Wednesday, 2025-09-24

*** vindict_ is now known as vindict01:08
opendevreviewRodolfo Alonso proposed openstack/neutron master: WIP == [OVN] The external networks GW chassis must the same as the GW LRP  https://review.opendev.org/c/openstack/neutron/+/96215507:14
opendevreviewRodolfo Alonso proposed openstack/neutron stable/2024.1: [stable] Pin rally to 4.1.0  https://review.opendev.org/c/openstack/neutron/+/96215707:38
tafkamaxHi I have a question regarding networking in openstack. We have deployed a 2024.2 release cluster and we are seeing that with floating ip we are getting an issue where the ping responds even if the associated VM is in poweroff state.07:38
tafkamaxWhy is this happening?07:38
ralonsohtafkamax, from the documentation07:40
ralonsohIn OVN (Open Virtual Network), the logical router is responsible for responding to a ping (ICMP echo request) directed to one of its IP addresses. This is a built-in feature of OVN's architecture designed to handle these requests efficiently at the logical level.07:40
ralonsohWhen a virtual machine or container sends a ping to an OVN logical router's IP address, the packet doesn't need to be forwarded to a physical router or a dedicated network service VM. Instead, OVN's control plane proactively installs flows in the Open vSwitch (OVS) integration bridge on the hypervisor where the sending VM resides. These flows are specifically designed to recognize and respond to such ICMP requests 07:40
ralonsohlocally.07:40
tafkamaxThanks07:46
tafkamaxCould you please link me the docs reference?07:48
tafkamaxI can't seem to find it when searching myself :S07:48
opendevreviewRodolfo Alonso proposed openstack/neutron stable/2024.1: [stable] Pin rally=4.1.0, rally-openstack=3.0.0  https://review.opendev.org/c/openstack/neutron/+/96215707:55
opendevreviewRodolfo Alonso proposed openstack/neutron master: [eventlet-removal] Don't use eventlet in the unit/functional tests  https://review.opendev.org/c/openstack/neutron/+/95225808:41
ralonsohhello folks, any review is welcome: https://review.opendev.org/c/openstack/neutron-lib/+/96161208:41
zigoHi there! I've just validated Flamingo on my CI, and all worked, but I found out that Neutron is using WAAAYYY more memory than it used to. I found my 2 network nodes doing OOM-kill, being completely dead. On my CI, my network nodes used to run with 8 GB of RAM, I had to increase it (I've set 16 GB).08:43
zigoIs this something known by the team?08:43
ralonsohzigo, with ML2/OVS, right?08:46
zigoYeah.08:46
ralonsohI saw this in our CI too08:47
ralonsohbut only in one patch (maybe because is triggering this job)08:47
ralonsohthis could be related to the eventlet removal08:47
zigoMy thoughts as well.08:47
ralonsohbut for L3/OVS/DHCP agents, the eventlet removal was finished time ago08:47
ralonsohso this should have been happening months ago08:48
ralonsohzigo, I saw this issue here: https://zuul.opendev.org/t/openstack/builds?job_name=neutron-tempest-plugin-openvswitch&project=openstack/neutron08:49
ralonsohbut only in one patch08:50
zigoWhat I saw in my network node, was so many neutron-keepalived-state-change processes, but maybe that's a consequence, not a root cause.08:52
ralonsohyeah, this is a recurrent problem with this process08:53
ralonsohwe need one process per HA router and each one consumes more than 100MB08:53
ralonsohI started, 2 years ago (never finished), to move this code to a bash script08:53
zigoWhat does that process do exactly?08:53
ralonsohnothing: check the keepalived status (main, backup) and write a file08:54
ralonsohbut because we import neutron, oslo, etc, the memory usage is massive08:54
* zigo reads the code08:55
ralonsohzigo, https://review.opendev.org/c/openstack/neutron/+/83614008:55
ralonsohthat goes from 100MB to 7MB08:55
ralonsohSo according to https://zuul.opendev.org/t/openstack/builds?job_name=neutron-tempest-plugin-openvswitch&project=openstack%2Fneutron&change=952258&skip=0 08:56
ralonsoh^^ just this job for this specific change08:56
ralonsohthat started on Sep 2208:56
ralonsohsorry Sep 23, the Sep 22 job passed fine08:57
ralonsohI see nothing in requirements nor Neutron (relevant for this)08:58
zigoOh, so keepalived_state_change is the thing that monitors active/standby thing for HA routers?09:00
ralonsohzigo, yes, this is how Neutron knows the status of each HA instance09:01
ralonsohI know this is trivial and it could be done with a bash script (^^ the upper link)09:01
zigoYeah, indeed ! :P09:01
ralonsohthat could aleviate a lot the memory requirements for OVS envs with many HA routers (1/10 size)09:02
ralonsohbut, to be honest, now I can't spend time on this09:02
ralonsohin any case, I'll add this to the PTG topics, for sure09:02
ralonsohlajoskatona, hello! Do you happen to know if we have a PTG etherpad?09:02
ralonsohahh yes09:03
ralonsohhttps://etherpad.opendev.org/p/oct2025-ptg-neutron09:03
zigoJust checked one of our network node, we have 673 instances of neutron-keepalived-state-change running, so that's 65 GB of RAM !09:04
zigoWith your rewrite, that'd be only 4...09:04
ralonsohexactly (and 4GB just to monitor is still insane...)09:04
ralonsohwe can propose:09:04
zigoYour proposal is also: having a single daemon per L3 agent, monitoring *ALL* HA routers?09:07
zigoSounds good to me. :P09:07
ralonsohI made these 2 proposals: migrate to bash or, with the same code, monitor several instances09:08
ralonsohthe double shot is to migrate to bash (1 monitor per HA router) and then use this single bash script for multiple instances09:08
zigoCould it be both? ie: a bash script that would monitor several HA routers?09:08
ralonsohyes, but step by step09:08
zigo:P09:08
zigoWould be awesome.09:09
ralonsohthe migration could fix many problems initially09:09
zigoThanks for all of these details.09:09
zigoralonsoh: Did you compare dash to bash? Writting in pure sh isn't hard, I could help, and it should also make the footprint a way nicer.09:11
ralonsohzigo, I did: https://review.opendev.org/c/openstack/neutron/+/83614009:12
ralonsoh^^ that was almost finished, but I didn't have more time09:12
zigoThat's full of bashism ! :P09:12
zigo"function" must be removed, double-test (ie: [[ ) must be removed too.09:13
zigoPlus there's no arrays in dash.09:13
ralonsohwell, that is supposed to be a bash script09:15
zigoDash would be 1/3rd of the memory footprint.09:16
ralonsohI think the migration from python to bash is more than enough09:16
zigoProbably ! :)09:16
ralonsohwe can continue with this script I proposed, I think09:16
*** dmellado7 is now known as dmellado09:23
opendevreviewRodolfo Alonso proposed openstack/neutron master: [OVN] Remove the check for external ports support  https://review.opendev.org/c/openstack/neutron/+/96181010:32
opendevreviewMartin Kalcok proposed openstack/neutron master: ovn_db_sync: Improve coexistence support.  https://review.opendev.org/c/openstack/neutron/+/95867911:04
opendevreviewFernando Royo proposed openstack/ovn-octavia-provider master: Optimize _get_current_operating_statuses to reduce octavia_lib calls  https://review.opendev.org/c/openstack/ovn-octavia-provider/+/96217911:49
opendevreviewMerged openstack/neutron stable/2024.1: [stable] Pin rally=4.1.0, rally-openstack=3.0.0  https://review.opendev.org/c/openstack/neutron/+/96215711:56
*** ykarel_ is now known as ykarel12:35
opendevreviewyatin proposed openstack/neutron master: [DNM] check fips-mode-setup  https://review.opendev.org/c/openstack/neutron/+/96120613:11
opendevreviewFernando Royo proposed openstack/ovn-octavia-provider master: Optimize _get_current_operating_statuses to reduce octavia_lib calls  https://review.opendev.org/c/openstack/ovn-octavia-provider/+/96217913:43
opendevreviewEduardo Olivares proposed openstack/neutron master: WIP: new job tempest-multinode-with-bgp  https://review.opendev.org/c/openstack/neutron/+/96218813:47
opendevreviewEduardo Olivares proposed openstack/neutron master: WIP: new job tempest-multinode-with-bgp  https://review.opendev.org/c/openstack/neutron/+/96218813:51
opendevreviewEduardo Olivares proposed openstack/neutron master: WIP: new job tempest-multinode-with-bgp  https://review.opendev.org/c/openstack/neutron/+/96218813:53
opendevreviewEduardo Olivares proposed openstack/neutron master: WIP: new job tempest-multinode-with-bgp  https://review.opendev.org/c/openstack/neutron/+/96218813:54
*** masayukig_ is now known as masayukig14:20
*** zseguin_ is now known as zseguin14:26
*** mnasiadka_ is now known as mnasiadka14:26
*** TheJulia_ is now known as TheJulia14:26
opendevreviewRodolfo Alonso proposed openstack/neutron master: [OVN] Always exclude the local chassis creating ``HA_Chassis_Group``  https://review.opendev.org/c/openstack/neutron/+/96219114:33
*** masayukig_ is now known as masayukig14:39
*** johnsom_ is now known as johnsom14:39
*** mnaser_ is now known as mnaser14:39
*** haleyb_ is now known as haleyb14:39
*** haleyb_ is now known as haleyb14:42
opendevreviewSlawek Kaplonski proposed openstack/neutron master: [S-RBAC] Fix tests for port RBAC policies  https://review.opendev.org/c/openstack/neutron/+/96219214:50
ralonsohhaleyb, hi! no rush: https://review.opendev.org/c/openstack/neutron-lib/+/961612. Just if you have a couple of mins14:54
*** haleyb_ is now known as haleyb15:05
*** cardoe_ is now known as cardoe15:32
*** gthiemon1e is now known as gthiemonge16:01
opendevreviewEduardo Olivares proposed openstack/neutron master: WIP: new job tempest-multinode-with-bgp  https://review.opendev.org/c/openstack/neutron/+/96218816:04
opendevreviewEduardo Olivares proposed openstack/neutron master: WIP: new job tempest-multinode-with-bgp  https://review.opendev.org/c/openstack/neutron/+/96218816:06
*** iurygregory_ is now known as iurygregory23:41

Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!