.8.1. ?It seems my program can run. ?Is this a right config ( >>> compile user program with 2.8.1, but use kernel module 2.7.4 )? >>> >>> >>> -Neutron >>> >>> >>>> Hi all, >>>> >>>> Sorry for t this silly question. I installed fuse 2.8.1 on my RHEL5.4 >>>> 2.6.18-128.7.1.el5. I used all default: >>>> >>>> ./configure >>>> make >>>> sudo make install >>>> >>>> I thought the fuse.ko should be installed to >>>> "/lib/modules/2.6.18-128.7.1.el5/kernel/fs/fuse/fuse.ko". ?But >>>> "modprobe" fail to find this file. ?Actually I couldn't see "fuse.ko" >>>> inside the fuse source dir after compiling. ?What's wrong with my >>>> config? >>>> >>>> >>>> -Neutron >>> >>> The 2.8 series uses the fuse.ko that is built as part of the kernel. >>> And fuse.ko is not included yet in 2.6.18-128... ?You would need the >>> latest 2.7 source to work with that kernel. ?Or upgrade to a later >>> redhat kernel. >> >> I don't know if that is a correct combination or not. ?At work, I've used 2.7.4 for everything (to support ntfs-3g), and just did the same thing at home. ?If I reinstall at home, it will be to a version with the kernel module, so I'll just go to 2.8 > > > > ------------------------------ > > Message: 6 > Date: Sat, 27 Mar 2010 12:02:52 +0530 (IST) > From: Amarnath BS <ama...@ya...> > Subject: [fuse-devel] ceph mounting using fuse > To: fus...@li... > Message-ID: <261...@we...> > Content-Type: text/plain; charset=iso-8859-1 > > Hi, > ?i have mounted ceph file system using fuse. when i try to create some files i get an error like > Error stating file : Input/output error. Please help me to solve this problem > --Amarnath > > > > The INTERNET now has a personality. YOURS! See your Yahoo! Homepage. http://in.yahoo.com/ > > ------------------------------ > > Message: 7 > Date: Mon, 29 Mar 2010 02:03:16 +0200 > From: Goswin von Brederlow <gos...@we...> > Subject: Re: [fuse-devel] Echoing poll()/select() calls to underlying > filesystems > To: Stef Bon <st...@gm...> > Cc: fus...@li... > Message-ID: <878w9ct2nv.fsf@frosties.localdomain> > Content-Type: text/plain; charset=iso-8859-1 > > Stef Bon <st...@gm...> writes: > >> Well thanks for your answer. >> >> Again I've go some questions. I'm reading about the epoll mechanism. >> >> First, to make this work I have to rewrite the fuse_session_loop. My >> fs is only written in the high level, not with the low level >> functions. Do I have to rewrite my fs to low level will it make any >> sense rewriting this function? > > I've only done this for the lowelevel interface but I don't see why you > can't use it for the highlevel interface too. Afaik the original loop is > the same for low and highlevel too. > >> I see that with select and poll, that after it reports there is some >> activity on a or some fd's, there is still a loop necessary to check >> all the fd there has been activity/change on every single instance. >> Isn't there a possibility that epoll or select return an array or >> something indicating the fd(s) which have changed? >> >> Stef > > Epoll returns an array of events, which contains the epoll_data_t > union. You can store the FD in there or a pointer to a struct with more > data. Whatever you stored in there for epoll_ctl will be returned when > the event is triggered. > >> 2010/3/17 Goswin von Brederlow <gos...@we...>: >>>> >>>> Apart from how fuse finds out about inotify requests and how to pass >>>> them to the underlying fs, is this correct? >>> >>> Yes. It might be better to use epoll instead of select and set callbacks >>> to handle the fd with activity better. With many FDs select becomes >>> increasingly inefficient and I think epoll is just easier to use for you >>> in the first place. >>> >>> But that might just be a matter of taste with the number of FDs you >>> probably have to handle. >>> >>>> Stef >>> >>> MfG >>> ?? ?? ?? ??Goswin >>> > > MfG > Goswin > > > > ------------------------------ > > Message: 8 > Date: Mon, 29 Mar 2010 01:01:51 -0400 > From: neutron <neu...@gm...> > Subject: [fuse-devel] fuse single/multi thread > To: fus...@li... > Message-ID: > <7d5...@ma...> > Content-Type: text/plain; charset=ISO-8859-1 > > Hi all, > > I encountered an issue of "out-of-order reads" with FUSE multi-thread mode. > > If I started fuse with "-s" (single thread mode), I found that there > can be only one ongoing file ops at given time. > For example, if one application thread is reading file-a and the read > blocks, another application thread which wants to read file-b cannot > proceed, even if the data for file-b is available. > > If I use fuse with multi-thread mode, since one large file-read from > the application is split into multiple 128KB fuse_read, and multiple > fuse-threads are started to execute those 128KB reads, those 128KB > reads can come out-of-order. > For example, my application does a read(fd, 1MB), FUSE will perform > multiple 128 KB reads, and sometime a read at offset 256KB comes > before a read at offset 128KB. > > Does FUSE provide the function such that, the reads belonging to a > same file always come in-order of the offset? Thanks! > > > > -Neutron > > > > ------------------------------ > > Message: 9 > Date: Mon, 29 Mar 2010 08:41:35 +0200 > From: Miklos Szeredi <mi...@sz...> > Subject: Re: [fuse-devel] fuse single/multi thread > To: neutron <neu...@gm...> > Cc: fus...@li... > Message-ID: <E1N...@po...> > > On Mon, 29 Mar 2010, neutron wrote: >> Hi all, >> >> I encountered an issue of "out-of-order reads" with FUSE multi-thread mode. >> >> If I started fuse with "-s" (single thread mode), I found that there >> can be only one ongoing file ops at given time. >> For example, if one application thread is reading file-a and the read >> blocks, another application thread which wants to read file-b cannot >> proceed, even if the data for file-b is available. >> >> If I use fuse with multi-thread mode, since one large file-read from >> the application is split into multiple 128KB fuse_read, and multiple >> fuse-threads are started to execute those 128KB reads, those 128KB >> reads can come out-of-order. >> For example, my application does a read(fd, 1MB), FUSE will perform >> multiple 128 KB reads, and sometime a read at offset 256KB comes >> before a read at offset 128KB. >> >> Does FUSE provide the function such that, the reads belonging to a >> same file always come in-order of the offset? Thanks! > > -osync_read > > Thanks, > Miklos > Yes "-osync_read" solved the problem. Thanks a ton Miklos!!! -Neutron > > > ------------------------------ > > ------------------------------------------------------------------------------ > Download Intel® Parallel Studio Eval > Try the new software tools for yourself. Speed compiling, find bugs > proactively, and fine-tune applications for parallel performance. > See why Intel Parallel Studio got high marks during beta. > http://p.sf.net/sfu/intel-sw-dev > > ------------------------------ > > _______________________________________________ > fuse-devel mailing list > fus...@li... > https://lists.sourceforge.net/lists/listinfo/fuse-devel > > > End of fuse-devel Digest, Vol 47, Issue 7 > ***************************************** > |