On 20/10, Ignazio Cassano wrote:
This is the entre log from when the migration started:
http://paste.openstack.org/show/799199/
Ignazio
Hi,
There are no os-brick calls in there. :-(
You should look for the call to connect_volume that should have something like:
==> disconnect_volume: call "{'args': (<os_brick.initiator.connectors.iscsi
And the second parameter to that call is a dictionary where you can see the target_lun, target_luns, target_portals, target_portal, target_iqn, target_iqns... This will allow us to check if we are actually connected to those targets-portals
The third parameter should contain two things that are relevant, the scsi_wwn and the path. You can check if the path exists and if that path actually has that wwn using /lib/udev/scsi_id --page 0x83 --whitelisted $path
Those are the things I would check, because the only reason I can think that os-brick is not disconnecting any volumes are that the connection info is not right, or that the volume is no longer connected.
Cheers, Gorka.
Il giorno mar 20 ott 2020 alle ore 11:23 Gorka Eguileor geguileo@redhat.com ha scritto:
On 20/10, Ignazio Cassano wrote:
Hello Gorka,this is what happens on nova compute with debug enabled,
when I
migrate an instance with iscsi volumes ( note Disconnecting from[] should be the issue):
Hi,
The disconnect from [] is the right clue, not necessarily the issue.
OS-Brick is saying that for the connection information that has been passed in the "disconnect_volume" call (which is not present in the emailed logs) there are no volumes present in the system.
You should check the connection info that Nova is passing to disconnect_volume and confirm if that data is correct. For example checking if the path present in the connection info dictionary is the same as the one in the instance's XML dump, or if the LUN from the connection info dict is actually present in the system.
There are multiple reasons why Nova could be passing the wrong connection info to os-brick. The ones that come to mind are:
- There was a failed migration at some point, and Nova didn't rollback the connection info on the BDM table.
- Nova is calling multiple times initialize_connection on Cinder for the same host and the driver being used is not idempotent.
Cheers, Gorka.
stderr= _run_iscsiadm_bare
/usr/lib/python2.7/site-packages/os_brick/initiator/connectors/iscsi.py:1122
2020-10-20 09:52:33.066 132171 DEBUG os_brick.initiator.connectors.iscsi [-] iscsi session list stdout=tcp: [10] 10.138.209.48:3260,9 iqn.1992-04.com.emc:cx.ckm00184400687.a3 (non-flash) tcp: [11] 10.138.215.17:3260,8 iqn.1992-04.com.emc:cx.ckm00184400687.a2 (non-flash) tcp: [12] 10.138.215.17:3260,8 iqn.1992-04.com.emc:cx.ckm00184400687.a2 (non-flash) tcp: [13] 10.138.215.18:3260,7 iqn.1992-04.com.emc:cx.ckm00184400687.b2 (non-flash) tcp: [14] 10.138.215.18:3260,7 iqn.1992-04.com.emc:cx.ckm00184400687.b2 (non-flash) tcp: [15] 10.138.209.47:3260,6 iqn.1992-04.com.emc:cx.ckm00184400687.b3 (non-flash) tcp: [16] 10.138.209.47:3260,6 iqn.1992-04.com.emc:cx.ckm00184400687.b3 (non-flash) tcp: [9] 10.138.209.48:3260,9 iqn.1992-04.com.emc:cx.ckm00184400687.a3 (non-flash) stderr= _run_iscsi_session
/usr/lib/python2.7/site-packages/os_brick/initiator/connectors/iscsi.py:1111
2020-10-20 09:52:33.078 132171 DEBUG os_brick.initiator.connectors.iscsi [-] Resulting device map defaultdict(<function <lambda> at
0x7f4f1b1f7cf8>,
