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 Thanks, Ilya _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |