Friday, 2026-05-08

opendevreviewHelen Chen proposed openstack/neutron master: Created OVN Agent EVPN extension & netlink monitor  https://review.opendev.org/c/openstack/neutron/+/98440902:09
opendevreviewBrian Haley proposed openstack/neutron master: WIP: Add project_id attributes for neutron-lib change  https://review.opendev.org/c/openstack/neutron/+/98696202:22
opendevreviewKyuyeong Lee proposed openstack/neutron-specs master: Add spec for Port Groups for Security Group Rules  https://review.opendev.org/c/openstack/neutron-specs/+/98778602:38
opendevreviewKyuyeong Lee proposed openstack/neutron-specs master: Add spec for Port Groups for Security Group Rules  https://review.opendev.org/c/openstack/neutron-specs/+/98778603:11
opendevreviewBrian Haley proposed openstack/neutron master: WIP: Add project_id attributes for neutron-lib change  https://review.opendev.org/c/openstack/neutron/+/98696204:29
opendevreviewBrian Haley proposed openstack/neutron master: WIP: Add project_id attributes for neutron-lib change  https://review.opendev.org/c/openstack/neutron/+/98696206:19
opendevreviewRodolfo Alonso proposed openstack/neutron master: Use ``query.first() is not None`` to check resources  https://review.opendev.org/c/openstack/neutron/+/98651306:41
opendevreviewRodolfo Alonso proposed openstack/neutron-lib master: Add security-groups-default-statefulness API definition  https://review.opendev.org/c/openstack/neutron-lib/+/98435407:01
ralonsohhey folks, if you have a couple of mins, please check https://review.opendev.org/c/openstack/neutron/+/98681407:10
ralonsohthanks!07:10
opendevreviewRodolfo Alonso proposed openstack/neutron-lib master: Add security-groups-default-statefulness API definition  https://review.opendev.org/c/openstack/neutron-lib/+/98435409:36
opendevreviewMerged openstack/neutron master: Add missing address_group CRUD and action policy rules  https://review.opendev.org/c/openstack/neutron/+/98681409:53
opendevreviewRodolfo Alonso proposed openstack/neutron master: Split L3RpcNotifierMixin to avoid RPC callbacks for OVN  https://review.opendev.org/c/openstack/neutron/+/98750210:10
opendevreviewRodolfo Alonso proposed openstack/neutron master: WIP == Add ``l3-agent-scheduler-ha-priority`` extension for ML2/OVN  https://review.opendev.org/c/openstack/neutron/+/98279211:21
ralonsoheolivare, hello! the non-voting tempest bgp job is a but unstable11:26
ralonsohand we see several retries that slow down the CI11:26
ralonsohdo you mind if I push a patch to set attemps=1 in this job?11:26
opendevreviewEduardo Olivares proposed openstack/neutron master: Add compute1 node to BGP multinode tempest job  https://review.opendev.org/c/openstack/neutron/+/98607511:29
opendevreviewRodolfo Alonso proposed openstack/neutron master: Set attempts=1 in ``neutron-ovn-bgp-tempest-multinode`` job  https://review.opendev.org/c/openstack/neutron/+/98782811:30
ralonsoheolivare, ^ 11:30
mlavallehaleyb: are we meeting today?13:03
haleybmlavalle: i had asked ralonsoh to run today's meeting yesterday as i have an appointment i need to leave for in a minute13:04
mlavalleack13:05
*** haleyb is now known as haleyb|out13:08
* mlavalle assumes no meeting today13:16
opendevreviewRodolfo Alonso proposed openstack/neutron master: Use Neutron FIP UUID as OVN ``NAT`` row UUID  https://review.opendev.org/c/openstack/neutron/+/98440813:17
ralonsohmlavalle, there is an agenda and I'm going to chair the meeting13:17
ralonsohthere are 2 topics to be discussed13:17
ralonsohupsss13:18
ralonsohthe meeting is at 1300UTC, my bad13:19
ralonsoh#startmeeting neutron_drivers13:19
opendevmeetMeeting started Fri May  8 13:19:11 2026 UTC and is due to finish in 60 minutes.  The chair is ralonsoh. Information about MeetBot at http://wiki.debian.org/MeetBot.13:19
opendevmeetUseful Commands: #action #agreed #help #info #idea #link #topic #startvote.13:19
opendevmeetThe meeting name has been set to 'neutron_drivers'13:19
ralonsohI need to update my calendar13:19
ralonsohone sec, the agenda is loading...13:19
ralonsohPing list: ykarel, mlavalle, mtomaska, slaweq, tobias-urdin, lajoskatona, haleyb, ralonsoh, cardoe 13:20
