Sunday, 2015-05-31

*** roeyc has joined #openstack-neutron-ovn08:04
*** roeyc has quit IRC09:56
*** roeyc has joined #openstack-neutron-ovn10:28
*** gsagie has quit IRC10:32
*** ajo has joined #openstack-neutron-ovn10:39
*** ajo has quit IRC10:42
*** roeyc has quit IRC11:10
*** roeyc has joined #openstack-neutron-ovn11:32
*** gsagie has joined #openstack-neutron-ovn13:40
gsagierussellb : got the same behaivor as you15:00
gsagiei can ping VM's on private network, but cant do it if i create my own network15:00
gsagiealso the DHCP problem was i was missing /usr/local/bin from rootwrap.conf15:00
rustlebeegsagie: glad you got it figured out15:00
*** rustlebee is now known as russellb15:01
gsagiei checked the L2GW project as well, it seems from first look that the API fits, but we will need to see as the current implementation i think relies on an agent15:02
russellbhopefully implementation is easy to plug in an alternative implementation15:02
gsagiei am afraid thats not the case from what i saw but maybe i am wrong15:02
russellbOVN_Northbound doesn't have anything for this yet, but i expect that's where we'd configure it15:02
russellbwell we'll have to -1 that in code review :-p15:02
gsagiethe gateway is not in the southbound API's ?15:03
russellbi think it's wrong to add anything in neutron that is tied to a specific implementation15:03
russellbthere are gateways in OVN_Southbound15:03
russellbbut Neutron configures things in OVN_Northbound15:03
russellbdon't see how we'd configure them right now15:03
gsagieyeah, but i will have to take a deeper look i saw it relies alot on RPC calls to the agent but maybe i am wrong..15:04
russellbanother thing we need to figure out soon is a proposal for now provider networks should work15:04
russellbthat's next on my priority list after i get tempest in a better state15:05
russellbon another topic, i heard your team might be looking at an ovsdb-server alternative?15:05
gsagiei am trying to understand the flows better in my setup, couldnt see yet how broadcast is sent across the subnet15:06
russellbwhere something else acts as an ovsdb server?15:06
russellboh, i can show you15:06
* russellb pulls up ovs-sandbox15:06
gsagieits a possbility, we do some DB synchronization projects but as far as i know there is nothin concrete right now, why?15:06
gsagiebut thats one area that is probably need alot of work15:07
russellbyeah, i'm just really interested since ovsdb-server is an obvious weak point of the architecture15:07
russellbcentral db that isn't distributed, will be a scale bottleneck (and HA concern)15:08
russellbthis is my simple 2 port setup in ovs-sandbox:
gsagieok, no tunnel yet15:09
russellb cookie=0x0, duration=15.457s, table=17, n_packets=0, n_bytes=0, priority=100,metadata=0x1,dl_dst=01:00:00:00:00:00/01:00:00:00:00:00 actions=set_field:0x2->reg7,resubmit(,18),set_field:0x1->reg7,resubmit(,18)15:09
russellbis the line that handles broadcast15:09
russellbif that bit in the MAC is set, resubmit to table 18 for each port15:09
gsagieyeah that i saw, but couldnt understand the tunneling of broadcast15:09
gsagieto other hypervisor15:10
russellbin my tunnel setup, it looks similar, but i don't have that set up right now15:10
gsagiefrom what it looks they put in tun_id an identifier which is actually the specific VM, the subnet identifier is stored in the metadata right?15:10
gsagiei will look at the code i guess15:11
gsagieand i will keep you updated if i get to start working on the DB thing15:12
gsagiethere is a really interesting project called RAMCloud, which we might be able to leverage for it, you can check it out15:12
russellbRAMCloud sounds familiar, i think that's what i heard about15:13
gsagieyes, i think either me or Eran talked with Chris about it15:13
russellbyep, heard from cdub :)15:14
russellbthis was from my multi hypervisor setup:
russellbso yeah, it sets tun_id to an ID which identifies the specific port15:16
russellband then outputs it to the tunnel to the hypervisor where that port resides15:16
gsagieso i wonder how you exchange the subnet id15:16
gsagiebasically i guess that means no broadcast accross tunnels15:17
russellbwait, maybe that tun_id identifies the subnet15:17
* russellb checks code15:17
gsagiein my setup i saw that its not as i got 2 VM's on the compute node and they got different tun_id15:17
gsagieunless its a bug :)15:17
gsagiejust wondering how the DHCP works15:17
russellbno, you're right, it's a different tun_id per port15:18
russellbthough i wonder if it sends a copy of the packet over the tunnel for every VM on the other hypervisor then15:18
russellbwhich seems pretty wasteful15:18
russellbor every port i mean, they're not all VMs15:19
russellbneed to set this back up i guess15:19
russellbdoesn't look like it's smart enough to only send one copy of the packet :(15:21
russellbwouldn't be the first thing in this setup that needs to be optimized15:21
russellbit also creates tunnels to *every* hypervisor15:21
russellbinstead of only creating them where needed15:23
gsagieok, saw and i understand how DHCP works15:26
gsagiebasically it copies the packet in the broadcast flow for every port in that subnet and resubmit it15:26
gsagieso it will send the packet on the tunnel X amount of times depending on the number of ports in the L2 domain15:28
gsagierussellb : from your blog, where you write the flows:  cookie=0x0, duration=15264.413s, table=17, n_packets=12978, n_bytes=1420946, priority=100,metadata=0x1,dl_dst=01:00:00:00:00:00/01:00:00:00:00:00 actions=set_field:0x6->reg7,resubmit(,18),set_field:0x1->reg7,resubmit(,18),set_field:0x2->reg7,resubmit(,18),set_field:0x4->reg7,resubmit(,18),set_field:0x5->reg7,resubmit(,18)15:29
gsagieit puts the destination port in reg7 and resubmit X number of ports in the network15:29
gsagiebut i guess they will change that soon15:30
gsagiegot to go before my wife gets mad :) talk to you tommorow15:32
gsagieand i will let you know if we start working on the RAMcloud, i am trying to push for it15:33
*** roeyc has quit IRC15:52
*** gsagie_ has joined #openstack-neutron-ovn16:14
gsagie_russellb : here by any chance? you confused me earlier, i thought it was monday :)16:15
*** gsagie_ has quit IRC17:30
*** shettyg has joined #openstack-neutron-ovn22:09
*** shettyg has quit IRC22:22

Generated by 2.14.0 by Marius Gedminas - find it at!