On Wed, Aug 30, 2017 at 6:47 PM, Dominik Holler <dholler@redhat.com> wrote:
> On Wed, 30 Aug 2017 13:31:38 +0200
> Dominik Holler <dholler@redhat.com> wrote:
>
>> On Wed, 30 Aug 2017 14:18:49 +0300
>> Dan Kenigsberg <danken@redhat.com> wrote:
>>
>> > On Wed, Aug 30, 2017 at 1:40 PM, Barak Korren <bkorren@redhat.com>
>> > wrote:
>> > > Test failed: [ 002_bootstrap.add_hosts ]
>> > >
>> > > Link to suspected patches:
>> > >
>> > > We suspect this is due to change to LLDPAD in upstream CentOS
>> > > repos. We can't tell the exact point it was introduced because
>> > > other ovirt-engine regressions introduced too much noise into the
>> > > system.
>> > >
>> > > It also seems that the failure is not 100% reproducible sine we
>> > > have runs that do not encounter it.
>> > >
>> > > Link to Job:
>> > > http://jenkins.ovirt.org/job/ovirt-master_change-queue-tester/2151
>> > >
>> > > (This job checked a specific patch to ovirt-log-collector but
>> > > failure seems unrelated, and also happens on vanille 'tested' repo
>> > > right now).
>> > >
>> > > Link to all logs:
>> > > VDSM logs seem to be most relevant here:
>> > >
>> > > http://jenkins.ovirt.org/job/ovirt-master_change-queue-tester/2151/artifact/exported-artifacts/basic-suit-master-el7/test_logs/basic-suite-master/post-002_bootstrap.py/lago-basic-suite-master-host-0/_var_log/vdsm/
>> > >
>> > > Error snippet from log:
>> > >
>> > > - From VDSM logs (note it exists very quickly):
>> > >
>> > > <error>
>> > >
>> > > 2017-08-30 05:55:52,208-0400 INFO  (MainThread) [vds] (PID: 4594)
>> > > I am the actual vdsm 4.20.2-133.git0ce4485.el7.centos
>> > > lago-basic-suite-master-host-0 (3.10.0-514.2.2.el7.x86_64)
>> > > (vdsmd:148) 2017-08-30 05:55:52,209-0400 INFO  (MainThread) [vds]
>> > > VDSM will run with cpu affinity: frozenset([1]) (vdsmd:254)
>> > > 2017-08-30 05:55:52,253-0400 INFO  (MainThread) [storage.check]
>> > > Starting check service (check:92)
>> > > 2017-08-30 05:55:52,257-0400 INFO  (MainThread)
>> > > [storage.Dispatcher] Starting StorageDispatcher... (dispatcher:48)
>> > > 2017-08-30 05:55:52,257-0400 INFO  (check/loop)
>> > > [storage.asyncevent] Starting <EventLoop running=True
>> > > closed=False at 0x45317520> (asyncevent:125)
>> > > 2017-08-30 05:55:52,288-0400 INFO  (MainThread) [vdsm.api] START
>> > > registerDomainStateChangeCallback(callbackFunc=<functools.partial
>> > > object at 0x2b47b50>) from=internal,
>> > > task_id=2cebe9ef-358e-47d7-81a6-8c4b54b9cd6d (api:46)
>> > > 2017-08-30 05:55:52,288-0400 INFO  (MainThread) [vdsm.api] FINISH
>> > > registerDomainStateChangeCallback return=None from=internal,
>> > > task_id=2cebe9ef-358e-47d7-81a6-8c4b54b9cd6d (api:52)
>> > > 2017-08-30 05:55:52,288-0400 INFO  (MainThread) [MOM] Preparing
>> > > MOM interface (momIF:53)
>> > > 2017-08-30 05:55:52,289-0400 INFO  (MainThread) [MOM] Using named
>> > > unix socket /var/run/vdsm/mom-vdsm.sock (momIF:62)
>> > > 2017-08-30 05:55:52,289-0400 INFO  (MainThread) [root]
>> > > Unregistering all secrets (secret:91)
>> > > 2017-08-30 05:55:52,307-0400 INFO  (MainThread) [vds] Setting
>> > > channels' timeout to 30 seconds. (vmchannels:221)
>> > > 2017-08-30 05:55:52,309-0400 INFO  (vmrecovery) [vds] recovery:
>> > > completed in 0s (clientIF:516)
>> > > 2017-08-30 05:55:52,310-0400 INFO  (MainThread)
>> > > [vds.MultiProtocolAcceptor] Listening at :::54321
