Hello,
I wanted to report that I had this morning a similar issue on another server where a few PHP-FPM processes get blocked on different GlusterFS volume mounted through a FUSE mount. This GlusterFS volume has no quota enabled so it might not be quota related after all.
Here would be the Linux kernel stack trace:
[Sun Sep 2 06:47:47 2018] INFO: task php5-fpm:25880 blocked for more than 120 seconds.
[Sun Sep 2 06:47:47 2018] Not tainted 3.16.0-4-amd64 #1
[Sun Sep 2 06:47:47 2018] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[Sun Sep 2 06:47:47 2018] php5-fpm D ffff88017ee12f40 0 25880 1 0x00000004
[Sun Sep 2 06:47:47 2018] ffff880101688b60 0000000000000282 0000000000012f40 ffff880059ca3fd8
[Sun Sep 2 06:47:47 2018] 0000000000012f40 ffff880101688b60 ffff8801093b51b0 ffff8801067ec800
[Sun Sep 2 06:47:47 2018] ffff880059ca3cc0 ffff8801093b5290 ffff8801093b51b0 ffff880059ca3e80
[Sun Sep 2 06:47:47 2018] Call Trace:
[Sun Sep 2 06:47:47 2018] [<ffffffffa00cd50d>] ? __fuse_request_send+0xbd/0x270 [fuse]
[Sun Sep 2 06:47:47 2018] [<ffffffff810a9610>] ? prepare_to_wait_event+0xf0/0xf0
[Sun Sep 2 06:47:47 2018] [<ffffffffa00d6990>] ? fuse_send_write+0xd0/0x100 [fuse]
[Sun Sep 2 06:47:47 2018] [<ffffffffa00d82af>] ? fuse_perform_write+0x26f/0x4b0 [fuse]
[Sun Sep 2 06:47:47 2018] [<ffffffffa00d86cd>] ? fuse_file_write_iter+0x1dd/0x2b0 [fuse]
[Sun Sep 2 06:47:47 2018] [<ffffffff811aa7f4>] ? new_sync_write+0x74/0xa0
[Sun Sep 2 06:47:47 2018] [<ffffffff811aaf32>] ? vfs_write+0xb2/0x1f0
[Sun Sep 2 06:47:47 2018] [<ffffffff811aadfd>] ? vfs_read+0xed/0x170
[Sun Sep 2 06:47:47 2018] [<ffffffff811aba72>] ? SyS_write+0x42/0xa0
[Sun Sep 2 06:47:47 2018] [<ffffffff811aaa8e>] ? SyS_lseek+0x7e/0xa0
[Sun Sep 2 06:47:47 2018] [<ffffffff8151a48d>] ? system_call_fast_compare_end+0x10/0x15
Did anyone already have time to have a look at the statedump file I sent around 3 weeks ago?
I never saw this type of problems in the past and it started to appear since I upgraded to GluterFS 3.12.12.
Best regards,
Mabi
‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
On August 15, 2018 9:21 AM, mabi <mabi@xxxxxxxxxxxxx> wrote:
Great, you will then find attached here the statedump of the client using the FUSE glusterfs mount right after two processes have blocked.Two notes here regarding the "path=" in this statedump file:- I have renamed all the "path=" which has the problematic directory as "path=PROBLEMATIC_DIRECTORY_HERE- All the other "path=" I have renamed them to "path=REMOVED_FOR_PRIVACY".Note also that funnily enough the number of "path=" for that problematic directory sums up to exactly 5000 entries. Coincidence or hint to the problem maybe?‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐On August 15, 2018 5:21 AM, Raghavendra Gowdappa <rgowdapp@xxxxxxxxxx> wrote:On Tue, Aug 14, 2018 at 7:23 PM, mabi <mabi@xxxxxxxxxxxxx> wrote:Bad news: the process blocked happened again this time with another directory of another user which is NOT over his quota but which also has quota enabled.The symptoms on the Linux side are the same:[Tue Aug 14 15:30:33 2018] INFO: task php5-fpm:14773 blocked for more than 120 seconds.[Tue Aug 14 15:30:33 2018] Not tainted 3.16.0-4-amd64 #1[Tue Aug 14 15:30:33 2018] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [Tue Aug 14 15:30:33 2018] php5-fpm D ffff8801fea13200 0 14773 729 0x00000000[Tue Aug 14 15:30:33 2018] ffff880100bbe0d0 0000000000000282 0000000000013200 ffff880129bcffd8[Tue Aug 14 15:30:33 2018] 0000000000013200 ffff880100bbe0d0 ffff880153ed0d68 ffff880129bcfee0[Tue Aug 14 15:30:33 2018] ffff880153ed0d6c ffff880100bbe0d0 00000000ffffffff ffff880153ed0d70[Tue Aug 14 15:30:33 2018] Call Trace:[Tue Aug 14 15:30:33 2018] [<ffffffff81522935>] ? schedule_preempt_disabled+0x25/0x70 [Tue Aug 14 15:30:33 