Tuesday, 2015-12-15

*** salv-orl_ has joined #openstack-neutron-ovn00:02
*** salv-orlando has quit IRC00:05
*** salv-orl_ has quit IRC00:11
*** azbiswas has joined #openstack-neutron-ovn00:42
*** azbiswas_ has joined #openstack-neutron-ovn00:52
*** azbiswas has quit IRC00:55
*** azbiswas has joined #openstack-neutron-ovn01:02
*** azbiswas_ has quit IRC01:05
*** openstack has joined #openstack-neutron-ovn01:06
*** azbiswas_ has joined #openstack-neutron-ovn01:09
Sam-I-Amrussellb: does starting ovn-northd automatically start ovsdb-server?01:09
*** azbiswas has quit IRC01:12
*** chandrav has quit IRC01:33
*** yamamoto has joined #openstack-neutron-ovn01:42
*** yamamoto has quit IRC01:52
*** yamamoto has joined #openstack-neutron-ovn01:52
*** yamamoto has quit IRC01:56
*** russellb has quit IRC02:07
*** russellb has joined #openstack-neutron-ovn02:09
*** salv-orlando has joined #openstack-neutron-ovn02:15
*** salv-orlando has quit IRC02:22
*** salv-orlando has joined #openstack-neutron-ovn02:31
*** azbiswas_ has quit IRC02:32
*** salv-orlando has quit IRC02:35
*** yamamoto has joined #openstack-neutron-ovn02:54
*** Guest647 has quit IRC03:12
*** chandrav has joined #openstack-neutron-ovn03:22
*** chandrav has quit IRC03:31
*** openstackstatus has quit IRC04:24
*** openstack has joined #openstack-neutron-ovn04:25
*** openstackstatus has joined #openstack-neutron-ovn04:25
*** ChanServ sets mode: +v openstackstatus04:25
*** salv-orl_ has joined #openstack-neutron-ovn04:37
*** salv-orl_ has quit IRC04:41
*** chandrav has joined #openstack-neutron-ovn04:42
*** salv-orlando has joined #openstack-neutron-ovn05:37
*** salv-orlando has quit IRC05:40
*** chandrav has quit IRC06:23
*** chandrav has joined #openstack-neutron-ovn07:04
*** fzdarsky_ has joined #openstack-neutron-ovn07:19
openstackgerritHan Zhou proposed openstack/networking-ovn: Disable ovsdb monitor2  https://review.openstack.org/25772108:07
*** roeyc has joined #openstack-neutron-ovn08:58
*** salv-orlando has joined #openstack-neutron-ovn09:19
*** fzdarsky_ is now known as fzdarsky|afk09:26
*** fzdarsky|afk has quit IRC09:30
openstackgerritgaryk proposed openstack/networking-ovn: Remove the openstack-common.conf file  https://review.openstack.org/25776209:35
openstackgerritgaryk proposed openstack/networking-ovn: Remove the openstack-common.conf file  https://review.openstack.org/25776209:39
*** openstackgerrit has quit IRC09:47
*** openstackgerrit has joined #openstack-neutron-ovn09:47
*** openstackgerrit has quit IRC10:32
*** openstackgerrit has joined #openstack-neutron-ovn10:32
*** roeyc has quit IRC10:38
*** ig0r_ has quit IRC11:20
*** roeyc has joined #openstack-neutron-ovn11:20
*** salv-orlando has quit IRC11:22
*** salv-orlando has joined #openstack-neutron-ovn11:22
*** fzdarsky|afk has joined #openstack-neutron-ovn11:33
*** fzdarsky|afk is now known as fzdarsky11:33
*** yamamoto has quit IRC11:42
*** yamamoto has joined #openstack-neutron-ovn11:44
*** yamamoto has quit IRC11:51
*** yamamoto has joined #openstack-neutron-ovn11:51
*** salv-orl_ has joined #openstack-neutron-ovn12:03
*** salv-orlando has quit IRC12:06
*** yamamoto has quit IRC12:14
*** rtheis has joined #openstack-neutron-ovn12:17
openstackgerritgaryk proposed openstack/networking-ovn: Introduce datastructure called PortInfo  https://review.openstack.org/25787712:43
*** salv-orl_ has quit IRC12:46
*** yamamoto has joined #openstack-neutron-ovn12:54
russellbSam-I-Am: it actually expects that openvswitch is already running (ovs-vswitchd and ovsdb-server).  When you start northd, it creates the OVN databases in the local ovsdb-server if they haven't been created yet13:36
Sam-I-Amrussellb: aight. i'll clarify.13:36
russellbthere's a bit of magic in there ... it was to avoid several more set up steps13:37