>> > > (protocoldetector:196)
>> > > 2017-08-30 05:55:52,496-0400 INFO  (http) [vds] Server running
>> > > (http:58) 2017-08-30 05:55:52,742-0400 INFO  (periodic/1)
>> > > [vdsm.api] START repoStats(domains=()) from=internal,
>> > > task_id=85768015-8ecb-48e3-9307-f671bfc33c65 (api:46)
>> > > 2017-08-30 05:55:52,743-0400 INFO  (periodic/1) [vdsm.api] FINISH
>> > > repoStats return={} from=internal,
>> > > task_id=85768015-8ecb-48e3-9307-f671bfc33c65 (api:52)
>> > > 2017-08-30 05:55:52,743-0400 WARN  (periodic/1) [throttled] MOM
>> > > not available. (throttledlog:103)
>> > > 2017-08-30 05:55:52,744-0400 WARN  (periodic/1) [throttled] MOM
>> > > not available, KSM stats will be missing. (throttledlog:103)
>> > > 2017-08-30 05:55:55,043-0400 INFO  (MainThread) [vds] Received
>> > > signal 15, shutting down (vdsmd:67)
>> > > 2017-08-30 05:55:55,045-0400 INFO  (MainThread)
>> > > [jsonrpc.JsonRpcServer] Stopping JsonRPC Server (__init__:759)
>> > > 2017-08-30 05:55:55,049-0400 INFO  (MainThread) [vds] Stopping
>> > > http server (http:79)
>> > > 2017-08-30 05:55:55,049-0400 INFO  (http) [vds] Server stopped
>> > > (http:69) 2017-08-30 05:55:55,050-0400 INFO  (MainThread) [root]
>> > > Unregistering all secrets (secret:91)
>> > > 2017-08-30 05:55:55,052-0400 INFO  (MainThread) [vdsm.api] START
>> > > prepareForShutdown(options=None) from=internal,
>> > > task_id=ffde5caa-fa44-49ab-bdd1-df81519680a3 (api:46)
>> > > 2017-08-30 05:55:55,089-0400 INFO  (MainThread) [storage.Monitor]
>> > > Shutting down domain monitors (monitor:222)
>> > > 2017-08-30 05:55:55,090-0400 INFO  (MainThread) [storage.check]
>> > > Stopping check service (check:105)
>> > > 2017-08-30 05:55:55,090-0400 INFO  (check/loop)
>> > > [storage.asyncevent] Stopping <EventLoop running=True
>> > > closed=False at 0x45317520> (asyncevent:220)
>> > > 2017-08-30 05:55:55,090-0400 INFO  (MainThread) [vdsm.api] FINISH
>> > > prepareForShutdown return=None from=internal,
>> > > task_id=ffde5caa-fa44-49ab-bdd1-df81519680a3 (api:52)
>> > > 2017-08-30 05:55:55,091-0400 INFO  (MainThread) [vds] Stopping
>> > > threads (vdsmd:159)
>> > > 2017-08-30 05:55:55,091-0400 INFO  (MainThread) [vds] Exiting
>> > > (vdsmd:170)
>> > >
>> > > </error>
>> > >
>> > >
>> > > - From SuperVDSM logs:
>> > >
>> > > <error>
>> > >
>> > > MainThread::ERROR::2017-08-30
>> > > 05:55:59,476::initializer::53::root::(_lldp_init) Failed to enable
>> > > LLDP on eth1
>> > > Traceback (most recent call last):
>> > >   File
>> > > "/usr/lib/python2.7/site-packages/vdsm/network/initializer.py",
>> > > line 51, in _lldp_init Lldp.enable_lldp_on_iface(device)
>> > >   File
>> > > "/usr/lib/python2.7/site-packages/vdsm/network/lldp/lldpad.py",
>> > > line 30, in enable_lldp_on_iface
>> > > lldptool.enable_lldp_on_iface(iface, rx_only) File
>> > > "/usr/lib/python2.7/site-packages/vdsm/network/lldpad/lldptool.py",
>> > > line 46, in enable_lldp_on_iface raise EnableLldpError(rc, out,
>> > > err, iface) EnableLldpError: (1,
>> > > "timeout\n'M00000001C3040000000c04eth1000badminStatus0002rx'
>> > > command timed out.\n", '', 'eth1')
>> > > MainThread::DEBUG::2017-08-30
>> > > 05:55:59,477::cmdutils::133::root::(exec_cmd) /usr/sbin/lldptool
>> > > get-lldp -i eth0 adminStatus (cwd None)
>> > >
>> > >
>> > > </error>
>> >
>> > The LLDP errors are due to SELinux, to be fixed in CentOS-7.4.1.
