On Tue, 30 Jul 2019 at 16:37, Diego Remolina <dijuremo@xxxxxxxxx> wrote:
This option is enabled. In which version has this been patched? This is a file server and disabling readdir-ahead will have a hard impact on performance.
This was fixed in 5.3 (https://bugzilla.redhat.com/show_bug.cgi?id=1659676).
This bug is only relevant if the gluster fuse client is the one that is using up memory.
The first thing to do would be to determine which process is using up the memory and to get a statedump.
ps <pid> should give you the details of the gluster process .
Regards,
Nithya
[root@ysmha01 ~]# gluster v get export readdir-ahead
Option Value
------ -----
performance.readdir-ahead on
The guide recommends enabling the setting:
https://docs.gluster.org/en/latest/Administrator%20Guide/Accessing%20Gluster%20from%20Windows/DiegoOn Mon, Jul 29, 2019 at 11:52 PM Nithya Balachandran <nbalacha@xxxxxxxxxx> wrote:Hi Diego,Please do the following:gluster v get <volname> readdir-aheadIf this is enabled, please disable it and see if it helps. There was a leak in the opendir codpath that was fixed in later released.Regards,NithyaOn Tue, 30 Jul 2019 at 09:04, Diego Remolina <dijuremo@xxxxxxxxx> wrote:Will this kill the actual process or simply trigger the dump? Which process should I kill? The brick process in the system or the fuse mount?DiegoOn Mon, Jul 29, 2019, 23:27 Nithya Balachandran <nbalacha@xxxxxxxxxx> wrote:On Tue, 30 Jul 2019 at 05:44, Diego Remolina <dijuremo@xxxxxxxxx> wrote:Unfortunately statedump crashes on both machines, even freshly rebooted.Do you see any statedump files in /var/run/gluster? This looks more like the gluster cli crashed.[root@ysmha01 ~]# gluster --print-statedumpdir
/var/run/gluster
[root@ysmha01 ~]# gluster v statedump export
Segmentation fault (core dumped)[root@ysmha02 ~]# uptime
20:12:20 up 6 min, 1 user, load average: 0.72, 0.52, 0.24
[root@ysmha02 ~]# gluster --print-statedumpdir
/var/run/gluster
[root@ysmha02 ~]# gluster v statedump export
Segmentation fault (core dumped)I rebooted today after 40 days. Gluster was eating up shy of 40GB of RAM out of 64.What would you recommend to be the next step?Diego_______________________________________________On Mon, Mar 4, 2019 at 5:07 AM Poornima Gurusiddaiah <pgurusid@xxxxxxxxxx> wrote:Could you also provide the statedump of the gluster process consuming 44G ram [1]. Please make sure the statedump is taken when the memory consumption is very high, like 10s of GBs, otherwise we may not be able to identify the issue. Also i see that the cache size is 10G is that something you arrived at, after doing some tests? Its relatively higher than normal.On Mon, Mar 4, 2019 at 12:23 AM Diego Remolina <dijuremo@xxxxxxxxx> wrote:Hi,I will not be able to test gluster-6rc because this is a production environment and it takes several days for memory to grow a lot.The Samba server is hosting all types of files, small and large from small roaming profile type files to bigger files like adobe suite, autodesk Revit (file sizes in the hundreds of megabytes).As I stated before, this same issue was present back with 3.8.x which I was running before.The information you requested:[root@ysmha02 ~]# gluster v info exportVolume Name: exportType: ReplicateVolume ID: b4353b3f-6ef6-4813-819a-8e85e5a95cffStatus: StartedSnapshot Count: 0Number of Bricks: 1 x 2 = 2Transport-type: tcpBricks:Brick1: 10.0.1.7:/bricks/hdds/brickBrick2: 10.0.1.6:/bricks/hdds/brickOptions Reconfigured:performance.stat-prefetch: onperformance.cache-min-file-size: 0network.inode-lru-limit: 65536performance.cache-invalidation: onfeatures.cache-invalidation: onperformance.md-cache-timeout: 600features.cache-invalidation-timeout: 600performance.cache-samba-metadata: ontransport.address-family: inetserver.allow-insecure: onperformance.cache-size: 10GBcluster.server-quorum-type: servernfs.disable: onperformance.io-thread-count: 64performance.io-cache: oncluster.lookup-optimize: oncluster.readdir-optimize: onserver.event-threads: 5client.event-threads: 5performance.cache-max-file-size: 256MBdiagnostics.client-log-level: INFOdiagnostics.brick-log-level: INFOcluster.server-quorum-ratio: 51%On Fri, Mar 1, 2019 at 11:07 PM Poornima Gurusiddaiah <pgurusid@xxxxxxxxxx> wrote:This high memory consumption is not normal. Looks like it's a memory leak. Is it possible to try it on test setup with gluster-6rc? What is the kind of workload that goes into fuse mount? Large files or small files? We need the following information to debug further:- Gluster volume info output- Statedump of the Gluster fuse mount process consuming 44G ram.Regards,PoornimaOn Sat, Mar 2, 2019, 3:40 AM Diego Remolina <dijuremo@xxxxxxxxx> wrote:_______________________________________________I am using glusterfs with two servers as a file server sharing files via samba and ctdb. I cannot use samba vfs gluster plugin, due to bug in current Centos version of samba. So I am mounting via fuse and exporting the volume to samba from the mount point.Upon initial boot, the server where samba is exporting files climbs up to ~10GB RAM within a couple hours of use. From then on, it is a constant slow memory increase. In the past with gluster 3.8.x we had to reboot the servers at around 30 days . With gluster 4.1.6 we are getting up to 48 days, but RAM use is at 48GB out of 64GB. Is this normal?The particular versions are below,[root@ysmha01 home]# uptime16:59:39 up 48 days, 9:56, 1 user, load average: 3.75, 3.17, 3.00[root@ysmha01 home]# rpm -qa | grep glustercentos-release-gluster41-1.0-3.el7.centos.noarchglusterfs-server-4.1.6-1.el7.x86_64glusterfs-api-4.1.6-1.el7.x86_64centos-release-gluster-legacy-4.0-2.el7.centos.noarchglusterfs-4.1.6-1.el7.x86_64glusterfs-client-xlators-4.1.6-1.el7.x86_64libvirt-daemon-driver-storage-gluster-3.9.0-14.el7_5.8.x86_64glusterfs-fuse-4.1.6-1.el7.x86_64glusterfs-libs-4.1.6-1.el7.x86_64glusterfs-rdma-4.1.6-1.el7.x86_64glusterfs-cli-4.1.6-1.el7.x86_64samba-vfs-glusterfs-4.8.3-4.el7.x86_64[root@ysmha01 home]# rpm -qa | grep sambasamba-common-tools-4.8.3-4.el7.x86_64samba-client-libs-4.8.3-4.el7.x86_64samba-libs-4.8.3-4.el7.x86_64samba-4.8.3-4.el7.x86_64samba-common-libs-4.8.3-4.el7.x86_64samba-common-4.8.3-4.el7.noarchsamba-vfs-glusterfs-4.8.3-4.el7.x86_64[root@ysmha01 home]# cat /etc/redhat-releaseCentOS Linux release 7.6.1810 (Core)RAM view using topTasks: 398 total, 1 running, 397 sleeping, 0 stopped, 0 zombie%Cpu(s): 7.0 us, 9.3 sy, 1.7 ni, 71.6 id, 9.7 wa, 0.0 hi, 0.8 si, 0.0 stKiB Mem : 65772000 total, 1851344 free, 60487404 used, 3433252 buff/cacheKiB Swap: 0 total, 0 free, 0 used. 3134316 avail MemPID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND9953 root 20 0 3727912 946496 3196 S 150.2 1.4 38626:27 glusterfsd9634 root 20 0 48.1g 47.2g 3184 S 96.3 75.3 29513:55 glusterfs14485 root 20 0 3404140 63780 2052 S 80.7 0.1 1590:13 glusterfs[root@ysmha01 ~]# gluster v status exportStatus of volume: exportGluster process TCP Port RDMA Port Online Pid------------------------------------------------------------------------------Brick 10.0.1.7:/bricks/hdds/brick 49157 0 Y 13986Brick 10.0.1.6:/bricks/hdds/brick 49153 0 Y 9953Self-heal Daemon on localhost N/A N/A Y 14485Self-heal Daemon on 10.0.1.7 N/A N/A Y 21934Self-heal Daemon on 10.0.1.5 N/A N/A Y 4598Task Status of Volume export------------------------------------------------------------------------------There are no active volume tasks
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
https://lists.gluster.org/mailman/listinfo/gluster-users
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
https://lists.gluster.org/mailman/listinfo/gluster-users
_______________________________________________ Gluster-users mailing list Gluster-users@xxxxxxxxxxx https://lists.gluster.org/mailman/listinfo/gluster-users
