RDO tickets are now tracked in Jira https://issues.redhat.com/projects/RDO/issues/
Bug 1193429 - failed to flow_del
Summary: failed to flow_del
Keywords:
Status: CLOSED CURRENTRELEASE
Alias: None
Product: RDO
Classification: Community
Component: openvswitch
Version: Juno
Hardware: Unspecified
OS: Unspecified
unspecified
unspecified
Target Milestone: Milestone3
: Kilo
Assignee: Alan Pevec (Fedora)
QA Contact: Ofer Blaut
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2015-02-17 10:29 UTC by Edouard Bourguignon
Modified: 2016-04-26 18:03 UTC (History)
7 users (show)

Fixed In Version: openvswitch-2.3.1-2.el7
Doc Type: Bug Fix
Doc Text:
Clone Of:
Environment:
Last Closed: 2015-04-11 20:10:25 UTC
Embargoed:


Attachments (Terms of Use)

Description Edouard Bourguignon 2015-02-17 10:29:49 UTC
Description of problem:

flooding error messages about failed to flow_del

/var/log/openvswitch/ovs-vswitchd.log-20150215:2015-02-13T17:43:54.555Z|00011|dpif(revalidator_7)|WARN|system@ovs-system: failed to flow_del (No such file or directory) skb_priority(0),in_port(6),skb_mark(0),eth(src=00:1a:a0:28:ca:cc,dst=ff:ff:ff:ff:ff:ff),eth_type(0x8100),vlan(vid=4001,pcp=0),encap(eth_type(0x0806),arp(sip=10.156.29.184,tip=10.156.20.110,op=1,sha=00:1a:a0:28:ca:cc,tha=00:00:00:00:00:00))

Version-Release number of selected component (if applicable):

We are using kernel 3.10.0-123.20.1.el7.x86_64 and openvswitch-2.1.2-2.el7.centos.1.x86_64

How reproducible:

static

Steps to Reproduce:
1. install openstack juno
2. start a few instances
3. tail -f /var/log/openvswitch/ovs-vswitchd.log

Actual results:
Warning messages appear in /var/log/openvswitch/ovs-vswitchd.log

Expected results:
No warning messages

Additional info:
Known bug
https://github.com/openvswitch/ovs/commit/3601bd879

Fixed in upstream 2.1.3. Is it possible to have it backported to el7?

Best regards.

Comment 1 Edouard Bourguignon 2015-02-25 09:34:45 UTC
From RedHat Customer Service, this bug seems to be fixed in openvswitch-2.1.2-2.el7_0.1. Where can we find it? Why it's not in RDO repository?

Comment 2 Alan Pevec (Fedora) 2015-02-25 19:16:25 UTC
> Known bug
> https://github.com/openvswitch/ovs/commit/3601bd879
> 
> Fixed in upstream 2.1.3. Is it possible to have it backported to el7?

https://repos.fedorapeople.org/repos/openstack/openstack-juno/epel-7/openvswitch-2.1.2-2.el7.centos.1.x86_64.rpm was built[*] from openvswitch-2.1.2-2.el7_0.1 SRPM which has changelog entry:
- updated with git snapshot up to 3f0588b

which includes 3601bd879.

[*] https://copr.fedoraproject.org/coprs/jruzicka/rdo-juno-epel-7/build/52323/


> Fixed in upstream 2.1.3. Is it possible to have it backported to el7?

Has that been confirmed? There are only few unrelated patches 3f0588b..v2.1.3

We'll be able to rebase to 2.3.0 after centos 7.1 is available (kernel changes are required).

Comment 3 Edouard Bourguignon 2015-02-26 15:01:23 UTC
Nope it seems redhat customer support is wrong, we still have the error with their package:
2015-02-23T02:23:57.879Z|646607|dpif(revalidator_7)|WARN|system@ovs-system: failed to flow_del (No such file or directory) skb_priority(0),tunnel(tun_id=0x68,src=192.168.40.104,dst=192.168.40.101,tos=0x0,ttl=64,flags(key)),in_port(8),skb_mark(0),eth(src=02:00:c0:a8:c7:d2,dst=02:00:c0:a8:c7:d0),eth_type(0x8100),vlan(vid=199,pcp=0),encap(eth_type(0x0800),ipv4(src=192.168.199.210,dst=192.168.199.208,proto=6,tos=0,ttl=64,frag=no),tcp(src=50631,dst=8300),tcp_flags(0x002))

Is there a release date for CentOS 7.1?

Comment 4 Edouard Bourguignon 2015-02-26 15:03:52 UTC
Oops sorry wrong log (should have read the date). What is strange is the redhat package has the same git commit for its sources, and same spec file. But with this redhat package we don't have the flow_del error...

Comment 5 Alan Pevec 2015-03-01 22:10:55 UTC
> But with this redhat package we don't have the flow_del error...