Sam-I-Amlooks like start openvswitch (which starts ovsdb-server) and then start ovn-northd13:38
*** yamamoto has quit IRC13:41
russellbSam-I-Am: yes, that's right13:51
*** regXboi has joined #openstack-neutron-ovn13:54
*** yamamoto has joined #openstack-neutron-ovn13:56
openstackgerritRussell Bryant proposed openstack/networking-ovn: Disable ovsdb monitor2  https://review.openstack.org/25772114:06
openstackgerritMerged openstack/networking-ovn: Remove the openstack-common.conf file  https://review.openstack.org/25776214:17
*** roeyc has quit IRC14:35
*** roeyc has joined #openstack-neutron-ovn14:36
*** dslevin has quit IRC14:42
*** chandrav has quit IRC15:05
*** dslevin has joined #openstack-neutron-ovn15:06
openstackgerritMerged openstack/networking-ovn: Disable ovsdb monitor2  https://review.openstack.org/25772115:59
*** dslevin has quit IRC16:06
*** yamamoto has quit IRC16:25
*** yamamoto has joined #openstack-neutron-ovn16:26
Sam-I-Amrussellb: moo16:38
russellbSam-I-Am: quack16:38
Sam-I-Ami saw you chatting about vlan aware vms16:39
russellbit's true!16:39
*** yamamoto has quit IRC16:39
Sam-I-Ami had to get away for a few mins and make coffee before bad things happened16:39
russellbi also need more coffee ...16:39
Sam-I-Amfirst question... ovs uses vlans internally which neutron modifies to vlan IDs it knows about16:40
Sam-I-Amhow does ovs pass vlans end-to-end?16:40
russellbnot sure i understand the question16:40
russellbmaybe an ascii diagram would help?  :-p16:41
* Sam-I-Am drops mic16:41
russellbok so, are you trying to reconcile the use of VLANs today in this new vlan-aware-vm world?16:41
russellbgot it, good question16:41
Sam-I-Amseems internal usage of vlans in ovs poses a problem16:41
russellbat any point in the path, a packet only has a single VLAN tag16:42
russellbtoday, there are 2 different cases of vlan tags, right?  the internal one for segmenting networks, and potentially the one used for the network itself16:42
* russellb is asking "right" a lot16:42
russellbok, in the new vlan-aware-vm world, we add a 3rd case that is still not overlapping16:43
russellbthe use of a vlan tag here is *only* as a packet goes in/out of the VM16:43
russellbit's an encapsulation to segment traffic over a single VIF16:43
russellbinside the VM, it can do whatever16:43
Sam-I-Amright. so you can have sub-ints on a vif that the guest OS can see?16:44
russellboutside the VM, we'll use that VLAN tag to know which neutron network to send the packet on to16:44
russellb"this packet had vlan 101, which means it was from logical port Foo, so we string the vlan tag and send on to network Bar"16:44
russellbnext step "network Bar uses vlan 201 internally, so apply that for now"16:44
Sam-I-Amok, so it sounds like the same implementation as the "other end" of a neutron network... where neutron tags traffic to ext nets16:44
Sam-I-Amexcept in this case, thats attached to a vm16:45
russellbon its way out to the network "vlan 201 means network Bar, and network Bar is actually vlan 301 on the real network here, so use that tag now"16:45
Sam-I-Amah. ok. so there's no q-in-q or trunking within ovs, so to speak.16:45
Sam-I-Amwow, all this without coffee16:45
Sam-I-Amso how about linux bridge?16:45
* russellb doesn't care about linux bridge16:46
russellbto be blunt16:46
russellbat some point, if you want to have a programmable network, you use the tools designed for a programmable network16:46
Sam-I-Amseems like a lot of this relies on nova knowing what to do16:46
Sam-I-Amyeah, this is why ovn will be nice.16:47
Sam-I-Amfinally using ovs without all the conventional baggage16:47
russellbmore seriously, i'm not aware of any plans to implement this for LB16:47
Sam-I-Am(which hopefully does vlan-aware vms too)16:47
russellbyes, OVN already has support for this16:47
russellbin fact, we have a binding:profile hack for doing vlan-aware-vms already16:47
*** roeyc has quit IRC16:47
russellbit was intended for containers-in-vms, but it's this model16:47
*** roeyc has joined #openstack-neutron-ovn16:48
Sam-I-Amlinuxbridge has appeal for certain operators, but if ovn is done right, i think it will gain adoption16:48
Sam-I-Amcan you think of a migration path from nova-net to ovn? or even linuxbridge to ovn?16:49
Sam-I-Amthere are some docs floating around for nova-net to neutron, but using linuxbridge in the most simplest sense.16:50
russellbLB is great if the use case doesn't need more complex programming capabilities16:50
Sam-I-Amin theory i guess you can deploy both linuxbridge and ovn/ovs in one deployment16:51
Sam-I-Amnext question - neutron-ovs-bridge-name bp16:53
russellbwelllllll not really16:54
russellbi'm not sure if that works with ml2+ovs either16:55
russellbbridge-name bp, plz hold, i need to meet wife for lunch16:55
Sam-I-Amha ok16:55
*** roeyc has quit IRC17:16
*** palexster has quit IRC17:22
*** chandrav has joined #openstack-neutron-ovn17:40
russellbSam-I-Am: i have returned.18:10
Sam-I-Amrussellb: so...18:21
Sam-I-Amprior to the bridge-name bp18:22
russellbSam-I-Am: shall we chat in #openstack-neutron ?18:22
Sam-I-Amin neutron.conf, you can specify ovs and lb mechanisms at the same time18:22
Sam-I-Amsure, we can go over there18:22
russellbwell, here is fine18:22
russellbdoes using both actually work?18:23
russellbhow does it choose which to use?18:23
russellbdepending on which agent that host is running?18:23
Sam-I-Amnot sure. devstack has both in the mechanism drivers.18:24
Sam-I-Ambut only includes one agent or the other18:24
russellbi see18:24
russellbi don't think vlan-aware-vms affects that18:25
russellbon another note, if you use ovn, that's all you use18:25
Sam-I-Amno, that was just a side note in my brain about helping people migrate from linuxbridge to ovs/ovn18:25
Sam-I-Amyeah, no ml2 there18:25
russellbre: migrating, it seems doable, it's all software right?18:25
russellbbut it would require some migration scripts ...18:25
Sam-I-Amthinking about a migration path makes sense though because it comes up a lot18:25
* russellb nods18:26
russellbi should lab up a devstack ml2+ovs and migrate that to ovn to start with18:26
russellbcould even do that with a grenade job18:26
russellbnova-net is a lot more work18:27
russellbbecause it requires data migration from nova to neutron18:27
russellbmigrating neutron ml2+ovs to ovn == probably pretty easy18:27
russellbmigrating neutron ml2+lb to ovn == probably still doable, but would require a script to re-plug all ports to ovs br-int18:27
russellbmigrating from nova-net == punt18:27
russellbthat's not solved in general, much less for ovn18:28
russellbi don't think ovn makes it any easier18:28
russellbno harder either18:28
Sam-I-Amunfortunately migration from nova-net would be the most popular :/18:29
*** armax has quit IRC18:29
russellbi think there's a lot more people running ml2+ovs than nova-net at this point18:29
Sam-I-Ami can see that18:30
Sam-I-Amin any case, adoption is quicker when people can migrate rather than greenfield18:31
Sam-I-Amwhich is why linuxbridge has gained a little steam with nova-net people18:31
*** salv-orlando has joined #openstack-neutron-ovn18:31
russellbSam-I-Am: but nobody knows how to migrate that either18:35
russellbso that doesn't really make sense to me ...18:35
russellbi think LB gained steam because it's more well known to the average linux person18:35
Sam-I-Amrussellb: there are some docs for nova-net to a basic neutron-linuxbridge implementation that work18:36
russellband the fact that the ovs agent uses OVS largely like a linux bridge, it's hard to argue a strong case for the additional complexity18:36
russellbah ok, well that's cool18:36
Sam-I-Amunofficial docs, but they exist18:36
Sam-I-Ami'm trying to make them more official18:36
Sam-I-Amyeah, the extra complexity of OVS as just-a-switch doesn't tickle people... but i think OVN will if it eliminates the bottlenecks of conventional agents18:37
russellbthat's the hope ... we're getting there18:37
Sam-I-Amparticularly distributed routing and implementing secgroups without an additional linuxbridge18:37
russellbthe sec group thing is being done for the ovs agent as well18:38
russellbthe routing implementation is drastically different though18:38
russellbwe can do some more serious performance comparisons and such once we finish nat18:38
Sam-I-Amsadly, dvr did not have a good start18:39
Sam-I-Amthere was huge demand for distributed routing for a long time... and dvr released without any docs or testing on more than one node18:39
Sam-I-Ampeople jumped on dvr and quickly became frustrated18:40
Sam-I-Amhence my push to make ovn docs available prior to release18:40
*** arosen has joined #openstack-neutron-ovn18:40
russellbyou have my full support :)18:41
*** palexster has joined #openstack-neutron-ovn18:42
Sam-I-Amwhen is ovs 2.5 coming out?18:42
Sam-I-Amseems thats a prereq for making ovn easier to install18:43
russellbbranch has been created18:57
russellbrelease very early 201618:57
Sam-I-Amso soon after that distro packages should exist?19:02
Sam-I-Amat least the ones being referenced in the deployment integ guide19:02
*** armax has joined #openstack-neutron-ovn19:06
Sam-I-Amrussellb: if using the conventional l3 agent, should service_plugins include 'router' ?19:12
russellbSam-I-Am: we hard code it to empty19:18
russellbin our devstack19:18
russellbthat doesn't mean it's correct ...19:18
Sam-I-Amrussellb: yeah... curious how that impacts conventional l3 agent vs. ovn l319:19
russellbnot sure19:19
Sam-I-Amhave you tested conventional l3 support with empty service_plugins ?19:19
switchcadeSam-I-Am: ETA January, maybe late19:20
Sam-I-Amswitchcade: ovn l3 support?19:20
switchcadesorry, was just catching up, OVS-2.5.19:21
switchcaderussellb: would know the OVN pieces better than I19:21
russellbSam-I-Am: "tested" yes :)19:22
russellbSam-I-Am: that's what it's set to for every devstack run19:22
russellband it passes tempest19:22
russellbnote that we have a custom tempest config19:22
russellbthat disables some stuff19:22
Sam-I-Amdoes it test any l3 stuff?19:23
russellbbut it definitely uses a neutron router19:23
russellbdefault devstack network setup, public and private nets with logical router in between19:23
russellbtests boot VMs on private, create floating IP on public, talk to VM through that floating IP19:23
russellball is happy19:23
Sam-I-Amwell then19:23
Sam-I-Amdo you have a deployment i can poke at?19:24
russellbSam-I-Am: nope ... but we're only a stack.sh away from having one19:25
russellbswitchcade: can I get an ACK on this little v2?  :-)  http://openvswitch.org/pipermail/dev/2015-December/063390.html19:26
Sam-I-Amok, i can just build one here19:26
russellbSam-I-Am: ok19:26
russellbSam-I-Am: i'm sure you've seen http://docs.openstack.org/developer/networking-ovn/testing.html19:26
Sam-I-Amrussellb: yup19:29
*** openstackgerrit has quit IRC19:32
*** openstackgerrit has joined #openstack-neutron-ovn19:32
switchcaderussellb: oh, sorry it slipped through. I'll take a look now.19:34
*** salv-orlando has quit IRC19:37
*** azbiswas has joined #openstack-neutron-ovn19:40
russellbswitchcade: no worries!19:44
russellbswitchcade: that's a follow-up to a fix i applied to branch-2.5, as well, so I guess I should put this in branch-2.5 as well?19:47
* russellb not sure how strict you guys usually are ...19:49
* russellb pushes it to branch-2.5 and hopes not to get wrist slapped :)19:51
