Improve networking options for libvirtd target

[teto]

Follow up of NixOS/nixops#922.
The use of qemu-agent and some more configurability for libvirt scenarios

I suppose the ipaddress dependency could be dropped of but:

• it's convenient
• and part of python3 batteries anyway (I hope we support python3 at some point ?!)

The only thing I don't understand is why the first deploy fails without:
networking.networkmanager.enable = true;
(nixops ssh seems to work fine, just the nix-copy-closure fails )

[teto]

so nix-copy-closure was stuck at

debug1: Sending command: nix-store --serve --write
debug2: channel 0: request exec confirm 1
debug3: send packet: type 98
debug2: channel_input_open_confirmation: channel 0: callback done
debug2: channel 0: open confirm rwindow 0 rmax 32768
debug2: channel 0: rcvd adjust 2097152
debug3: receive packet: type 99
debug2: channel_input_status_confirm: type 99 id 0
debug2: exec request accepted on channel 0

while opening wireshark, I saw the MTU was at 9000 for the libvirtd bridge. (I think TCP_NODELAY might be set, ) and found the related issue NixOS/nix#1988 . I changed the bridge mtu to 1400 and it fixed it. I wonder if that's a kernel issue or something else.

[teto]

@nh2 @sorkin @mbrgm if you wanna test, it should be ready.

[AmineChikhaoui]

This seems to break existing deployments ?

[nix-shell:~/src/nixops]$ nixops deploy -d libvirt
[<nixopsvirtd.backends.libvirtd.LibvirtdNetwork instance at 0x7f22352a3a28>]
len=1
machine> Connecting to qemu:///system...
machine> connecting...
libvirt: QEMU Driver error : argument unsupported: QEMU guest agent is not configured
machine> argument unsupported: QEMU guest agent is not configured
libvirt: QEMU Driver error : argument unsupported: QEMU guest agent is not configured
machine> argument unsupported: QEMU guest agent is not configured
libvirt: QEMU Driver error : argument unsupported: QEMU guest agent is not configured
machine> argument unsupported: QEMU guest agent is not configured
ssh: Could not resolve hostname none: Name or service not known
libvirt: QEMU Driver error : argument unsupported: QEMU guest agent is not configured
machine> argument unsupported: QEMU guest agent is not configured
machine> could not connect to ‘root@None’, retrying in 1 seconds...
libvirt: QEMU Driver error : argument unsupported: QEMU guest agent is not configured
machine> argument unsupported: QEMU guest agent is not configured
ssh: Could not resolve hostname none: Name or service not known
libvirt: QEMU Driver error : argument unsupported: QEMU guest agent is not configured
machine> argument unsupported: QEMU guest agent is not configured
machine> could not connect to ‘root@None’, retrying in 2 seconds...
^Cerror: interrupted

Also there is debug statements left and many commented lines, can you make sure to remove those ?

[teto]

In order to be backwards compatible, I could check for both values of https://libvirt.org/html/libvirt-libvirt-domain.html#virDomainInterfaceAddressesSource, e.g., to search for IPs seen by the libvirt router or by the agent.
Would you prefer to make it configurable ? (with/without agent), not sure that's worth the effort IMO.

[dsg22]

Overall I like this. I would like to extend the networking configuration more (allowing setting VLAN tags in the interface XML, etc). Adding those changes will be simpler after this is merged.

[dsg22]

I think this needs to be fixed before merge. Is there an upstream libvirt bug about the mismatch in docs and reality mentioned in the comment above?

Explicitly specifying owner, group and SELinux label will break many setups, including mine.

[teto]

it might have become unnecessary with the other PR merged today. I will look into it

[teto]

so if I understand the doc correctly, the pool is created with the permission of the folder but the volums are created by default with the user/group that libvirtd is running as. Seems like the libvirtd service is running as root without any nix option to change it so this looks a bit more complex than I have time for (for now) so feel free to look into it.