lajoskatonao/13:20
cardoeo/13:20
ralonsohsorry for the delay, I had this meeting in my calendar 1 hour later, my bad13:20
cardoeso did I13:20
ralonsohhmmmm I don't know if other folks are going to join13:21
ralonsohthat was my bad for calling this meeting so late13:21
ralonsohslaweq? mlavalle?13:22
slaweqhi13:22
slaweqsorry, I forgot what time it is13:22
slaweqI am here13:22
ralonsohok, we have a couple of topics, I hope we don't need to vote13:23
ralonsohfirst one13:23
ralonsoh#link https://review.opendev.org/c/openstack/neutron/+/98573213:23
ralonsohproposed by cardoe 13:23
cardoeI'm just looking for some feedback as to what path forward you'd like13:24
cardoeIt's not clear to me from the review comments as to what's our next path13:24
ralonsohso, IMO, we should do a couple of things13:25
cardoeCause it says to add a new exception but it's got one13:25
ralonsoh1) create a specific exception for this case13:25
ralonsoh2) handle the HPB, just in case we raise this exception in a higher level, to undo the previous ones13:26
ralonsohand in MechanismManager.bind_port, report another log error13:27
slaweqI will need to go deeper into this but for now I don't really get why to do that. Drivers are kinda independed of each other, why one want to be so much "selfish" to say something like 'if I can't bind this port, nobody will'13:28
ralonsohyeah that is a very exceptional case, to be honest13:29
ralonsohthat is required by Ironic and should be handled with care. As you said, we should let all drivers to first check if they can bind13:29
ralonsohbut this is something exceptional13:29
lajoskatonaslaweq: +113:30
cardoeSo this is for layered binding13:30
cardoeSo the lower segments say this cannot bind. It's definitely not a common case but a special case where things break.13:31
lajoskatonacan't we say that there is a cfg option and if you set it your drivers can stop the binding process and make it fail for all, but the default is to have it as it is now?13:31
cardoeSure. There's no ability to raise a stop right now at all.13:31
ralonsohlajoskatona, if we implement this kind of capability (stop binding), I would prefer not to have a config option13:32
ralonsohis the mech driver reponsibility to be correctly implemented and to provide a responsible and justified exception in that case13:32
slaweqralonsoh I agree, if we want to do it that way, we should try to do it without another config knob13:32
lajoskatonathat can break things, but not sure how other drivers use the HPB13:32
ralonsohlajoskatona, I know I know13:33
cardoeWhat's HPB?13:33
ralonsohhierchical port binding13:33
lajoskatonahyerarchical port binding, sorry13:33
cardoeah okay13:33
slaweqcardoe sorry for having those doubts now, I know it was discussed during PTG already, I was just not that much  into this topic then. I am not going to block this as it was already agreed that should be changed. I just wanted to understand it better13:33
cardoeThis is for HPB13:34
cardoeSo the top layer works but the bottom layer cannot work.13:34
ralonsohso, for example, apart from Ironic, I don't know other mech drivers that will implement this. If we do this, it will be the responsibility of this driver to provide a justification for this binding stop13:34
ralonsohthat's a lot of "power" for a single mech driver13:35
cardoee.g. I'm touching a fabric in layer 0 and then touching a switch with the physical port in layer 1.13:35
cardoelayer 0 works and configures the fabric plane13:38
cardoeBut then the switch itself fails to take the changes13:38
cardoeYou need to not say that this binding worked and let Nova think that this port is bound13:38
cardoeAnd you need to have the ability to unwind the change.13:38
slaweqcardoe isn't port in `binding_failed` state in such case?13:38
ralonsohTheJulia, hi! ^^ can you answer that?13:40
cardoeBut neutron will still attempt to bind with another driver no?13:40
cardoeIt also won't unwind for us in reverse order?13:40
ralonsohI think the problem was that: if the expected layer 2 fails and Neutron tries, succesfully, to bind this port to another driver13:41
