#midonet

hi agabert, I've set this up in neutron with the following commands:

$ neutron net-create ext-net --shared --router:external True

$ neutron subnet-create ext-net --name ext-subnet \

--allocation-pool start=115.146.82.132,end=115.146.82.200 \

--disable-dhcp --gateway 115.146.82.129 115.146.82.128/25

good

aeh

no

thats not right

hold on a second

thats not how routing really works

neutron net-create public --router:external=true

oh, I'm not sure what has been wrong, could you please point it out

you said the floating ip range would be 115.146.82.132 to 115.146.82.200

that range must not fall inside the network 115.146.82.128/25

because if it does its not routing.

to keep it really simple

i would build it this way (you need to reconfigure your 115.146.82.129 router though)

agabert@p2003005C4B7DCDD5021D92FFFEF79964:~$ ipcalc 115.146.82.128/30

Address: 115.146.82.128 01110011.10010010.01010010.100000 00

Netmask: 255.255.255.252 = 30 11111111.11111111.11111111.111111 00

Wildcard: 0.0.0.3 00000000.00000000.00000000.000000 11

=>

Network: 115.146.82.128/30 01110011.10010010.01010010.100000 00

HostMin: 115.146.82.129 01110011.10010010.01010010.100000 01

HostMax: 115.146.82.130 01110011.10010010.01010010.100000 10

Broadcast: 115.146.82.131 01110011.10010010.01010010.100000 11

Hosts/Net: 2 Class A

this is what i would use as transfer network between your router and the midonet gateway

ryu_mido: ping

hey apuimedo!

ryu_mido: how is it going?

a bit tired, but otherwise not too bad, yourself?

bit tired too :P

and wondering if I'll get back home today

since the road next to my town is cut due to "snow"

xD

your 115.146.82.129 will send all traffic destined for 115.146.82.192/26 to 115.146.82.130

where are you?

in the office

in bcn

it's snowing??

Address: 115.146.82.192 01110011.10010010.01010010.11 000000

Netmask: 255.255.255.192 = 26 11111111.11111111.11111111.11 000000

Wildcard: 0.0.0.63 00000000.00000000.00000000.00 111111

=>

Network: 115.146.82.192/26 01110011.10010010.01010010.11 000000

HostMin: 115.146.82.193 01110011.10010010.01010010.11 000001

HostMax: 115.146.82.254 01110011.10010010.01010010.11 111110

Broadcast: 115.146.82.255 01110011.10010010.01010010.11 111111

in barcelona it only snowed for a moment, then rain

hhuang: otherwise its not routing.

hhuang: you see what i mean?

around my town I guess it snowed more

ah ok, hope you get to find yourself home!

your port will be 115.146.82.130 on the eth2.75

;-)

I was checking the nova libvirt vif driver and I was wondering why do we use the ethernet backend config

since it marks the domains as tainted and such

(which means that libvirt devs would get off the hook for support)

hhuang: does that make sense?

:P

yes, actually could you hold on a sec...someone having an issue in the office...be back in a few minutes

ok

you are throwing away almost half of your public ip space but thats something that can be fixed if you would use a separate, small, border network and be able to route the whole /25 to the provider router port

you cannot have the same network in the provider router neutron network and on the physical link side, you would have to use bridging then.

(which is the default model of neutron ovs i suppose, thats why everybody is used to it so much)

that's right, I did need a bridge for ovs

apuimedo, i'm back, yes you're right, libvirt folks discourage the use of 'ethernet' type

but unfortunately we don't really have a choice because other modes do not work for us

ovs modes actually run ovs-brctl command to configure the interface which doesn't work for us

ryu_mido: what would happen if we created the taps on the midonet vswitch?

apuimedo: at what timing would it be created?

ryu_mido: the tap device?

do you mean that midonet agent handles the tap creation?

no

do we have a way to create tap devices in mm-ctl?

no not at the moment

but of course, trivial to do so if we choose to

well, the ideal would be to teach libvirt about how to create a vNIC on midonet

i think the issue is that the libvirt config that gets generated dictates libvirt to configure the tap/bridge on the host and its' quite opinionated

hhuang: would it be possible to use routing?

hhuang: otherwise we need to build you a bridging scenario.

apuimedo, agreed 100%

but that would take a longish time to make it into stable libvirts in distros now :(

probably el7.2

ryu_mido: you could call ovs userland from under midonet agent in java by running exec() and doing funny things to make duarte lose his temper.

if libvirt becomes midonet-aware (which is not something out of the question considering it's open sourced) it would be great

(and ubuntu I don't even know :P )

I'm actually not a networking guy and got very limited details about the networking from my colleague. but I can modify the current networking setting to use your approach

hhuang: do you have access to the gateway?

hhuang: you need to flip the netmask of that interface to /30

ryu_mido: if we were to do it

hhuang: the rest is easy

what would be the ideal thing?

that libvirt would call mm-ctl to create the tap device with a port binding already?

(provided we added mm-ctl support for that, of course)

apuimedo: yes, basically instead of ovs-vsctl, it would do mm-ctl

hhuang: and you need to add a route to the gateway 115.146.82.129: send traffic for 115.146.82.192/26 to 115.146.82.130 please

in fact, libvirt probably creates tap interface for its ovs mode too (though not 100% sure)

so I'll look into it

hhuang: midonet provider router will take it from there if you put this port on the interface

so it could be that libvirt creates tap and calls mm-ctl

and talk to the libvirt people if they would be open to such patchsets

or mm-ctl handles both... i think it does make sense that the hypervisor side handles the tap management

indeed

better let libvirt handle tap life-cycle

and we simply provide the script (mm-ctl) to attach two sides

yup

agabert: I like duarte too much for me to do such thing to him!

it would simplify the migration flow

yeah

what did I do? *scared*

nothing

hhuang: send me an email if you want to sync up, i have to run away now

bberg1:

hell

relevant updates from the importer side? blockers? what did you decide re. L2 support?

Hey, hey! This is a public channel. Say "Heck."

srvaroa that was completely uncalled for

I'm working on reorganizing the code to make sharing code between different translators (which we need) easier. I'm implementing a bunch of *Manager traits, like PortManager, ChainManager, etc.

ryu_mido: what was!

This is for shared code that is used by multiple translators.

oh hell, I didn't wrte that

I wrote hello, I think !

I also investigated a user issue (see above), submitted a small cleanup patch, and reviewed Tomohiko's small cleanup patch.

Regarding L2 stuff, Tomohiko said he'd take the Subnet work.

...I think. This was literally hours ago, so my memory's a bit fuzzy.

Then I spent like 40 minutes trying to get the standup meeting started.

cool

who deals with office infrastructure?

I don't know. Not sure what's up with the conference room computer. I think someone messed with it, since the speaker and camera were unplugged.

ok

well, so the L2 stuff is unblocked then

cool

Also, the windows were open when I came in. There may have been a ninja invasion.

I go next.

Mostly as Brandon reported.

ok, I'll reply to your quesiton in the email thread