{(u'10.138.215.17:3260', u'iqn.1992-04.com.emc:cx.ckm00184400687.a2'): (set([]), set([u'sdg', u'sdi'])), (u'10.138.209.47:3260', u'iqn.1992-04.com.emc:cx.ckm00184400687.b3'): (set([]), set([u'sdo', u'sdq'])), (u'10.138.209.48:3260', u'iqn.1992-04.com.emc:cx.ckm00184400687.a3'): (set([]), set([u'sdd', u'sdb'])), (u'10.138.215.18:3260', u'iqn.1992-04.com.emc:cx.ckm00184400687.b2'): (set([]), set([u'sdm', u'sdk']))}) _get_connection_devices
/usr/lib/python2.7/site-packages/os_brick/initiator/connectors/iscsi.py:844
2020-10-20 09:52:33.078 132171 DEBUG os_brick.initiator.connectors.iscsi [-] Disconnecting from: [] _disconnect_connection
/usr/lib/python2.7/site-packages/os_brick/initiator/connectors/iscsi.py:1099
2020-10-20 09:52:33.079 132171 DEBUG oslo_concurrency.lockutils [-] Lock "connect_volume" released by "os_brick.initiator.connectors.iscsi.disconnect_volume" :: held 1.058s inner /usr/lib/python2.7/site-packages/oslo_concurrency/lockutils.py:339 2020-10-20 09:52:33.079 132171 DEBUG os_brick.initiator.connectors.iscsi [-] <== disconnect_volume: return (1057ms) None trace_logging_wrapper /usr/lib/python2.7/site-packages/os_brick/utils.py:170 2020-10-20 09:52:33.079 132171 DEBUG nova.virt.libvirt.volume.iscsi [-] [instance: 0c846f66-f194-40de-b31e-d53652570fa7] Disconnected iSCSI
Volume
disconnect_volume /usr/lib/python2.7/site-packages/nova/virt/libvirt/volume/iscsi.py:78 2020-10-20 09:52:33.080 132171 DEBUG os_brick.utils [-] ==> get_connector_properties: call u"{'execute': None, 'my_ip': '10.138.208.178', 'enforce_multipath': True, 'host': 'podiscsivc-kvm02', 'root_helper': 'sudo nova-rootwrap /etc/nova/rootwrap.conf', 'multipath': True}" trace_logging_wrapper /usr/lib/python2.7/site-packages/os_brick/utils.py:146 2020-10-20 09:52:33.125 132171 DEBUG os_brick.initiator.linuxfc [-] No Fibre Channel support detected on system. get_fc_hbas /usr/lib/python2.7/site-packages/os_brick/initiator/linuxfc.py:157 2020-10-20 09:52:33.126 132171 DEBUG os_brick.initiator.linuxfc [-] No Fibre Channel support detected on system. get_fc_hbas /usr/lib/python2.7/site-packages/os_brick/initiator/linuxfc.py:157 2020-10-20 09:52:33.145 132171 DEBUG os_brick.utils [-] <== get_connector_properties: return (61ms) {'initiator': u'iqn.1994-05.com.redhat:fbfdc37eed4c', 'ip': u'10.138.208.178', 'system uuid': u'4C4C4544-0051-4E10-8057-B6C04F425932', 'platform': u'x86_64', 'host': u'podiscsivc-kvm02', 'do_local_attach': False, 'os_type': u'linux2', 'multipath': True} trace_logging_wrapper /usr/lib/python2.7/site-packages/os_brick/utils.py:170
Best regards Ignazio
Il giorno gio 15 ott 2020 alle ore 10:57 Gorka Eguileor <
geguileo@redhat.com>
ha scritto:
On 14/10, Ignazio Cassano wrote:
Hello, thank you for the answer. I am using os-brick 2.3.8 but I got same issues on stein with
os.brick
2.8
For explain better the situation I send you the output of multipath
-ll
on
a compute node: root@podvc-kvm01 ansible]# multipath -ll Oct 14 18:50:01 | sdbg: alua not supported Oct 14 18:50:01 | sdbe: alua not supported Oct 14 18:50:01 | sdbd: alua not supported Oct 14 18:50:01 | sdbf: alua not supported 360060160f0d049007ab7275f743d0286 dm-11 DGC ,VRAID size=30G features='1 retain_attached_hw_handler' hwhandler='1 alua'
wp=rw
