On Wed, Apr 12, 2017 at 10:51 AM, Ilya Dryomov <idryomov@xxxxxxxxx> wrote: > On Wed, Apr 12, 2017 at 4:28 PM, Alex Gorbachev <ag@xxxxxxxxxxxxxxxxxxx> wrote: >> Hi Ilya, >> >> On Wed, Apr 12, 2017 at 4:58 AM Ilya Dryomov <idryomov@xxxxxxxxx> wrote: >>> >>> On Tue, Apr 11, 2017 at 3:10 PM, Alex Gorbachev <ag@xxxxxxxxxxxxxxxxxxx> >>> wrote: >>> > Hi Ilya, >>> > >>> > On Tue, Apr 11, 2017 at 4:06 AM, Ilya Dryomov <idryomov@xxxxxxxxx> >>> > wrote: >>> >> On Tue, Apr 11, 2017 at 4:01 AM, Alex Gorbachev >>> >> <ag@xxxxxxxxxxxxxxxxxxx> wrote: >>> >>> On Mon, Apr 10, 2017 at 2:16 PM, Alex Gorbachev >>> >>> <ag@xxxxxxxxxxxxxxxxxxx> wrote: >>> >>>> I am trying to understand the cause of a problem we started >>> >>>> encountering a few weeks ago. There are 30 or so per hour messages >>> >>>> on >>> >>>> OSD nodes of type: >>> >>>> >>> >>>> ceph-osd.33.log:2017-04-10 13:42:39.935422 7fd7076d8700 0 bad crc in >>> >>>> data 2227614508 != exp 2469058201 >>> >>>> >>> >>>> and >>> >>>> >>> >>>> 2017-04-10 13:42:39.939284 7fd722c42700 0 -- 10.80.3.25:6826/5752 >>> >>>> submit_message osd_op_reply(1826606251 >>> >>>> rbd_data.922d95238e1f29.00000000000101bf [set-alloc-hint object_size >>> >>>> 16777216 write_size 16777216,write 6328320~12288] v103574'18626765 >>> >>>> uv18626765 ondisk = 0) v6 remote, 10.80.3.216:0/1934733503, failed >>> >>>> lossy con, dropping message 0x3b55600 >>> >>>> >>> >>>> On a client sometimes, but not corresponding to the above: >>> >>>> >>> >>>> Apr 10 11:53:15 roc-5r-scd216 kernel: [4906599.023174] libceph: osd96 >>> >>>> 10.80.3.25:6822 socket error on write >>> >>>> >>> >>>> And from time to time, slow requests: >>> >>>> >>> >>>> 2017-04-10 13:00:04.280686 osd.91 10.80.3.45:6808/5665 231 : cluster >>> >>>> [WRN] slow request 30.108325 seconds old, received at 2017-04-10 >>> >>>> 12:59:34.172283: osd_op(client.11893449.1:324079247 >>> >>>> rbd_data.8fcdfb238e1f29.00000000000187e7 [set-alloc-hint object_size >>> >>>> 16777216 write_size 16777216,write 10772480~8192] 14.ed0bcdec >>> >>>> ondisk+write e103545) currently waiting for subops from 2,104 >>> >>>> 2017-04-10 13:00:06.280949 osd.91 10.80.3.45:6808/5665 232 : cluster >>> >>>> [WRN] 2 slow requests, 1 included below; oldest blocked for > >>> >>>> 32.108610 secs >>> >>>> >>> >>>> Questions: >>> >>>> >>> >>>> 1. Is there any way to drill further into the "bad crc" message? >>> >>>> sometimes they have nothing before or after them, but how to >>> >>>> determine >>> >>>> from/to what this came from - another OSD, client, which one? >>> >>>> >>> >>>> 2. Network seems OK - no errors on NICs, regression testing does not >>> >>>> show any issues. I realize this can be disk response, but using >>> >>>> Christian Balzer's atop recommendation shows a pretty normal system. >>> >>>> What is my best course of troubleshooting here - dump historic ops on >>> >>>> OSD, wireshark the links or anything else? >>> >>>> >>> >>>> 3. Christian, if you are looking at this, what would be your red >>> >>>> flags in atop? >>> >>>> >>> >>> >>> >>> One more note: OSD nodes are running kernel 4.10.2-041002-generic and >>> >>> clients - 4.4.23-040423-generic >>> >> >>> >> Hi Alex, >>> >> >>> >> Did you upgrade the kernel client from < 4.4 to 4.4 somewhere in that >>> >> time >>> >> frame by any chance? >>> > >>> > Yes, they were upgraded from 4.2.8 to 4.4.23 in October. >>> >>> There is a block layer bug in 4.4 and later kernels [1]. It >>> effectively undoes krbd commit [2], which prevents pages from being >>> further updated while in-flight. The timeline doesn't fit though, so >>> it's probably unrelated -- not every workload can trigger it... >>> >>> [1] http://tracker.ceph.com/issues/19275 >>> [2] >>> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=bae818ee1577c27356093901a0ea48f672eda514 >> >> >> Would the workaround be to drop to 4.3? > > Unfortunately, short of installing a custom kernel or disabling data > CRCs entirely, yes. I'll poke the respective maintainer again later > today... I downgraded to 4.3.6 and so far all the CRC messages and lossy con went away completely. > > Thanks, > > Ilya _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |