Re: help re-configuring bond and ipoib devices/networks

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Some network types use longer or shorter addresses, not all of them try to mimic Ethernet.

For example FireWire uses 64-bit hardware addresses but IP-over-FW extends it to 128-bit addresses in ARP for technical reasons, and I think it's the same for Infiniband and IPoIB.

Unfortunately Networkd doesn't understand any of it.

On Tue, Jul 30, 2024, 04:42 serenissi <serenissi@xxxxxxxxxxxxx> wrote:

I can't tell more about the IPoIB going down after networkd restart without additional debugging info. But from the complains, did you try removing the problematic keys (ipoib is part of netdev, not network. network has no knowledge of the device type)?

Also are you sure 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:6f:85:11 is the mac address? It is clearly more than 48 bits.

On 7/29/24 08:43, Chandler Sobel-Sorenson wrote:

    I'm quite frustrated having spent many hours and little success, when things were perfectly fine before our backup generator decided not to kick in and power surges ensued, messing up all kinds of things stupid electrons.  We have a server that is a bit more important than the others, runs LDAP and hosts our home directories, and runs the subnet manager for an Infiniband network.

    The main uplink is a 2x10gbe bond device, and there's a local 1gbe 10-net interface/network, and an Infiniband with IPoIB.  The main issue is getting the IPoIB working.  The bond and its 2 links work fine after the machine boots, but whenever I restart systemd-networkd, it becomes unreachable even though there aren't any problems reported in the system logs and networkctl reports it's still routable.  The only thing I've figured to get it reachable again is to reboot the system, so I'd like to figure out the problem there because that shouldn't be happening.

    The IPoIB is driving me nuts, I read through all of systemd.network and systemd.netdev docs, and it's somehow become rogue and unmanaged when it used to be configured.  systemd-networkd keeps telling me about unknown keys and unknown sections, even though they were all added in versions previous to the current, which is:

systemd 252 (252.26-1~deb12u2) running on Debian 11 Linux 5.10.0-20-amd64

Below are the configs and messages from systemd.  Hope you can help.

# networkctl
IDX LINK     TYPE       OPERATIONAL SETUP     
  1 lo       loopback   carrier     unmanaged
  2 ens1f0   ether      enslaved    configured
<--10gbe link 1
  3 enp4s0f0 ether      off         unmanaged 
<--not connected
  4 ens1f1   ether      enslaved    configured <--10gbe link 2
  5 enp4s0f1 ether      routable    configured <--1gbe local 10-net
  6 ibs3     infiniband off         unmanaged  <--lazy, rogue
  7 ibs3d1   infiniband off         unmanaged  <--not connected
  8 bond007  bond       routable    configured

Infiniband, IPoIB, ibs3

# cat 10-ibs3.netdev
[Match]

[NetDev]
Name=ibs3
Kind=ipoib
MTUBytes=65520
MACAddress=80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:6f:85:11

[IPoIB]
Mode=connected


# cat ibs3-10_10_11_203.network
[Match]
Name=ibs3
PermanentMACAddress=80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:6f:85:11
Path=/devices/pci0000:00/0000:00:01.0/0000:01:00.0/net/ibs3
Driver=ib_ipoib
Type=infiniband
Kind=ipoib
Property=ID_NET_MANAGED_BY=io.systemd.Network

[Link]
MTUBytes=65520

[Network]
Kind=ipoib
Address=10.10.11.203/24
Gateway=10.10.11.203
LinkLocalAddressing=no
IPv4AcceptLocal=yes
KeepConfiguration=static

[IPoIB]
Mode=connected

systemd-networkd complaints

/etc/systemd/network/10-ibs3.netdev:8: Not a valid hardware address, ignoring assignment: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:6f:85:11
/etc/systemd/network/10-ibs3.netdev:8: Not a valid hardware address, ignoring assignment: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:6f:85:11
/etc/systemd/network/ibs3-10_10_11_203.network:15: Unknown key 'Kind' in section [Network], ignoring.
/etc/systemd/network/ibs3-10_10_11_203.network:22: Unknown section 'IPoIB'. Ignoring.

bond

# cat 10-bond007.netdev
[NetDev]
Name=bond007
Kind=bond
MTUBytes=9000

[Bond]
Mode=802.3ad
MIIMonitorSec=1000
UpDelaySec=1000
DownDelaySec=2000


# cat Intel_X710_DA2-bond007.network
[Match]
Path=pci-0000:05:00.0
Path=pci-0000:05:00.1

[Network]
Bond=bond007


# cat bond007-10_140_78_70.network
[Match]
Name=bond007

[Network]
Address=10.140.78.70/28
Gateway=10.140.78.65
DNS=128.196.11.233
DNS=128.196.11.234
DNS=128.196.11.235
LinkLocalAddressing=no
IPv6AcceptRA=no

Best Regards,
Chandler

The University of Arizona block 'A' logo.
Chandler Sobel-Sorenson
Sr. Systems Administrator
Arizona Genomics Institute
School of Plant Sciences—Research
THE UNIVERSITY OF ARIZONA

Thomas W. Keating Bioresearch Bldg. | Rm. 200A24
1657 E. Helen St. | Tucson, AZ 85721
Office: 520-626-9589 | Cell: 520-907-4352

chandler@xxxxxxxxxxxxxxxxxx
Pronouns: he/him/his
www.genome.arizona.edu

Integrity, Compassion, Exploration, Adaptation, Inclusion, Determination

PNG image

PNG image

Attachment: signature-footer.png
Description: PNG image

Attachment: ua-sig-logo.png
Description: PNG image


[Index of Archives]     [LARTC]     [Bugtraq]     [Yosemite Forum]     [Photo]

  Powered by Linux