slaweqahh, and then we end up in situation where port is bound on different layers with potentially "uncompatible" drivers, correct?13:42
ralonsohfrom the LP bug:13:42
ralonsohbut the other issue is the silent failure can still cause the binding process to continue into other ML2 plugins which might believe they have successfully bound the port, which means the port likely suggests binding was successful.13:43
cardoeslaweq: exactly13:43
ralonsohso this "exception" wants to prevent this explicitly: if a driver knows that must be bound to itself but that doesn't happen, stop any further try13:43
slaweqand to avoid that in case if layer 1 binding with proper mechanism driver failed tell neutron "stop, we need to revert uper level bindings and don't try to bind on this layer with others"13:43
TheJuliahuh, what13:44
cardoeSo like in my chain, we've got the l2vni as the top layer and then n-g-s as the lower layer and then BGP as the lowest layer.13:44
TheJuliawhere should I start reading?13:44
ralonsohTheJulia, we are talking about https://review.opendev.org/c/openstack/neutron/+/985732/1/neutron/plugins/ml2/common/exceptions.py13:44
ralonsohin particular, why do we need this new "exception"13:44
ralonsohI copy/pasted a snippet ^13:45
cardoesorry split brain because my boss is having a 1:1 with me13:45
lajoskatonaso this is not changing the contract we give to binding users by stopping it but actually making it work as it should have worked from the beggining?13:45
TheJuliaI mean, any exception13:45
ralonsohlajoskatona, well, the current contract is "I can't, you try now". That will break it13:46
TheJuliaAny special exception would do the needful, the key is "this will never work, the plugin has enough visibility to know that its impossible. It has a duty to set context otherwise there is a casm of lacking context and expectations of users not met.13:46
slaweqralonsoh maybe we can avoid doing that "hard stop" for the level 0 binding13:46
TheJuliaSo, y'all can extend trust that a plugin author is doing the needful, or you can rip your plugin interface out if you want to expect all drivers to always be successful because thats not operational reality13:46
slaweqand allow it only in other levels13:46
TheJuliaAnd I'm sorry if that is coming off as harsh, but I'm sort of a bit frustrated with having an interface, and then second guessing what a plugin knows/can verify13:47
TheJuliasince its a plugin13:47
ralonsohslaweq, but level 0 can work but not the upper one. This is why we need a mechanism to revert the binding13:47
ralonsohTheJulia, I think this is doable but we need to document it well and that will leave on the driver shoulders the responsibility of knowing when the binding should hard stop13:48
slaweqisn't it that it starts with level=0 always and then move on only for HPB? In case of "normal" ports where there is only one level of binding, it would always be level=0, no?13:48
TheJuliaTo be super blunt, I've also not had time to revisit the original patch, because I've been burried under high priority items.13:48
TheJuliain theory, you only find out say level 1 or 2 (if there was a level 2) has a failure if the driver is in the weeds13:49
TheJuliaat which point, either allow port deletion/unbind to unwind level 0, or we handle the flow and unwind in code13:50
TheJuliaThe latter being way more complicated, but also you need some of that to understand the root failure too13:50
TheJuliaso, in effect, unwind is also erasing context unless logging is clear enough13:50
TheJuliaeither way is navigatable13:50
TheJuliaralonsoh: totally fair to document, just humans only have so much capacity and focus at any given moment.13:51
ralonsohslaweq, so far, only Ironic is using the HPB. And it will be Ironic the only driver using this new "hard stop" plug, for now13:52
ralonsohwe always start on level=0, yes. And for in-tree drivers is the only one used13:52
cardoeI'd happy to pick up the work.13:54
cardoeI'd just like to understand what the next steps13:54
lajoskatonathanks, it is a complicated part of the networking :-)13:54
ralonsohI would first implement the exception in neutron-lib, to be shared with other projects13:55
ralonsohand then I would continue with the Neutron patch, considering carefully the unwinding part in case of HPB error in upper levels13:56
TheJuliaralonsoh: fair, we'll likely want some additional logging there in that case as well, but nothing horrible.13:56
slaweqmakes sense, I am fine with it13:57
slaweqand I have now slightly better understanding of the issue :)13:57
TheJuliacool!13:57
ralonsohok, I think we can jump to the next topic13:58
ralonsoh#link https://lists.openstack.org/archives/list/openstack-discuss@lists.openstack.org/thread/GIDECROTNN2SOF3QXIDKYNJNLDZJRKV6/13:58
ralonsohcardoe, please13:58
cardoeThe next one is hairy as well.13:58
opendevreviewMerged openstack/neutron stable/2026.1: Fix deletion of floating IPs with managed DNS records  https://review.opendev.org/c/openstack/neutron/+/98774213:58
cardoeSo we need multiple pools of VNIs for VXLAN13:58
cardoeWe've seen that with Jakub and Helen's spec and implementation, they're making a pool of VNIs.13:59
ralonsohfor evpn, yes13:59
ralonsohthere is a provisional implementation of the DB layer14:00
cardoeWhich they've already said they'll need more pools in the future.14:00
cardoeAnd it's not expandable.14:00
ralonsohsorry, what does it mean?14:00
slaweqbut what they are doing is one pool which will be allocated for routers now and potentially for networks in future, it is not multiple pools AFAIU14:00
cardoeToday Neutron assumes that VNIs span across the entire Neutron install.14:01
cardoeBut we've had users as recently as this week asking how to identify when that's not the case.14:01
cardoeThe VXLAN-EVPN spec is to add VNI pools for EVPN and other pools.14:02
slaweqso you want something like vlan ids per provider network but with VNIs for vxlan networks. So that same VNI can be used by networks using different physical network, is my understanding correct?14:03
TheJuliamy $0.02 is physical networks, because there is the network neutron presents to VM workloads, but there are the networks which are attached, and the needful is to begin to pull the lower level to be able to support >1 VNI which si the same number because distributed and disjointed networks are just totally a thing in the real world.14:04
TheJuliaThe only way to better bridge to that is to be able to model that14:05
cardoeSo far what I've been modeling has been that VXLAN can use physical_network but different than VLAN.14:05
cardoeIt can have "None" or it can have a string.14:05
cardoeThe current behavior of VXLAN is an overlay that touches everything is the physical_network=None case.14:06
cardoeBut if I ask a provider come along and create VXLAN VNI 20000 on physical_network=foo, that's a DIFFERENT pool / fabric / (insert name of your choice) than physical_network = None.14:06
ralonsohbut your proposal doesn't clash with https://review.opendev.org/c/openstack/neutron/+/98725014:07
cardoeralonsoh: no. making a separate allocation pool implementation that's a copy and paste of the VxlanAllocation class with a sed to the name and the table in the DB is silly.14:08
ralonsohI mean, your proposal is something different. The EVPN feature that is implemented ^^ does not map VXLAN neutron networks with physical networks14:08
cardoeYou're creating another VXLAN fabric in their spec.14:09
cardoeThat's separate from the current "overlay network" as you've called it that OVN operates with.14:09
TheJuliacardoe: to explicitly keep it so separate, modeled purely around point to point connections, is that largely where your saying its going in the wrong path?14:10
slaweqcardoe that makes sense for me what you are saying except part with `physical_network = None` - I'm not sure how that should work really in practice14:10
TheJulia^^^ comes to mind, specifically because modeling it purely as point-to-point VNI allocations largely prevents effective type-2 integration with any level of integration beyond basic IP traffic flow.14:10
TheJuliain a point to point context14:10
cardoeTheJulia: yeah. It's a waste of a DB table and duplicated code that could very easily be generic and cover that use case and others.14:10
cardoeslaweq: sure I was just trying to maintain compat.14:11
TheJuliaYeah, its heading down a path to build a duplicate hyperfocused table14:11
cardoeToday we don't allow physical_network=None in our neutron.14:11
cardoeJakub and Helen have already said they're gonna need to come back and add type-2 support and that's a different fabric / pool / (pick your name here).14:12
TheJuliaAnd to do type-2 right, it really needs to be integrated14:12
cardoeSo that means they're gonna have to copy and paste https://review.opendev.org/c/openstack/neutron/+/987250 and create yet another DB table for type-214:12
cardoeBut all these implementations are only for OVN.14:12
ralonsohwell, the spec was very clear on the type-5 support only14:13
cardoeWe're using the teleco model with router flavors today.14:13
cardoeralonsoh: it was and it said type-2 will come afterwards.14:13
TheJuliaSo, to reframe this perhaps, I think cardoe is advocating taking the generic path now as opposed to tool down a hyper-specific path which will create technical debt14:13
cardoe^^^^^^^14:13
ralonsohthat makes sense...14:13
cardoeSo what Jakub and Helen are doing today... we've got that implemented today but not for OVN14:14
TheJuliaand that seems like the worst migration imaginable because given how deep it is, it will likely need to be done in code dynamically which... is not a habit openstack has been good at in general14:14
cardoeBut with a custom router flavor for the specific switch hardware we're using14:14
ralonsohok, cardoe, do me a favor: send a mail to the opendiscuss mail14:14
ralonsohI'll make Helen and Jakub to read it, for sure14:14
ralonsohwe can rethink the DB layer now in order to support type-2 too14:14
cardoeralonsoh: what I'm also asking for is to let other router flavors use their own pools14:15
cardoeCause my router's type-5 pool of VNIs is gonna be different than OVN's type-5 pool.14:15
ralonsohwell, you can have you own flavor and DB definitions in your repo14:16
TheJuliaFWIW, adding it as a column difference would also make it super easy for plugins to support in the existing model around hierarchical port binding where the type-2 route is facilitating a layer-2 bridge connection as opposed to routing just traffic over a tunnel.14:16
cardoeralonsoh: I don't need it to be though.14:16
cardoeCause it's literally just a pool of VNIs.14:16
cardoeThe approach that the type-5 spec takes is also awkward for the user.14:17
cardoeBecause they've said they don't wanna implement automatic VNI allocation. But that they want to focus on the user supplying the VNI to the openstack router create call.14:17
ralonsohplease, comment in the DB definition 9https://review.opendev.org/c/openstack/neutron/+/987250/4/neutron/db/migration/alembic_migrations/versions/2026.2/expand/8c7db84c96c8_add_l3_evpn_tables.py) how it should be to be a generic implementation and usable in other drivers14:17
cardoeBut then their range of VNIs is hardcoded in the neutron.conf14:18
ralonsohyes, it will be the user14:18
cardoeThere's no API call to fetch what the valid VNI range is.14:18
cardoeMy proposed implementation uses the network_segment_range extension.14:18
cardoeAnd you create type=vxlan, physical_network=<your-pool-name-here>, minimum=FOO, maximum=BAR14:19
cardoeAnd your driver uses that network_segment_range with that physical_network value.14:19
cardoeThat's what the first design proposal in my email was.14:19
TheJuliaI still think there would need the ability for a creator to supply a vni though, specifically as an example, say I have a whole fabric of hosts in my garage, and I want to connect them to my office for a type-5 connetion, the vni on the remote side I'll want to map to will be constrained by the far side14:20
cardoe100%14:20
TheJuliabut, that should ideally fall into a scope which can be allocated from14:20
cardoeBut now I can query the network_segment_range API to know what's the valid range14:20
TheJuliaeven if its 1-16 million14:20
