|
View:
New views
10 Messages
—
Rating Filter:
Alert me
|
|
|
test statusTwo test runs using "dbench -D /mnt/dbench-test-dir 60" (in this case
the older dbench 3.04) : One against Samba 3.0.28a had the usual errors in the cleanup phase (which are hard to avoid due to their broken posix open support but we need to look into more) The test run against more current Samba server (3.4.0-GIT-b660e51-devel) failed though: 60 1834 1.79 MB/sec execute 304 sec 60 1837 1.79 MB/sec execute 305 sec 60 1840 1.78 MB/sec execute 306 sec 60 1843 1.77 MB/sec execute 307 sec 60 1846 1.77 MB/sec execute 308 sec (1861) open /mnt1/dbench-test-dir/clients/client27/~dmtmp/PARADOX/STUDENTS.DB failed for handle 10277 (Too many open files) (1877) open /mnt1/dbench-test-dir/clients/client51/~dmtmp/PARADOX/__43904.DB failed for handle 10283 (Too many open files) (1878) ERROR: handle 10283 was not found (1834) open /mnt1/dbench-test-dir/clients/client59/~dmtmp/PARADOX/COURSES.VAL failed for handle 10269 (Too many open files) (1835) ERROR: handle 10269 was not found (1861) open /mnt1/dbench-test-dir/clients/client40/~dmtmp/PARADOX/STUDENTS.DB failed for handle 10277 (Too many open files) (1808) open /mnt1/dbench-test-dir/clients/client58/~dmtmp/PARADOX/__42B96.DB failed for handle 10262 (Too many open files) (1809) ERROR: handle 10262 was not found (1954) open /mnt1/dbench-test-dir/clients/client29/~dmtmp/PARADOX/__43212.DB failed for handle 10288 (Too many open files) (1955) ERROR: handle 10288 was not found (1868) open /mnt1/dbench-test-dir/clients/client5/~dmtmp/PARADOX/__43904.DB failed for handle 10281 (Too many open files) 60 1850 1.76 MB/sec execute 309 sec Child failed with status 1 Can anyone else reproduce this with Jeff's inode patch series? -- Thanks, Steve _______________________________________________ linux-cifs-client mailing list linux-cifs-client@... https://lists.samba.org/mailman/listinfo/linux-cifs-client |
|
|
Re: test statusOn Fri, 2009-06-26 at 11:52 -0500, Steve French wrote:
> Two test runs using "dbench -D /mnt/dbench-test-dir 60" (in this case > the older dbench 3.04) : > One against Samba 3.0.28a had the usual errors in the cleanup phase > (which are hard to avoid due to their broken posix open support but we > need to look into more) > The test run against more current Samba server > (3.4.0-GIT-b660e51-devel) failed though: > > 60 1834 1.79 MB/sec execute 304 sec > 60 1837 1.79 MB/sec execute 305 sec > 60 1840 1.78 MB/sec execute 306 sec > 60 1843 1.77 MB/sec execute 307 sec > 60 1846 1.77 MB/sec execute 308 sec > (1861) open /mnt1/dbench-test-dir/clients/client27/~dmtmp/PARADOX/STUDENTS.DB > failed for handle 10277 (Too many open files) > (1877) open /mnt1/dbench-test-dir/clients/client51/~dmtmp/PARADOX/__43904.DB > failed for handle 10283 (Too many open files) > (1878) ERROR: handle 10283 was not found > (1834) open /mnt1/dbench-test-dir/clients/client59/~dmtmp/PARADOX/COURSES.VAL > failed for handle 10269 (Too many open files) > (1835) ERROR: handle 10269 was not found > (1861) open /mnt1/dbench-test-dir/clients/client40/~dmtmp/PARADOX/STUDENTS.DB > failed for handle 10277 (Too many open files) > (1808) open /mnt1/dbench-test-dir/clients/client58/~dmtmp/PARADOX/__42B96.DB > failed for handle 10262 (Too many open files) > (1809) ERROR: handle 10262 was not found > (1954) open /mnt1/dbench-test-dir/clients/client29/~dmtmp/PARADOX/__43212.DB > failed for handle 10288 (Too many open files) > (1955) ERROR: handle 10288 was not found > (1868) open /mnt1/dbench-test-dir/clients/client5/~dmtmp/PARADOX/__43904.DB > failed for handle 10281 (Too many open files) > 60 1850 1.76 MB/sec execute 309 sec > Child failed with status 1 > > > Can anyone else reproduce this with Jeff's inode patch series? Strange... Is this reproducible without that patchset in place? -- Jeff Layton <jlayton@...> _______________________________________________ linux-cifs-client mailing list linux-cifs-client@... https://lists.samba.org/mailman/listinfo/linux-cifs-client |
