Dmitry Isakbayev
2018-Dec-28 20:37 UTC
[Gluster-users] java application crushes while reading a zip file
These 3 options seem to trigger both (reading zip file and renaming files) problems. Options Reconfigured: performance.io-cache: off performance.stat-prefetch: off performance.quick-read: off performance.parallel-readdir: off *performance.readdir-ahead: on* *performance.write-behind: on* *performance.read-ahead: on* performance.client-io-threads: off nfs.disable: on transport.address-family: inet On Fri, Dec 28, 2018 at 10:24 AM Dmitry Isakbayev <isakdim at gmail.com> wrote:> Turning a single option on at a time still worked fine. I will keep > trying. > > We had used 4.1.5 on KVM/CentOS7.5 at AWS without these issues or log > messages. Do you suppose these issues are triggered by the new environment > or did not exist in 4.1.5? > > [root at node1 ~]# glusterfs --version > glusterfs 4.1.5 > > On AWS using > [root at node1 ~]# hostnamectl > Static hostname: node1 > Icon name: computer-vm > Chassis: vm > Machine ID: b30d0f2110ac3807b210c19ede3ce88f > Boot ID: 52bb159a0aa94043a40e7c7651967bd9 > Virtualization: kvm > Operating System: CentOS Linux 7 (Core) > CPE OS Name: cpe:/o:centos:centos:7 > Kernel: Linux 3.10.0-862.3.2.el7.x86_64 > Architecture: x86-64 > > > > > On Fri, Dec 28, 2018 at 8:56 AM Raghavendra Gowdappa <rgowdapp at redhat.com> > wrote: > >> >> >> On Fri, Dec 28, 2018 at 7:23 PM Dmitry Isakbayev <isakdim at gmail.com> >> wrote: >> >>> Ok. I will try different options. >>> >>> This system is scheduled to go into production soon. What version would >>> you recommend to roll back to? >>> >> >> These are long standing issues. So, rolling back may not make these >> issues go away. Instead if you think performance is agreeable to you, >> please keep these xlators off in production. >> >> >>> On Thu, Dec 27, 2018 at 10:55 PM Raghavendra Gowdappa < >>> rgowdapp at redhat.com> wrote: >>> >>>> >>>> >>>> On Fri, Dec 28, 2018 at 3:13 AM Dmitry Isakbayev <isakdim at gmail.com> >>>> wrote: >>>> >>>>> Raghavendra, >>>>> >>>>> Thank for the suggestion. >>>>> >>>>> >>>>> I am suing >>>>> >>>>> [root at jl-fanexoss1p glusterfs]# gluster --version >>>>> glusterfs 5.0 >>>>> >>>>> On >>>>> [root at jl-fanexoss1p glusterfs]# hostnamectl >>>>> Icon name: computer-vm >>>>> Chassis: vm >>>>> Machine ID: e44b8478ef7a467d98363614f4e50535 >>>>> Boot ID: eed98992fdda4c88bdd459a89101766b >>>>> Virtualization: vmware >>>>> Operating System: Red Hat Enterprise Linux Server 7.5 (Maipo) >>>>> CPE OS Name: cpe:/o:redhat:enterprise_linux:7.5:GA:server >>>>> Kernel: Linux 3.10.0-862.14.4.el7.x86_64 >>>>> Architecture: x86-64 >>>>> >>>>> >>>>> I have configured the following options >>>>> >>>>> [root at jl-fanexoss1p glusterfs]# gluster volume info >>>>> Volume Name: gv0 >>>>> Type: Replicate >>>>> Volume ID: 5ffbda09-c5e2-4abc-b89e-79b5d8a40824 >>>>> Status: Started >>>>> Snapshot Count: 0 >>>>> Number of Bricks: 1 x 3 = 3 >>>>> Transport-type: tcp >>>>> Bricks: >>>>> Brick1: jl-fanexoss1p.cspire.net:/data/brick1/gv0 >>>>> Brick2: sl-fanexoss2p.cspire.net:/data/brick1/gv0 >>>>> Brick3: nxquorum1p.cspire.net:/data/brick1/gv0 >>>>> Options Reconfigured: >>>>> performance.io-cache: off >>>>> performance.stat-prefetch: off >>>>> performance.quick-read: off >>>>> performance.parallel-readdir: off >>>>> performance.readdir-ahead: off >>>>> performance.write-behind: off >>>>> performance.read-ahead: off >>>>> performance.client-io-threads: off >>>>> nfs.disable: on >>>>> transport.address-family: inet >>>>> >>>>> I don't