russellbactually that code is different in branch-2.5 so i'm leaving it alone19:51
* russellb talks to himself some more19:51
Sam-I-Amrussellb: thats a thing19:52
openstackgerritHan Zhou proposed openstack/networking-ovn: Fix remote_group handling when remote_group is not self  https://review.openstack.org/25706720:05
Sam-I-Amrussellb: how quickly do you think rh will build ovs 2.5 packages?20:07
russellbin 3 minutes20:07
russellbbut seriously, we'll have it in RDO pretty quick20:07
Sam-I-Amawesome... which probably means the delorean master builds of openstack should also include the bits?20:08
russellbgood question20:10
russellbdon't know if they build ovs master there20:10
* russellb in meeting now20:10
Sam-I-Amlooks like no - https://trunk.rdoproject.org/f22/current/20:13
switchcaderussellb: I think we're not super strict right now, since we've just branched20:37
russellbswitchcade: ok20:37
switchcadeI think after christmas we'll be tightening it up more.20:37
switchcadeSomething cosmetic like that would rarely get a backport once the release is actually made20:38
russellbright, i figured20:38
Sam-I-Amrussellb: do you have any push for getting ovs master into rdo... or another repo that might work?20:40
russellbi'm sure i could20:40
russellbjust a question of time and priorities20:40
Sam-I-Amseems like it might be a good first step for making ovn usable outside of devstack... install fedora, roll the master packages, config, go.20:41
*** dslevin has joined #openstack-neutron-ovn20:47
*** dslev has joined #openstack-neutron-ovn21:02
russellbconverting the ovs python lib to Python 3 has been a lot more work than i expected21:07
russellbgood thing i didn't try to scope it before i started21:07
russellb24 commits,  39 files changed, 509 insertions(+), 305 deletions(-)21:07
russellbso far ...21:07
switchcadehmm. I always thought it was mostly about changing print foo -> print(foo) ;-)21:08
russellbyeah, that's part of it :)21:08
russellbthey decided if they were breaking print, might as well break everything21:08
russellbsome of that is style stuff that i couldn't help myself from21:11
switchcadeI had wondered if anyone has run pep8 on the codebase before21:49
russellbswitchcade: i left several of the violations ignored, but they're some of the more obnoxious violations21:52
russellbthere's some errors about how you do alignment of multi-line statements21:53
russellbyou have to get it just right or throws a fit21:53
russellbI fixed a bunch, but here's what I left turned off -- https://gist.github.com/russellb/fc9bd7c6c2d55107f8f721:54
russellb21% test coverage!21:54
switchcaderussellb: ah yeah, seems reasonable to me. So long as there's no mixed tabs/spaces ;-)21:58
switchcadeI'm guessing that some of our C style has bled into the python22:00
russellbyeah, it's all good22:00
switchcadeand that's not the end of the world.22:00
russellbit works, that's what matters22:00
switchcade21% seems pretty low, although I'm really not aware of what the python code we have does22:01
russellbthe only "testing" is importing the code22:01
switchcadeah, I see22:01
russellbbaby steps22:01
switchcadeyep, makes sense22:02
russellbjust that much of a skeleton test env was broken for Python 3.4 until my last commit22:02
*** regXboi has quit IRC22:02
russellbi probably could have made it pass sooner, but i was fixing everything I knew about22:03
*** rtheis has quit IRC22:10
*** salv-orlando has joined #openstack-neutron-ovn22:29
*** dslev has quit IRC22:39
*** dslev_ has joined #openstack-neutron-ovn22:47
Sam-I-Amrussellb: which rpm includes ovn-controller ?23:20
Sam-I-Amor will include...23:20
Sam-I-Amopenvswitch-ovn ?23:20
*** salv-orlando has quit IRC23:23

Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!