|
|
Re: test statusHaven't had time to try yet, trying now
On Fri, Jun 26, 2009 at 12:27 PM, Jeff Layton<jlayton@...> wrote: > On Fri, 2009-06-26 at 11:52 -0500, Steve French wrote: >> Two test runs using "dbench -D /mnt/dbench-test-dir 60" (in this case >> the older dbench 3.04) : >> One against Samba 3.0.28a had the usual errors in the cleanup phase >> (which are hard to avoid due to their broken posix open support but we >> need to look into more) >> The test run against more current Samba server >> (3.4.0-GIT-b660e51-devel) failed though: >> >> 60 1834 1.79 MB/sec execute 304 sec >> 60 1837 1.79 MB/sec execute 305 sec >> 60 1840 1.78 MB/sec execute 306 sec >> 60 1843 1.77 MB/sec execute 307 sec >> 60 1846 1.77 MB/sec execute 308 sec >> (1861) open /mnt1/dbench-test-dir/clients/client27/~dmtmp/PARADOX/STUDENTS.DB >> failed for handle 10277 (Too many open files) >> (1877) open /mnt1/dbench-test-dir/clients/client51/~dmtmp/PARADOX/__43904.DB >> failed for handle 10283 (Too many open files) >> (1878) ERROR: handle 10283 was not found >> (1834) open /mnt1/dbench-test-dir/clients/client59/~dmtmp/PARADOX/COURSES.VAL >> failed for handle 10269 (Too many open files) >> (1835) ERROR: handle 10269 was not found >> (1861) open /mnt1/dbench-test-dir/clients/client40/~dmtmp/PARADOX/STUDENTS.DB >> failed for handle 10277 (Too many open files) >> (1808) open /mnt1/dbench-test-dir/clients/client58/~dmtmp/PARADOX/__42B96.DB >> failed for handle 10262 (Too many open files) >> (1809) ERROR: handle 10262 was not found >> (1954) open /mnt1/dbench-test-dir/clients/client29/~dmtmp/PARADOX/__43212.DB >> failed for handle 10288 (Too many open files) >> (1955) ERROR: handle 10288 was not found >> (1868) open /mnt1/dbench-test-dir/clients/client5/~dmtmp/PARADOX/__43904.DB >> failed for handle 10281 (Too many open files) >> 60 1850 1.76 MB/sec execute 309 sec >> Child failed with status 1 >> >> >> Can anyone else reproduce this with Jeff's inode patch series? > > Strange... > > Is this reproducible without that patchset in place? > > -- > Jeff Layton <jlayton@...> > > -- Thanks, Steve _______________________________________________ linux-cifs-client mailing list linux-cifs-client@... https://lists.samba.org/mailman/listinfo/linux-cifs-client |
|
|
Re: test statusOn Fri, 2009-06-26 at 12:29 -0500, Steve French wrote:
> Haven't had time to try yet, trying now > > On Fri, Jun 26, 2009 at 12:27 PM, Jeff Layton<jlayton@...> wrote: > > On Fri, 2009-06-26 at 11:52 -0500, Steve French wrote: > >> Two test runs using "dbench -D /mnt/dbench-test-dir 60" (in this case > >> the older dbench 3.04) : > >> One against Samba 3.0.28a had the usual errors in the cleanup phase > >> (which are hard to avoid due to their broken posix open support but we > >> need to look into more) > >> The test run against more current Samba server > >> (3.4.0-GIT-b660e51-devel) failed though: > >> > >> 60 1834 1.79 MB/sec execute 304 sec > >> 60 1837 1.79 MB/sec execute 305 sec > >> 60 1840 1.78 MB/sec execute 306 sec > >> 60 1843 1.77 MB/sec execute 307 sec > >> 60 1846 1.77 MB/sec execute 308 sec > >> (1861) open /mnt1/dbench-test-dir/clients/client27/~dmtmp/PARADOX/STUDENTS.DB > >> failed for handle 10277 (Too many open files) > >> (1877) open /mnt1/dbench-test-dir/clients/client51/~dmtmp/PARADOX/__43904.DB > >> failed for handle 10283 (Too many open files) > >> (1878) ERROR: handle 10283 was not found > >> (1834) open /mnt1/dbench-test-dir/clients/client59/~dmtmp/PARADOX/COURSES.VAL > >> failed for handle 10269 (Too many open files) > >> (1835) ERROR: handle 10269 was not found > >> (1861) open /mnt1/dbench-test-dir/clients/client40/~dmtmp/PARADOX/STUDENTS.DB > >> failed for handle 10277 (Too many open files) > >> (1808) open /mnt1/dbench-test-dir/clients/client58/~dmtmp/PARADOX/__42B96.DB > >> failed for handle 10262 (Too many open files) > >> (1809) ERROR: handle 10262 was not found > >> (1954) open /mnt1/dbench-test-dir/clients/client29/~dmtmp/PARADOX/__43212.DB > >> failed for handle 10288 (Too many open files) > >> (1955) ERROR: handle 10288 was not found > >> (1868) open /mnt1/dbench-test-dir/clients/client5/~dmtmp/PARADOX/__43904.DB > >> failed for handle 10281 (Too many open files) > >> 60 1850 1.76 MB/sec execute 309 sec > >> Child failed with status 1 > >> > >> > >> Can anyone else reproduce this with Jeff's inode patch series? > > > > Strange... > > > > Is this reproducible without that patchset in place? > > > > -- > > Jeff Layton <jlayton@...> > > > > > > > I wonder whether the server recorded any errors during this period? Looks like we translate ERRnofids from the server as EMFILE. -- Jeff Layton <jlayton@...> _______________________________________________ linux-cifs-client mailing list linux-cifs-client@... https://lists.samba.org/mailman/listinfo/linux-cifs-client |
|
|
Re: Re: test statusOn Fri, 2009-06-26 at 12:29 -0500, Steve French wrote:
> >> The test run against more current Samba server > >> (3.4.0-GIT-b660e51-devel) failed though: > >> > >> 60 1834 1.79 MB/sec execute 304 sec > >> 60 1837 1.79 MB/sec execute 305 sec > >> 60 1840 1.78 MB/sec execute 306 sec > >> 60 1843 1.77 MB/sec execute 307 sec > >> 60 1846 1.77 MB/sec execute 308 sec > >> (1861) open /mnt1/dbench-test-dir/clients/client27/~dmtmp/PARADOX/STUDENTS.DB > >> failed for handle 10277 (Too many open files) > >> (1877) open /mnt1/dbench-test-dir/clients/client51/~dmtmp/PARADOX/__43904.DB > >> failed for handle 10283 (Too many open files) Glancing through the code, looks like we translate ERRnofids from the server to EMFILE. Did the server happen to log any errors during this? -- Jeff Layton <jlayton@...> _______________________________________________ linux-cifs-client mailing list linux-cifs-client@... https://lists.samba.org/mailman/listinfo/linux-cifs-client |
|
|
Re: Re: test statusIt looks like this fails without your two inode patches as well,
although it took a little longer. I also updated the server to current git (samba master branch, Version 3.4.0-GIT-382a937-devel) and it still fails. I am going to add some temporary debug code to dump the number of files we think are open (to dmesg on the client) when the error occurs. I didn't spot any interesting log entries (server side) yet. Near the time of the failure, running smbstatus on the server showed about 1000 open files (they rapidly decrease after that as the dbench processes exit and handles are freed). On Fri, Jun 26, 2009 at 2:26 PM, Jeff Layton<jlayton@...> wrote: > On Fri, 2009-06-26 at 12:29 -0500, Steve French wrote: > >> >> The test run against more current Samba server >> >> (3.4.0-GIT-b660e51-devel) failed though: >> >> >> >> 60 1834 1.79 MB/sec execute 304 sec >> >> 60 1837 1.79 MB/sec execute 305 sec >> >> 60 1840 1.78 MB/sec execute 306 sec >> >> 60 1843 1.77 MB/sec execute 307 sec >> >> 60 1846 1.77 MB/sec execute 308 sec >> >> (1861) open /mnt1/dbench-test-dir/clients/client27/~dmtmp/PARADOX/STUDENTS.DB >> >> failed for handle 10277 (Too many open files) >> >> (1877) open /mnt1/dbench-test-dir/clients/client51/~dmtmp/PARADOX/__43904.DB >> >> failed for handle 10283 (Too many open files) > > Glancing through the code, looks like we translate ERRnofids from the > server to EMFILE. Did the server happen to log any errors during this? > > -- > Jeff Layton <jlayton@...> > > -- Thanks, Steve _______________________________________________ linux-cifs-client mailing list linux-cifs-client@... https://lists.samba.org/mailman/listinfo/linux-cifs-client |