know if it is related, but I am seeing a lot of >>>>> [2018-12-27 20:19:23.776080] W [MSGID: 114031] >>>>> [client-rpc-fops_v2.c:1932:client4_0_seek_cbk] 2-gv0-client-0: remote >>>>> operation failed [No such device or address] >>>>> [2018-12-27 20:19:47.735190] E [MSGID: 101191] >>>>> [event-epoll.c:671:event_dispatch_epoll_worker] 2-epoll: Failed to dispatch >>>>> handler >>>>> >>>> >>>> These msgs were introduced by patch [1]. To the best of my knowledge >>>> they are benign. We'll be sending a patch to fix these msgs though. >>>> >>>> +Mohit Agrawal <moagrawa at redhat.com> +Milind Changire >>>> <mchangir at redhat.com> . Can you try to identify why we are seeing >>>> these messages? If possible please send a patch to fix this. >>>> >>>> [1] >>>> https://review.gluster.org/r/I578c3fc67713f4234bd3abbec5d3fbba19059ea5 >>>> >>>> >>>>> And java.io exceptions trying to rename files. >>>>> >>>> >>>> When you see the errors is it possible to collect, >>>> * strace of the java application (strace -ff -v ...) >>>> * fuse-dump of the glusterfs mount (use option --dump-fuse while >>>> mounting)? >>>> >>>> I also need another favour from you. By trail and error, can you point >>>> out which of the many performance xlators you've turned off is causing the >>>> issue? >>>> >>>> The above two data-points will help us to fix the problem. >>>> >>>> >>>>> Thank You, >>>>> Dmitry >>>>> >>>>> >>>>> On Thu, Dec 27, 2018 at 3:48 PM Raghavendra Gowdappa < >>>>> rgowdapp at redhat.com> wrote: >>>>> >>>>>> What version of glusterfs are you using? It might be either >>>>>> * a stale metadata issue. >>>>>> * inconsistent ctime issue. >>>>>> >>>>>> Can you try turning off all performance xlators? If the issue is 1, >>>>>> that should help. >>>>>> >>>>>> On Fri, Dec 28, 2018 at 1:51 AM Dmitry Isakbayev <isakdim at gmail.com> >>>>>> wrote: >>>>>> >>>>>>> Attempted to set 'performance.read-ahead off` according to >>>>>>> https://jira.apache.org/jira/browse/AMQ-7041 >>>>>>> That did not help. >>>>>>> >>>>>>> On Mon, Dec 24, 2018 at 2:11 PM Dmitry Isakbayev <isakdim at gmail.com> >>>>>>> wrote: >>>>>>> >>>>>>>> The core file generated by JVM suggests that it happens because the >>>>>>>> file is changing while it is being read - >>>>>>>> https://bugs.java.com/bugdatabase/view_bug.do?bug_id=8186557. >>>>>>>> The application reads in the zipfile and goes through the zip >>>>>>>> entries, then reloads the file and goes the zip entries again. It does so >>>>>>>> 3 times. The application never crushes on the 1st cycle but sometimes >>>>>>>> crushes on the 2nd or 3rd cycle. >>>>>>>> The zip file is generated about 20 seconds prior to it being used >>>>>>>> and is not updated or even used by any other application. I have never >>>>>>>> seen this problem on a plain file system. >>>>>>>> >>>>>>>> I would appreciate any suggestions on how to go debugging this >>>>>>>> issue. I can change the source code of the java application. >>>>>>>> >>>>>>>> Regards, >>>>>>>> Dmitry >>>>>>>> >>>>>>>> >>>>>>>> _______________________________________________ >>>>>>> Gluster-users mailing list >>>>>>> Gluster-users at gluster.org >>>>>>> https://lists.gluster.org/mailman/listinfo/gluster-users >>>>>> >>>>>>-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20181228/d5fb862e/attachment.html>
Raghavendra Gowdappa
2018-Dec-29 23:46 UTC
[Gluster-users] java application crushes while reading a zip file