|-+- policy='round-robin 0' prio=0 status=enabled | |- 15:0:0:71 sdbg 67:160 failed faulty running | `- 12:0:0:71 sdbe 67:128 failed faulty running `-+- policy='round-robin 0' prio=0 status=enabled |- 11:0:0:71 sdbd 67:112 failed faulty running `- 13:0:0:71 sdbf 67:144 failed faulty running 360060160f0d049004cdb615f52343fdb dm-8 DGC ,VRAID size=80G features='2 queue_if_no_path retain_attached_hw_handler' hwhandler='1 alua' wp=rw |-+- policy='round-robin 0' prio=50 status=active | |- 15:0:0:210 sdau 66:224 active ready running | `- 12:0:0:210 sdas 66:192 active ready running `-+- policy='round-robin 0' prio=10 status=enabled |- 11:0:0:210 sdar 66:176 active ready running `- 13:0:0:210 sdat 66:208 active ready running 360060160f0d0490034aa645fe52265eb dm-12 DGC ,VRAID size=100G features='2 queue_if_no_path retain_attached_hw_handler' hwhandler='1 alua' wp=rw |-+- policy='round-robin 0' prio=50 status=active | |- 12:0:0:177 sdbi 67:192 active ready running | `- 15:0:0:177 sdbk 67:224 active ready running `-+- policy='round-robin 0' prio=10 status=enabled |- 11:0:0:177 sdbh 67:176 active ready running `- 13:0:0:177 sdbj 67:208 active ready running 360060160f0d04900159f225fd6126db9 dm-6 DGC ,VRAID size=40G features='2 queue_if_no_path retain_attached_hw_handler' hwhandler='1 alua' wp=rw |-+- policy='round-robin 0' prio=50 status=active | |- 11:0:0:26 sdaf 65:240 active ready running | `- 13:0:0:26 sdah 66:16 active ready running `-+- policy='round-robin 0' prio=10 status=enabled |- 12:0:0:26 sdag 66:0 active ready running `- 15:0:0:26 sdai 66:32 active ready running Oct 14 18:50:01 | sdba: alua not supported Oct 14 18:50:01 | sdbc: alua not supported Oct 14 18:50:01 | sdaz: alua not supported Oct 14 18:50:01 | sdbb: alua not supported 360060160f0d049007eb7275f93937511 dm-10 DGC ,VRAID size=40G features='1 retain_attached_hw_handler' hwhandler='1 alua'
wp=rw
|-+- policy='round-robin 0' prio=0 status=enabled | |- 12:0:0:242 sdba 67:64 failed faulty running | `- 15:0:0:242 sdbc 67:96 failed faulty running `-+- policy='round-robin 0' prio=0 status=enabled |- 11:0:0:242 sdaz 67:48 failed faulty running `- 13:0:0:242 sdbb 67:80 failed faulty running 360060160f0d049003a567c5fb72201e8 dm-7 DGC ,VRAID size=40G features='2 queue_if_no_path retain_attached_hw_handler' hwhandler='1 alua' wp=rw |-+- policy='round-robin 0' prio=50 status=active | |- 12:0:0:57 sdbq 68:64 active ready running | `- 15:0:0:57 sdbs 68:96 active ready running `-+- policy='round-robin 0' prio=10 status=enabled |- 11:0:0:57 sdbp 68:48 active ready running `- 13:0:0:57 sdbr 68:80 active ready running 360060160f0d04900c120625f802ea1fa dm-9 DGC ,VRAID size=25G features='2 queue_if_no_path retain_attached_hw_handler' hwhandler='1 alua' wp=rw |-+- policy='round-robin 0' prio=50 status=active | |- 11:0:0:234 sdav 66:240 active ready running | `- 13:0:0:234 sdax 67:16 active ready running `-+- policy='round-robin 0' prio=10 status=enabled |- 15:0:0:234 sday 67:32 active ready running `- 12:0:0:234 sdaw 67:0 active ready running 360060160f0d04900b8b0615fb14ef1bd dm-3 DGC ,VRAID size=50G features='2 queue_if_no_path retain_attached_hw_handler' hwhandler='1 alua' wp=rw |-+- policy='round-robin 0' prio=50 status=active | |- 11:0:0:11 sdan 66:112 active ready running | `- 13:0:0:11 sdap 66:144 active ready running `-+- policy='round-robin 0' prio=10 status=enabled |- 12:0:0:11 sdao 66:128 active ready running `- 15:0:0:11 sdaq 66:160 active ready running
The active running are related to running virtual machines. The faulty are related to virtual macnines migrated on other kvm
nodes.