>> > They do not cause supervdsmd to stop, so they are not the reason
>> > for vdsm's failure to start. Something ELSE is doing this.
>> >
>> > in
>> > http://jenkins.ovirt.org/job/ovirt-master_change-queue-tester/2151/artifact/exported-artifacts/basic-suit-master-el7/test_logs/basic-suite-master/post-002_bootstrap.py/lago-basic-suite-master-host-0/_var_log/messages/*view*/
>> > we have
>> >
>> > Aug 30 05:55:52 lago-basic-suite-master-host-0 journal: vdsm
>> > throttled WARN MOM not available.
>> > Aug 30 05:55:52 lago-basic-suite-master-host-0 journal: vdsm
>> > throttled WARN MOM not available, KSM stats will be missing.
>> > Aug 30 05:55:53 lago-basic-suite-master-host-0 systemd: Started
>> > Session 10 of user root.
>> > Aug 30 05:55:53 lago-basic-suite-master-host-0 systemd-logind: New
>> > session 10 of user root.
>> > Aug 30 05:55:53 lago-basic-suite-master-host-0 systemd: Starting
>> > Session 10 of user root.
>> > Aug 30 05:55:53 lago-basic-suite-master-host-0 python: ansible-setup
>> > Invoked with filter=* gather_subset=['all']
>> > fact_path=/etc/ansible/facts.d gather_timeout=10
>> > Aug 30 05:55:54 lago-basic-suite-master-host-0 python:
>> > ansible-command Invoked with warn=True executable=None
>> > _uses_shell=False _raw_params=bash -c "rpm -qi vdsm | grep -oE
>> > 'Version\\s+:\\s+[0-9\\.]+' | awk '{print $3}'" removes=None
>> > creates=None chdir=None
>> > Aug 30 05:55:54 lago-basic-suite-master-host-0 python:
>> > ansible-systemd Invoked with no_block=False name=libvirt-guests
>> > enabled=True daemon_reload=False state=started user=False
>> > masked=None Aug 30 05:55:54 lago-basic-suite-master-host-0 systemd:
>> > Reloading. Aug 30 05:55:55 lago-basic-suite-master-host-0 systemd:
>> > Cannot add dependency job for unit lvm2-lvmetad.socket, ignoring:
>> > Unit is masked. Aug 30 05:55:55 lago-basic-suite-master-host-0
>> > systemd: Stopped MOM instance configured for VDSM purposes.
>> > Aug 30 05:55:55 lago-basic-suite-master-host-0 systemd: Stopping
>> > Virtual Desktop Server Manager...
>> > Aug 30 05:55:55 lago-basic-suite-master-host-0 systemd: Starting
>> > Suspend Active Libvirt Guests...
>> > Aug 30 05:55:55 lago-basic-suite-master-host-0 systemd: Started
>> > Suspend Active Libvirt Guests.