cardoeand when I do openstack router create --vni 20000 foobar, it automatically marks 20000 as allocated14:21
cardoeSo that nobody else can create that same VNI again14:21
ralonsohwe had a conversation yesterday about this, that is what is expected14:22
ralonsohbut, of course, that means we are missing the physical network14:22
cardoeralonsoh: I think when we spoke before, you had said the current VXLAN implementation assumes 1 VRF?14:22
cardoeThat's probably the best way to model this.14:22
cardoeEach VRF is its own pool of VNIs.14:23
TheJuliaI mean, so, an interesting thing which might even be a needful is if you have, for example east/west style cloud (really, different norths) connections where you want VNI traffic to flow a particular way to reach those and the easiest way is in a sense, declare a different network entirely which can service the flow14:23
cardoeAnd physical_network on the VNI could just identify the VRF its a part of14:23
cardoeAgain, I'm trying to design this in the most generic way possible and compatible way.14:23
TheJuliacardoe: ++, I was typing something very similar.14:23
ralonsohcardoe, understood. I think your proposal actually collides with the current DB definition in patch 98725014:24
cardoeralonsoh: cause Helen and Jakub's proposal for type-5 is running things through a separate VRF technically14:24
cardoeralonsoh: yep. because they're making another DB table for just 1 more VRF14:24
cardoeThey're gonna have another VRF for type-214:24
ralonsohyes, each network in each compute node will create an independent VFR (if I'm not wrong)14:24
ralonsohplease, write a mail and send it asap to the opendiscuss list14:25
ralonsohwe need to address this now that we are implementing the code14:25
ralonsohor comment in the patch 98725014:25
cardoeokay separate email or replying to my big long email? :-D14:25
ralonsohany of them, I'll ask Helen and Jakub to read it and consider it14:26
TheJuliaThe whole concern I have with adding yet another VRF for type-2 is that it actually feels wrong because type-2's use model and case is entirely different where your not actually routing traffic.14:26
TheJuliaIn my mind, it feels cleaner to be able to model and bridge directly where the needful needs to be instead of using a whole vrf style config for traffic to just be bridged14:27
ralonsohsorry folks, I know we started the meeting late but is even later now14:28
ralonsohso we need to finish the Neutron Drivers meeting now14:28
cardoeThanks for taking the time to design withme14:28
cardoeI'll send those emails.14:28
ralonsohcardoe, please, send this mail and also comment on the patch14:29
lajoskatona+114:29
cardoewill do14:29
ralonsohthanks everyone for attending ( TheJulia thanks for joining)14:29
TheJuliaThanks folks!14:29
ralonsoh#endmeeting14:29
opendevmeetMeeting ended Fri May  8 14:29:38 2026 UTC.  Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4)14:29
opendevmeetMinutes:        https://meetings.opendev.org/meetings/neutron_drivers/2026/neutron_drivers.2026-05-08-13.19.html14:29
opendevmeetMinutes (text): https://meetings.opendev.org/meetings/neutron_drivers/2026/neutron_drivers.2026-05-08-13.19.txt14:29
opendevmeetLog:            https://meetings.opendev.org/meetings/neutron_drivers/2026/neutron_drivers.2026-05-08-13.19.log.html14:29
* TheJulia wonders off back to the land of crypto controls14:29
lajoskatonao/14:29
slaweqo/14:29
cardoeI'll just mention I submitted something like... https://review.opendev.org/c/openstack/neutron-lib/+/98776914:31
opendevreviewMerged openstack/neutron stable/2025.2: Fix deletion of floating IPs with managed DNS records  https://review.opendev.org/c/openstack/neutron/+/98774314:31
cardoeI've also got some other patches to Neutron for the tests to make the FakeTypeDriver actually implement ML2TypeDriver14:31
cardoeCause right now it doesn't and the methods it implements have slightly different arguments.14:32
cardoeMy only guess is that it doesn't get hit due to other mocking? I'm not really sure.14:32
ralonsohthis kind of change can be something merged independently, an improvement of the current code14:33