Every volume has 4 path because iscsi on unity needs two different
vlans,
each one with 2 addresses. I think this issue can be related to os-brick because when I migrate
a
virtual machine from host A host B in the cova compute log on host A
I
read:
2020-10-13 10:31:02.769 118727 DEBUG
os_brick.initiator.connectors.iscsi
[req-771ede8c-6e1b-4f3f-ad4a-1f6ed820a55c
66adb965bef64eaaab2af93ade87e2ca
85cace94dcc7484c85ff9337eb1d0c4c - default default] *Disconnecting
from:
[]*
Ignazio
Hi,
That's definitely the right clue!! Though I don't fully agree with
this
being an os-brick issue just yet. ;-)
Like I mentioned before, RCA is usually non-trivial, and explaining how to debug these issues over email is close to impossible, but if this were my system, and assuming you have tested normal attach/detach procedure and is working fine, this is what I would do:
- Enable DEBUG logs on Nova compute node (I believe you already have)
- Attach a new device to an instance on that node with --debug to get the request id
- Get the connection information dictionary that os-brick receives on the call to connect_volume for that request, and the data that os-brick returns to Nova on that method call completion.
- Check if the returned data to Nova is a multipathed device or not (in 'path'), and whether we have the wwn or not (in 'scsi_wwn'). It should be a multipath device, and then I would check its status in
the
multipath daemon.
- Now do the live migration (with --debug to get the request id) and
see
what information Nova passes in that request to os-brick's disconnect_volume.
- Is it the same? Then it's likely an os-brick issue, and I can have
a
look at the logs if you put the logs for that os-brick detach process in a pastebin [1].
- Is it different? Then it's either a Nova bug or a Cinder driver specific bug.
- Is there a call from Nova to Cinder, in the migration request,
for
that same volume to initialize_connection passing the source host connector info (info from the host that is currently attached)? If there is a call, check if the returned data is different from the one we used to do the attach, if that's the case then it's a Nova and Cinder driver bug that was solved on the Nova side in 17.0.10 [2]. - If there's no call to Cinder's initialize_connection, the it's most likely a Nova bug. Try to find out if this connection info makes any sense for that host (LUN, target, etc.) or if this is the one from the destination volume.I hope this somehow helps.
Cheers, Gorka.
Il giorno mer 14 ott 2020 alle ore 13:41 Gorka Eguileor <
geguileo@redhat.com>
ha scritto:
On 09/10, Ignazio Cassano wrote: > Hello Stackers, I am using dell emc iscsi driver on my centos 7
queens
> openstack. It works and instances work as well but on compute
nodes I
got a > lot a faulty device reported by multipath il comand. > I do know why this happens, probably attacching and detaching
volumes and
> live migrating instances do not close something well. > I read this can cause serious performances problems on compute
nodes.
> Please, any workaround and/or patch is suggested ? > Regards > Ignazio
Hi,
There are many, many, many things that could be happening there,
and
it's not usually trivial doing the RCA, so the following questions
are
just me hoping this is something "easy" to find out.
What os-brick version from Queens are you running? Latest
(2.3.9), or
maybe one older than 2.3.3?
When you say you have faulty devices reported, are these faulty
devices
alone in the multipath DM? Or do you have some faulty ones with
some
that are ok?
If there are some OK and some that aren't, are they consecutive
devices?
(as in /dev/sda /dev/sdb etc).
Cheers, Gorka.