2018] [<ffffffff815243d3>] ? __mutex_lock_slowpath+0xd3/0x1d0 [Tue Aug 14 15:30:33 2018] [<ffffffff811d7a93>] ? write_inode_now+0x93/0xc0[Tue Aug 14 15:30:33 2018] [<ffffffff815244eb>] ? mutex_lock+0x1b/0x2a[Tue Aug 14 15:30:33 2018] [<ffffffffa00df69f>] ? fuse_flush+0x8f/0x1e0 [fuse][Tue Aug 14 15:30:33 2018] [<ffffffff811af393>] ? vfs_read+0x93/0x170[Tue Aug 14 15:30:33 2018] [<ffffffff811ad14a>] ? filp_close+0x2a/0x70[Tue Aug 14 15:30:33 2018] [<ffffffff811ad1af>] ? SyS_close+0x1f/0x50[Tue Aug 14 15:30:33 2018] [<ffffffff81525d0d>] ? system_call_fast_compare_end+0x10/0x15 and if I check this process it has state "D" which is "D = uninterruptible sleep".Now I also managed to take a statedump file as recommended but I see in its content under the "[io-cache.inode]" a "path=" which I would need to remove as it contains filenames for privacy reasons. Can I remove every "path=" line and still send you the statedump file for analysis?Yes. Removing path is fine and statedumps will still be useful for debugging the issue.Thank you.‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐On August 14, 2018 10:48 AM, Nithya Balachandran <nbalacha@xxxxxxxxxx> wrote:Thanks for letting us know. Sanoj, can you take a look at this?Thanks.NithyaOn 14 August 2018 at 13:58, mabi <mabi@xxxxxxxxxxxxx> wrote:As you mentioned after creating the /var/run/gluster directory I got a statedump file in there.As a workaround I have now removed the quota for this specific directory and as it is a production server I can currently not "play" with it by adding the quota back and having the same problem as it requires me to reboot the server with downtime...But I can confirm that by removing the quota from that directory, the problem is gone (no more blocking processes such as "ls") so there must be an issue or bug with the quota part of gluster.‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐On August 10, 2018 4:19 PM, Nithya Balachandran <nbalacha@xxxxxxxxxx> wrote:On 9 August 2018 at 19:54, mabi <mabi@xxxxxxxxxxxxx> wrote:Thanks for the documentation. On my client using FUSE mount I found the PID by using ps (output below):root 456 1 4 14:17 ? 00:05:15 /usr/sbin/glusterfs --volfile-server=gfs1a --volfile-id=myvol-private /mnt/myvol-privateThen I ran the following commandsudo kill -USR1 456but now I can't find where the files are stored. Are these supposed to be stored on the client directly? I checked /var/run/gluster and /var/log/gluster but could not see anything and /var/log/gluster does not even exist on the client.They are usually created in /var/run/gluster. You will need to create the directory on the client if it does not exist.‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐On August 9, 2018 3:59 PM, Raghavendra Gowdappa <rgowdapp@xxxxxxxxxx> wrote:On Thu, Aug 9, 2018 at 6:47 PM, mabi <mabi@xxxxxxxxxxxxx> wrote:Hi Nithya,Thanks for the fast answer. Here the additional info:1. gluster volume infoVolume Name: myvol-privateType: ReplicateVolume ID: e7a40a1b-45c9-4d3c-bb19-0c59b4eceec5 Status: StartedSnapshot Count: 0Number of Bricks: 1 x (2 + 1) = 3Transport-type: tcpBricks:Brick1: gfs1a:/data/myvol-private/brick Brick2: gfs1b:/data/myvol-private/brick Brick3: gfs1c:/srv/glusterfs/myvol-private/brick (arbiter) Options Reconfigured:features.default-soft-limit: 95%transport.address-family: inetfeatures.quota-deem-statfs: onfeatures.inode-quota: onfeatures.quota: onnfs.disable: onperformance.readdir-ahead: onclient.event-threads: 4server.event-threads: 4auth.allow: 192.168.100.922. Sorry I have no clue how to take a "statedump" of a process on Linux. Which command should I use for that? and which process would you like, the blocked process (for example "ls")?Statedumps are gluster specific. Please refer to https://docs.gluster.org/en/v3/Troubleshooting/statedump/ for instructions.Regards,M.‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐On August 9, 2018 3:10 PM, Nithya Balachandran <nbalacha@xxxxxxxxxx> wrote:Hi,Please provide the following:
- gluster volume info
- statedump of the fuse process when it hangs
Thanks,NithyaOn 9 August 2018 at 18:24, mabi <mabi@xxxxxxxxxxxxx> wrote:Hello,I recently upgraded my GlusterFS replica 2+1 (aribter) to version 3.12.12 and now I see a weird behaviour on my client (using FUSE mount) where I have processes (PHP 5.6 FPM) trying to access a specific directory and then the process blocks. I can't kill the process either, not even with kill -9. I need to reboot the machine in order to get rid of these blocked processes.This directory has one particularity compared to the other directories it is that it has reached it's quota soft-limit as you can see here in the output of gluster volume quota list:Path Hard-limit Soft-limit Used Available Soft-limit exceeded? Hard-limit exceeded?------------------------------------------------------------ ------------------------------ ------------------------------ ------- /directory 100.0GB 80%(80.0GB) 90.5GB 9.5GB Yes NoThat does not mean that it is the quota's fault but it might be a hint where to start looking for... And by the way can someone explain me what the soft-limit does? or does it not do anything special?Here is an the linux stack of a blocking process on that directory which happened with a simple "ls -la":[Thu Aug 9 14:21:07 2018] INFO: task ls:2272 blocked for more than 120 seconds.[Thu Aug 9 14:21:07 2018] Not tainted 3.16.0-4-amd64 #1[Thu Aug 9 14:21:07 2018] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [Thu Aug 9 14:21:07 2018] ls D ffff88017ef93200 0 2272 2268 0x00000004[Thu Aug 9 14:21:07 2018] ffff88017653f490 0000000000000286 0000000000013200 ffff880174d7bfd8[Thu Aug 9 14:21:07 2018] 0000000000013200 ffff88017653f490 ffff8800eeb3d5f0 ffff8800fefac800[Thu Aug 9 14:21:07 2018] ffff880174d7bbe0 ffff8800eeb3d6d0 ffff8800fefac800 ffff8800ffe1e1c0[Thu Aug 9 14:21:07 2018] Call Trace:[Thu Aug 9 14:21:07 2018] [<ffffffffa00dc50d>] ? __fuse_request_send+0xbd/0x270 [fuse][Thu Aug 9 14:21:07 2018] [<ffffffff810abce0>] ? prepare_to_wait_event+0xf0/0xf0 [Thu Aug 9 14:21:07 2018] [<ffffffffa00e0791>] ? fuse_dentry_revalidate+0x181/0x300 [fuse] [Thu Aug 9 14:21:07 2018] [<ffffffff811b944e>] ? lookup_fast+0x25e/0x2b0[Thu Aug 9 14:21:07 2018] [<ffffffff811bacc5>] ? path_lookupat+0x155/0x780[Thu Aug 9 14:21:07 2018] [<ffffffff81195715>] ? kmem_cache_alloc+0x75/0x480[Thu Aug 9 14:21:07 2018] [<ffffffffa00dfca9>] ? fuse_getxattr+0xe9/0x150 [fuse][Thu Aug 9 14:21:07 2018] [<ffffffff811bb316>] ? filename_lookup+0x26/0xc0[Thu Aug 9 14:21:07 2018] [<ffffffff811bf594>] ? user_path_at_empty+0x54/0x90[Thu Aug 9 14:21:07 2018] [<ffffffff81193e08>] ? kmem_cache_free+0xd8/0x210[Thu Aug 9 14:21:07 2018] [<ffffffff811bf59f>] ? user_path_at_empty+0x5f/0x90[Thu Aug 9 14:21:07 2018] [<ffffffff811b3d46>] ? vfs_fstatat+0x46/0x90[Thu Aug 9 14:21:07 2018] [<ffffffff811b421d>] ? SYSC_newlstat+0x1d/0x40[Thu Aug 9 14:21:07 2018] [<ffffffff811d34b8>] ? SyS_lgetxattr+0x58/0x80[Thu Aug 9 14:21:07 2018] [<ffffffff81525d0d>] ? system_call_fast_compare_end+0x10/0x15 My 3 gluster nodes are all Debian 9 and my client Debian 8.Let me know if you need more information.Best regards,Mabi_______________________________________________ Gluster-users mailing list_______________________________________________ Gluster-users mailing list_______________________________________________ Gluster-users mailing list_______________________________________________ Gluster-users mailing list
_______________________________________________ Gluster-users mailing list Gluster-users@xxxxxxxxxxx https://lists.gluster.org/mailman/listinfo/gluster-users