Thanks Dmitry. Can you provide the following debug info I asked earlier: * strace -ff -v ... of java application * dump of the I/O traffic seen by the mountpoint (use --dump-fuse while mounting). regards, Raghavendra On Sat, Dec 29, 2018 at 2:08 AM Dmitry Isakbayev <isakdim at gmail.com> wrote:> These 3 options seem to trigger both (reading zip file and renaming files) > problems. > > Options Reconfigured: > performance.io-cache: off > performance.stat-prefetch: off > performance.quick-read: off > performance.parallel-readdir: off > *performance.readdir-ahead: on* > *performance.write-behind: on* > *performance.read-ahead: on* > performance.client-io-threads: off > nfs.disable: on > transport.address-family: inet > > > On Fri, Dec 28, 2018 at 10:24 AM Dmitry Isakbayev <isakdim at gmail.com> > wrote: > >> Turning a single option on at a time still worked fine. I will keep >> trying. >> >> We had used 4.1.5 on KVM/CentOS7.5 at AWS without these issues or log >> messages. Do you suppose these issues are triggered by the new environment >> or did not exist in 4.1.5? >> >> [root at node1 ~]# glusterfs --version >> glusterfs 4.1.5 >> >> On AWS using >> [root at node1 ~]# hostnamectl >> Static hostname: node1 >> Icon name: computer-vm >> Chassis: vm >> Machine ID: b30d0f2110ac3807b210c19ede3ce88f >> Boot ID: 52bb159a0aa94043a40e7c7651967bd9 >> Virtualization: kvm >> Operating System: CentOS Linux 7 (Core) >> CPE OS Name: cpe:/o:centos:centos:7 >> Kernel: Linux 3.10.0-862.3.2.el7.x86_64 >> Architecture: x86-64 >> >> >> >> >> On Fri, Dec 28, 2018 at 8:56 AM Raghavendra Gowdappa <rgowdapp at redhat.com> >> wrote: >> >>> >>> >>> On Fri, Dec 28, 2018 at 7:23 PM Dmitry Isakbayev <isakdim at gmail.com> >>> wrote: >>> >>>> Ok. I will try different options. >>>> >>>> This system is scheduled to go into production soon. What version >>>> would you recommend to roll back to? >>>> >>> >>> These are long standing issues. So, rolling back may not make these >>> issues go away. Instead if you think performance is agreeable to you, >>> please keep these xlators off in production. >>> >>> >>>> On Thu, Dec 27, 2018 at 10:55 PM Raghavendra Gowdappa < >>>> rgowdapp at redhat.com> wrote: >>>> >>>>> >>>>> >>>>> On Fri, Dec 28, 2018 at 3:13 AM Dmitry Isakbayev <isakdim at gmail.com> >>>>> wrote: >>>>> >>>>>> Raghavendra, >>>>>> >>>>>> Thank for the suggestion. >>>>>> >>>>>> >>>>>> I am suing >>>>>> >>>>>> [root at jl-fanexoss1p glusterfs]# gluster --version >>>>>> glusterfs 5.0 >>>>>> >>>>>> On >>>>>> [root at jl-fanexoss1p glusterfs]# hostnamectl >>>>>> Icon name: computer-vm >>>>>> Chassis: vm >>>>>> Machine ID: e44b8478ef7a467d98363614f4e50535 >>>>>> Boot ID: eed98992fdda4c88bdd459a89101766b >>>>>> Virtualization: vmware >>>>>> Operating System: Red Hat Enterprise Linux Server 7.5 (Maipo) >>>>>> CPE OS Name: cpe:/o:redhat:enterprise_linux:7.5:GA:server >>>>>> Kernel: Linux 3.10.0-862.14.4.el7.x86_64 >>>>>> Architecture: x86-64 >>>>>> >>>>>> >>>>>> I have configured the following options >>>>>> >>>>>> [root at jl-fanexoss1p glusterfs]# gluster volume info >>>>>> Volume Name: gv0 >>>>>> Type: Replicate >>>>>> Volume ID: 5ffbda09-c5e2-4abc-b89e-79b5d8a40824 >>>>>> Status: Started >>>>>> Snapshot Count: 0 >>>>>> Number of Bricks: 1 x 3 = 3 >>>>>> Transport-type: tcp >>>>>> Bricks: >>>>>> Brick1: jl-fanexoss1p.cspire.net:/data/brick1/gv0 >>>>>> Brick2: sl-fanexoss2p.cspire.net:/data/brick1/gv0 >>>>>> Brick3: nxquorum1p.cspire.net:/data/brick1/gv0 >>>>>> Options Reconfigured: >>>>>> performance.io-cache: off >>>>>> performance.stat-prefetch: off >>>>>> performance.quick-read: off >>>>>> performance.parallel-readdir: off >>>>>> performance.readdir-ahead: off >>>>>> performance.write-behind: off >>>>>> performance.read-ahead: off >>>>>> performance.client-io-threads: off >>>>>> nfs.disable: on >>>>>> transport.address-family: inet >>>>>> >>>>>> I don't know if it is related, but I am seeing a lot of >>>>>> [2018-12-27 20:19:23.776080] W [MSGID: 114031] >>>>>> [client-rpc-fops_v2.c:1932:client4_0_seek_cbk] 2-gv0-client-0: remote >>>>>> operation failed [No such device or address] >>>>>> [2018-12-27 20:19:47.735190] E [MSGID: 101191] >>>>>> [event-epoll.c:671:event_dispatch_epoll_worker] 2-epoll: Failed to dispatch >>>>>> handler >>>>>> >>>>> >>>>> These msgs were introduced by patch [1]. To the best of my knowledge >>>>> they are benign. We'll be sending a patch to fix these msgs though. >>>>> >>>>> +Mohit Agrawal <moagrawa at redhat.com> +Milind Changire >>>>> <mchangir at redhat.com> . Can you try to identify why we are seeing >>>>> these messages? If possible please send a patch to fix this. >>>>> >>>>> [1] >>>>> https://review.gluster.org/r/I578c3fc67713f4234bd3abbec5d3fbba19059ea5 >>>>> >>>>> >>>>>> And java.io exceptions trying to rename files. >>>>>> >>>>> >>>>> When you see the errors is it possible to collect, >>>>> * strace of the java application (strace -ff -v ...) >>>>> * fuse-dump of the glusterfs mount (use option --dump-fuse while >>>>> mounting)? >>>>> >>>>> I also need another favour from you. By trail and error, can you point >>>>> out which of the many performance xlators you've turned off is causing the >>>>> issue? >>>>> >>>>> The above two data-points will help us to fix the problem. >>>>> >>>>> >>>>>> Thank You, >>>>>> Dmitry >>>>>> >>>>>> >>>>>> On Thu, Dec 27, 2018 at 3:48 PM Raghavendra Gowdappa < >>>>>> rgowdapp at redhat.com> wrote: >>>>>> >>>>>>> What version of glusterfs are you using? It might be either >>>>>>> * a stale metadata issue. >>>>>>> * inconsistent ctime issue. >>>>>>> >>>>>>> Can you try turning off all performance xlators? If the issue is 1, >>>>>>> that should help. >>>>>>> >>>>>>> On Fri, Dec 28, 2018 at 1:51 AM Dmitry Isakbayev <isakdim at gmail.com> >>>>>>> wrote: >>>>>>> >>>>>>>> Attempted to set 'performance.read-ahead off` according to >>>>>>>> https://jira.apache.org/jira/browse/AMQ-7041 >>>>>>>> That did not help. >>>>>>>> >>>>>>>> On Mon, Dec 24, 2018 at 2:11 PM Dmitry Isakbayev <isakdim at gmail.com> >>>>>>>> wrote: >>>>>>>> >>>>>>>>> The core file generated by JVM suggests that it happens because >>>>>>>>> the file is changing while it is being read - >>>>>>>>> https://bugs.java.com/bugdatabase/view_bug.do?bug_id=8186557. >>>>>>>>> The application reads in the zipfile and goes through the zip >>>>>>>>> entries, then reloads the file and goes the zip entries again. It does so >>>>>>>>> 3 times. The application never crushes on the 1st cycle but sometimes >>>>>>>>> crushes on the 2nd or 3rd cycle. >>>>>>>>> The zip file is generated about 20 seconds prior to it being used >>>>>>>>> and is not updated or even used by any other application. I have never >>>>>>>>> seen this problem on a plain file system. >>>>>>>>> >>>>>>>>> I would appreciate any suggestions on how to go debugging this >>>>>>>>> issue. I can change the source code of the java application. >>>>>>>>> >>>>>>>>> Regards, >>>>>>>>> Dmitry >>>>>>>>> >>>>>>>>> >>>>>>>>> _______________________________________________ >>>>>>>> Gluster-users mailing list >>>>>>>> Gluster-users at gluster.org >>>>>>>> https://lists.gluster.org/mailman/listinfo/gluster-users >>>>>>> >>>>>>>-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20181230/34c8b9b0/attachment.html>