cardoeJust not sure if adding typing is wanted?14:34
ralonsohit is expected but usually not used14:39
ralonsohif you add it, better14:39
opendevreviewRodolfo Alonso proposed openstack/neutron master: Use Neutron FIP UUID as OVN ``NAT`` row UUID  https://review.opendev.org/c/openstack/neutron/+/98440814:58
opendevreviewMerged openstack/neutron stable/2025.1: Fix deletion of floating IPs with managed DNS records  https://review.opendev.org/c/openstack/neutron/+/98774414:59
opendevreviewMerged openstack/neutron master: Add --port option to remove-duplicated-port-bindings script  https://review.opendev.org/c/openstack/neutron/+/98262214:59
opendevreviewRodolfo Alonso proposed openstack/neutron-lib master: Add security-groups-default-statefulness API definition  https://review.opendev.org/c/openstack/neutron-lib/+/98435415:04
opendevreviewRodolfo Alonso proposed openstack/neutron master: Add security-groups-default-statefulness API extension  https://review.opendev.org/c/openstack/neutron/+/98435615:19
*** haleyb|out is now known as haleyb15:23
marbindrakono/ folks. I'm playing with a proof of concept for providing operator-defined local services (i.e. NTP, DNS, Windows KMS, Package Repos) within isolated tenant networks (https://github.com/marbindrakon/neutron-local-services-poc). My thinking is to keep this out of tree assuming I can get localport binding figured out. Would this ultimately be better off in-tree given the mech driver dependency for OVN? 15:37
cardoeralonsoh: okay. If its cool I'll share some typing patches and you can tell me if that style is good?15:41
cardoemarbindrakon: absolutely think that's great.15:43
cardoeI can tell you in a previous hat on, I worked on that same stuff.15:43
cardoeThat's what Rackspace's "quark" driver does.15:44
marbindrakonI've previously met this use case with Contrail's local services so I've cribbed the same use model. The PoC does have to masquerade as octavia though because of some hard-coding in the mech driver15:45
cardoeThat was their OVS implementation before there was an OVS driver in neutron.15:45
cardoeI just don't work in the VM space anymore.15:45
cardoeI would have stopped working on that around Mitaka.15:46
TheJuliamarbindrakon: neat! One thing to keep in mind is in-tree is likely going to be much bigger hurdle when optional or easy to bolt on in a separate repo may be easier. For example, ironic has an ml2 plugin which can be loaded to trigger localnet ports, it does require ovn, but really that is really a detail17:18
marbindrakonTheJulia: Thanks! I'll check out what Ironic is doing there.17:30
TheJuliamarbindrakon: fwiw, its in the networking-baremetal repository17:30
winiciusallan[m]ralonsoh: hi! we've talked a last month iirc about cutting a new release of n-lib and bumping it in neutron for stable branches (25.2 and 25.1). is it still planned to happen?19:15
winiciusallan[m]I'm needing this https://review.opendev.org/c/openstack/neutron-lib/+/976808 in these branches for a downstream fix19:15
opendevreviewBrian Haley proposed openstack/neutron master: Upgrade to tox 4.28.0  https://review.opendev.org/c/openstack/neutron/+/98551219:15
opendevreviewMerged openstack/neutron-fwaas-dashboard master: Update Babel configuration  https://review.opendev.org/c/openstack/neutron-fwaas-dashboard/+/98748719:27
opendevreviewMerged openstack/neutron-vpnaas-dashboard master: Update Babel configuration  https://review.opendev.org/c/openstack/neutron-vpnaas-dashboard/+/98747219:49
opendevreviewBrian Haley proposed openstack/neutron master: WIP: Add project_id attributes for neutron-lib change  https://review.opendev.org/c/openstack/neutron/+/98696221:13
opendevreviewBrian Haley proposed openstack/neutron-tempest-plugin master: DNM: test hack for tenant_id changes  https://review.opendev.org/c/openstack/neutron-tempest-plugin/+/98795922:39
opendevreviewBrian Haley proposed openstack/neutron master: WIP: Add project_id attributes for neutron-lib change  https://review.opendev.org/c/openstack/neutron/+/98696222:40
*** kinrui is now known as fungi22:44

Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!