Was that on the same machine (was it RHEL or CentOS?)
i.e. you had openvswitch-2.1.2-2.el7.centos.1 installed and got flow_del errors, then _only_ upgraded to openvswitch-2.1.2-2.el7_0.1 and errors were gone?

Comment 6 Edouard Bourguignon 2015-03-02 12:38:19 UTC
Same machine, CentOS 7 (but we can have RHEL we are on our way to pay a subscription on Openstack platform).

So we have flow_del errors on openvswitch-2.1.2-2.el7.centos.1 running on CentOS. Then _only_ upgrading to openvswitch-2.1.2-2.el7_0.1 fix all the errors. At least it's running since a couple of days, still nothing. We don't understand why. If you have any idea?

Comment 7 Edouard Bourguignon 2015-03-02 16:45:34 UTC
A bit more fun, not sure if quite relevant:

Here is a dump from in order redhat srpm rebuilt in mock against centos/epel7 repo, redhat openvswitch package, and rdo package (only ovs-vswitchd):
SIZE    CHECKSUMS
1569432 0b144c4f31595d7a98c804df2b22a0d330d1bb656ea56e8263cd65dbb920a2af
1569448 f6806bf3435ca7b60de4bac2ac2acc2f671813117c505d28a661b6f41ae7ee7f
1569448 3d8c6ef31a67700760f9b92f530223e2ee3ca77c2cc50c9ac5263525094a2786

so 3 different binaries from same sources and same spec file. is it normal?

Comment 8 Edouard Bourguignon 2015-03-04 09:40:15 UTC
We just had this error again so definitively redhat package is not fixing the problem:

2015-03-03T01:17:33.257Z|00001|dpif(revalidator_7)|WARN|system@ovs-system: failed to flow_del (No such file or directory) skb_priority(0),in_port(6),skb_mark(0),eth(src=00:1a:64:a2:24:fe,dst=ff:ff:ff:ff:ff:ff),eth_type(0x8100),vlan(vid=4001,pcp=0),encap(eth_type(0x0806),arp(sip=10.156.1.186,tip=10.156.28.48,op=1,sha=00:1a:64:a2:24:fe,tha=00:00:00:00:00:00))

Which make more sense since it's the same sources as rdo package.

Comment 9 Alan Pevec 2015-03-11 18:38:01 UTC
Haikel tells me ovs 2.3.1 works on EL7.0 kernel just that there might be some races. (Haikel, do you have details?)
Please try CBS build https://cbs.centos.org/koji/buildinfo?buildID=568
if that works for you, we'll push it as RDO Juno update asap.

Comment 10 Edouard Bourguignon 2015-03-12 10:01:14 UTC
With this package we have more warnings as soon as we start the openvswitch service:

2015-03-12T09:58:41.985Z|00005|ofproto_dpif|INFO|system@ovs-system: Datapath does not support recirculation
2015-03-12T09:58:41.985Z|00006|dpif|WARN|system@ovs-system: execute userspace(pid=0,userdata(00000000)) failed (Invalid argument) on packet metadata=0,in_port=0,vlan_tci=0x0000,dl_src=00:00:00:00:00:00,dl_dst=00:00:00:00:00:00,dl_type=0x1234
2015-03-12T09:58:41.985Z|00007|ofproto_dpif|WARN|system@ovs-system: variable-length userdata feature probe failed (Invalid argument)
2015-03-12T09:58:41.985Z|00008|dpif|WARN|system@ovs-system: failed to put[create] (Invalid argument) skb_priority(0),skb_mark(0),in_port(0),eth(src=00:00:00:00:00:00,dst=00:00:00:00:00:00),eth_type(0x8847),mpls(label=0,tc=0,ttl=0,bos=1)

We also have this, could be related to selinux:
2015-03-12T09:56:25.212Z|00001|ofproto_dpif_xlate(handler10)|WARN|/proc/sys/net/core/netdev_max_backlog: open failed (Permission denied)

Comment 11 hguemar 2015-03-13 08:31:30 UTC
Flavio Leitner says it should work and I haven't run into any issues with it so far.

"You should be able to use openvswitch with any RHEL-7 kernel, but
versions prior to the above missed one bugfix, so I'd recommend
to use at least that version."

http://openvswitch.org/pipermail/dev/2014-November/048319.html 
http://openvswitch.org/pipermail/discuss/2014-November/015511.html

Comment 12 Alan Pevec 2015-03-18 02:25:03 UTC
Warnings from comment 10 are also present with 7.1 kernel 3.10.0-229.el7.x86_64

Comment 13 Alan Pevec 2015-03-26 13:07:44 UTC
ovs 2.1.2 build we have in RDO repos crashes with 7.1 kernel (bug 1200918) so we need to push ovs 2.3.1 update asap.

Comment 14 Alan Pevec 2015-03-30 16:51:13 UTC
RDO Juno update https://review.gerrithub.io/228852


Note You need to log in before you can comment on or make changes to this bug.