>> > Aug 30 05:55:55 lago-basic-suite-master-host-0 journal: libvirt
>> > version: 2.0.0, package: 10.el7_3.9 (CentOS BuildSystem
>> > <http://bugs.centos.org>, 2017-05-25-20:52:28, c1bm.rdu2.centos.org)
>> > Aug 30 05:55:55 lago-basic-suite-master-host-0 journal: hostname:
>> > lago-basic-suite-master-host-0.lago.local
>> > Aug 30 05:55:55 lago-basic-suite-master-host-0 vdsmd_init_common.sh:
>> > vdsm: Running run_final_hooks
>> > Aug 30 05:55:55 lago-basic-suite-master-host-0 journal: End of file
>> > while reading data: Input/output error
>> > Aug 30 05:55:55 lago-basic-suite-master-host-0 systemd: Stopped
>> > Virtual Desktop Server Manager.
>> >
>> > suggesting that systemd stopped vdsmd since mom did not start on
>> > time. We have an extremely ugly two-way dependency between the two
>> > services. Each depends on the other one. I am guessing that
>> > systemd's grace period for such a cycle has expired, and the couple
>> > was taken down.
>> >
>> > Does anybody have a guess what could cause such slowlness?
>>
>> Even if the LLDP errors does not have a negative impact of the
>> functional behavior of supervdsmd, they have a impact timing behavior
>> of the initialization of supervdsmd.
>>
>> I will post a patch today which moves the initialization of LLDP in an
>> extra thread, to make the initialization of LLDP asynchronly.
>>
>>
>
> I analyzed this issue further by running OST [1] while disabled [2] the
> suspected initialization of LLDP in supervdsmd.
> Like expected, there are no log entries about enabling LLDP in [3],
> but the OST is still failing with same symptoms.
>
> So the reason for this failure is still unknown.
>
>
> [1]
>   http://jenkins.ovirt.org/view/oVirt%20system%20tests/job/ovirt-system-tests_manual/1045/
>
> [2]
>   https://gerrit.ovirt.org/#/c/81232/1/lib/vdsm/network/initializer.py
>
> [3]
>   http://jenkins.ovirt.org/view/oVirt%20system%20tests/job/ovirt-system-tests_manual/1045/artifact/exported-artifacts/test_logs/basic-suite-master/post-002_bootstrap.py/lago-basic-suite-master-host-0/_var_log/vdsm/supervdsm.log/*view*/
>

So we're back in square one.
Another possible culprit may be ansible: Vdsm is stopped two seconds
after it logs to the host.

Aug 30 11:26:24 lago-basic-suite-master-host-0 systemd: Starting
Session 10 of user root.
Aug 30 11:26:25 lago-basic-suite-master-host-0 python: ansible-setup
Invoked with filter=* gather_subset=['all']
fact_path=/etc/ansible/facts.d gather_timeout=10
Aug 30 11:26:25 lago-basic-suite-master-host-0 python: ansible-command
Invoked with warn=True executable=None _uses_shell=False
_raw_params=bash -c "rpm -qi vdsm | grep -oE
'Version\\s+:\\s+[0-9\\.]+' | awk '{print $3}'" removes=None
creates=None chdir=None
Aug 30 11:26:26 lago-basic-suite-master-host-0 python: ansible-systemd
Invoked with no_block=False name=libvirt-guests enabled=True
daemon_reload=False state=started user=False masked=None
Aug 30 11:26:26 lago-basic-suite-master-host-0 systemd: Reloading.
Aug 30 11:26:26 lago-basic-suite-master-host-0 systemd: Cannot add
dependency job for unit lvm2-lvmetad.socket, ignoring: Unit is masked.
Aug 30 11:26:26 lago-basic-suite-master-host-0 systemd: Stopped MOM
instance configured for VDSM purposes.
Aug 30 11:26:26 lago-basic-suite-master-host-0 systemd: Stopping
Virtual Desktop Server Manager...


could it be that it triggers a systemd-reload that makes systemd croak
on the vdsm-mom cycle?