|
|
Re: Re: test statusOn Fri, 2009-06-26 at 14:39 -0500, Steve French wrote:
> It looks like this fails without your two inode patches as well, > although it took a little longer. I also updated the server to > current git (samba master branch, Version 3.4.0-GIT-382a937-devel) and > it still fails. I am going to add some temporary debug code to dump > the number of files we think are open (to dmesg on the client) when > the error occurs. > > I didn't spot any interesting log entries (server side) yet. > > Near the time of the failure, running smbstatus on the server showed > about 1000 open files (they rapidly decrease after that as the dbench > processes exit and handles are freed). > Ok, good to know that it's not a regression. Might also be interesting to get a wire capture of the run during the time that the client starts returning errors and see whether the server is sending ERRnofids. Maybe also a run of lsof against the smbd process just to see whether it might have a ton of files open? -- Jeff Layton <jlayton@...> _______________________________________________ linux-cifs-client mailing list linux-cifs-client@... https://lists.samba.org/mailman/listinfo/linux-cifs-client |
|
|
|
|
|
Re: test statusWorked with dbench 50 (50 processes) but failed with 55.
On Fri, Jun 26, 2009 at 7:48 PM, Steve French<smfrench@...> wrote: > Current linux cifs client (with or without Jeff's patches) mounted to > current Samba 3 server (master branch): > > Typical failure (SMB response): > > failing smb: : dump of 48 bytes of data at 0xc7050480 > 00000023 424d53ff 00011f32 c00180c0 # . . . � S M B 2 . . . � . . > 00000000 00000000 00000000 70d60001 . . . . . . . . . . . . . . � p > 915d0064 00000000 00000a00 00000040 d . ] . . . . . . . . . @ . . . > > On the client I added debug code and dumped the following when the > first EMFILE is returned: > Number of Opens (that succeeded): 15777 Number of closes: 14763 > > This indicates the 1015th open is being rejected by Samba server - has > Samba server changed its max open file limit? > > -- > Thanks, > > Steve > -- Thanks, Steve _______________________________________________ linux-cifs-client mailing list linux-cifs-client@... https://lists.samba.org/mailman/listinfo/linux-cifs-client |
|
|
Re: Re: test statusOn Fri, 2009-06-26 at 21:19 -0500, Steve French wrote:
> Worked with dbench 50 (50 processes) but failed with 55. > > On Fri, Jun 26, 2009 at 7:48 PM, Steve French<smfrench@...> wrote: > > Current linux cifs client (with or without Jeff's patches) mounted to > > current Samba 3 server (master branch): > > > > Typical failure (SMB response): > > > > failing smb: : dump of 48 bytes of data at 0xc7050480 > > 00000023 424d53ff 00011f32 c00180c0 # . . . � S M B 2 . . . � . . > > 00000000 00000000 00000000 70d60001 . . . . . . . . . . . . . . � p > > 915d0064 00000000 00000a00 00000040 d . ] . . . . . . . . . @ . . . > > > > On the client I added debug code and dumped the following when the > > first EMFILE is returned: > > Number of Opens (that succeeded): 15777 Number of closes: 14763 > > > > This indicates the 1015th open is being rejected by Samba server - has > > Samba server changed its max open file limit? > > > > -- > > Thanks, > > > > Steve > > I did some testing with dbench too and saw failures and deadlocks when testing against samba-3.4.0-0pre1.37.fc12.x86_64 in the fedora rawhide repo. It seems to be better after applying the patch that I just posted for the fh_mutex regression. -- Jeff Layton <jlayton@...> _______________________________________________ linux-cifs-client mailing list linux-cifs-client@... https://lists.samba.org/mailman/listinfo/linux-cifs-client |
| Free embeddable forum powered by Nabble | Forum Help |