Andrey is correct. It appears that balance-tcp requires successful LACP negotiation. See here: https://github.com/osrg/openvswitch/blob/master/lib/bond.c#L610 and here: https://github.com/osrg/openvswitch/blob/master/lib/bond.c#L1438. This also means that when we create bonds with balance-tcp we need to configure lacp as well.
On Mon, Feb 24, 2014 at 3:14 PM, Andrey Danin <[email protected]> wrote: > And yes, the bug https://bugs.launchpad.net/fuel/+bug/1272842 and current > problem can be unrelated but they have similar error messages in OVS logs. > > > On Tue, Feb 25, 2014 at 2:55 AM, Andrey Danin <[email protected]> wrote: > >> Guys, I set up hardware (2 nodes) and software (3 nodes) labs today with >> ISO #181 to test bonding. Unfortunately, balance-tcp mode is totally >> broken. When I use it during deployment or switch to it in a working >> cluster, all traffic stops. Playing with rebalance interval doesn't help. >> On the contrary, balance-slb works fine. Both Ubuntu (Hhardware nodes) >> and CentOS (virtual env) works without any traffic lost. I'm running a >> flooded ping between virtual instances inside of clouds for a night and >> will check a number of lost packets. Also I want to play with iperf. >> >> Next things we can do: >> * Build an ISO with stable (1.9.3) or newest (2.0.x) version of OVS and >> play with them. Yesterday we decided to build Ubuntu 12.04 with Debian Sid >> 1.9.3 version of OVS. There is the ticket about that >> https://mirantis.jira.com/browse/OSCI-1089 Also Igor built its own >> version of an ISO with Sid package. >> * Dump openflow rules in balance-tcp mode and try to fix them. It's hard >> to do that because Aliens developed their syntax. >> * Run Igor's tests again and again until balance-slb starts block a >> traffic. Then dig into openflow rules. >> * Play with LACP on a real hardware. Maybe balance-tcp can be used only >> with lacp=active. >> * Ask the openvswitch community about our problems. >> >> Andrew, yes, the PXE network still nailed to an interface. I hope we will >> fix it in 5.0. >> >> >> On Tue, Feb 25, 2014 at 12:20 AM, Igor Shishkin >> <[email protected]>wrote: >> >>> Hello, Dmitry. >>> >>> It's 100% reproducible on virtual environment when we're trying to >>> deploy bonding in balance tcp or balance slb mode. >>> Tests related as a way to reproduce and a warning why these tests should >>> fail when they'll be merged. >>> >>> As we can see problem is in rebalance procedure openvswitch tries to do >>> since it started bonded interface. And in this time bonded interfaces stops >>> to accept ARPs. >>> >>> I just built openvswitch=1.9.3 which is LTS and wanna try it in the same >>> case and try to descrease bond-rebalance-interval to 0(as Andrey K. >>> suggested). If any of this will help - this could be the solution(but I'm >>> really not sure bond-rebalance-interval=0 is a good way). >>> -- >>> Igor Shishkin >>> QA Engineer >>> >>> >>> >>> On 24 Feb 2014, at 23:59, Dmitry Borodaenko <[email protected]> >>> wrote: >>> >>> > Mike, Igor, >>> > >>> > Can you provide more details on how the integration test in review >>> > #75161 helps to reproduce bug #1272842? >>> > >>> > As far as I understand, the bug is a highly intermittent problem with >>> > ARP that was only showing up after an environment with LACP bonding >>> > was operational for at least a few hours. >>> > >>> > On the other hand, the problem Igor is reporting based on the >>> > integration test sounds like something 100% reproducible that doesn't >>> > require real hardware or LACP and is not necessarily related to ARP. >>> > >>> > Are you sure you're not confusing two unrelated problems? >>> > >>> > Thanks, >>> > -DmitryB >>> > >>> > >>> > On Mon, Feb 24, 2014 at 9:18 AM, Mike Scherbakov >>> > <[email protected]> wrote: >>> >> The issue is here: https://bugs.launchpad.net/fuel/+bug/1272842. >>> >> Those who know what can be wrong with our openvswitch/kernel, please >>> provide >>> >> your input.. >>> >> >>> >> >>> >> On Mon, Feb 24, 2014 at 9:04 PM, Igor Shishkin < >>> [email protected]> >>> >> wrote: >>> >>> >>> >>> Hello, >>> >>> >>> >>> Currently we have this review https://review.openstack.org/#/c/75161with >>> >>> test cases for our brand new shiny bonding feature but >>> >>> balance-tcp/balance-slb modes are not working for now. >>> >>> >>> >>> Steps to reproduce are very simple: >>> >>> Create cluster with simple or HA configuration, select balance-tcp or >>> >>> balance-slb bonding mode and start deployment. >>> >>> >>> >>> Deployment will not finish with success because of rebalance >>> procedure >>> >>> problems. >>> >>> -- >>> >>> Igor Shishkin >>> >>> QA Engineer >>> >>> >>> >>> >>> >>> >>> >> >>> >> >>> >> >>> >> -- >>> >> Mike Scherbakov >>> >> #mihgen >>> >> >>> >> -- >>> >> Mailing list: https://launchpad.net/~fuel-dev >>> >> Post to : [email protected] >>> >> Unsubscribe : https://launchpad.net/~fuel-dev >>> >> More help : https://help.launchpad.net/ListHelp >>> >> >>> > >>> > >>> > >>> > -- >>> > Dmitry Borodaenko >>> >>> >>> -- >>> Mailing list: https://launchpad.net/~fuel-dev >>> Post to : [email protected] >>> Unsubscribe : https://launchpad.net/~fuel-dev >>> More help : https://help.launchpad.net/ListHelp >>> >> >> >> >> -- >> Andrey Danin >> [email protected] >> skype: gcon.monolake >> > > > > -- > Andrey Danin > [email protected] > skype: gcon.monolake > > -- > Mailing list: https://launchpad.net/~fuel-dev > Post to : [email protected] > Unsubscribe : https://launchpad.net/~fuel-dev > More help : https://help.launchpad.net/ListHelp > >
-- Mailing list: https://launchpad.net/~fuel-dev Post to : [email protected] Unsubscribe : https://launchpad.net/~fuel-dev More help : https://help.launchpad.net/ListHelp
