Hi Stefan, We run the 18.2.7 version. I turned on the debug logs and this looks really weird: [DBG] public networks ['PREFIX:22::/64'] [DBG] cluster networks [] [DBG] processing data from 12 hosts [DBG] checking public network membership for: ['host-20', 'host-13', 'host-19', 'host-18', 'host-12', 'host-16', 'host-14', 'host-15', 'host-22', 'host-17', 'host-23', 'host-21'] [DBG] checking network PREFIX:22::/64 [DBG] subnet data - {"subnet": "PREFIX:22::/64", "mtu_map": {"9100": ["host-13", "host-12", "host-16", "host-14", "host-15", "host-17", "host-21"]}, "speed_map": {"20000": ["host-13", "host-12", "host-16", "host-14", "host-17", "host-21"], "10000": ["host-15"]}} [DBG] processing mtu map : {"9100": ["host-13", "host-12", "host-16", "host-14", "host-15", "host-17", "host-21"]} [DBG] MTU problems detected [DBG] most hosts using 9100 [DBG] processing subnet : {"subnet": "PREFIX:22::/64", "mtu_map": {"9100": ["host-13", "host-12", "host-16", "host-14", "host-15", "host-17", "host-21"]}, "speed_map": {"20000": ["host-13", "host-12", "host-16", "host-14", "host-17", "host-21"], "10000": ["host-15"]}} [DBG] linkspeed issue(s) detected [DBG] most hosts using 10000 But when I check one of the hosts (host-22). that do not show up in the subnet or mtu map and run a "cephadm --image quay.io/ceph/ceph@sha256:1b9158ce28975f95def6a0ad459fa19f1336506074267a4b47c1bd914a00fec0 gather-facts" I get the following for the interface which should have the prefix "bond0.22": { "driver": "", "iftype": "logical", "ipv4_address": "", "ipv6_address": "fe80::.../64", "lower_devs_list": [ "bond0" ], "mtu": 9100, "nic_type": "ethernet", "operstate": "up", "speed": 20000, "upper_devs_list": [] } But doing a "cephadm --image quay.io/ceph/ceph@sha256:1b9158ce28975f95def6a0ad459fa19f1336506074267a4b47c1bd914a00fec0 list-networks" I get ..., "PREFIX:22::/64": { "bond0.22": [ "PREFIX:22::186" ] }, ..., "fe80::/64": { ..., "bond0.22": [ "fe80::..." ], ... } } So why doesn't it show up in the 9100 mtu_map and in the 20000 speed_map? And why does it use the link local address in the gather-facts section? I cross checked with some other host that acutally shows up and in the gather-facts it got the correct prefix, but the list-networks look the same with the fe80::/64 network. Is there a way to priortize the configured IP addresses in the output and only use the link-local addresses when there is no other IP address? I really wouldn't like to disable the link-local address, because I want to have the ceph-nodes in different clusters configured as same as possible. We have some clusters that use RA on the mgmt interface and some that have a static gateway configured. Am Mi., 17. Sept. 2025 um 09:06 Uhr schrieb Stefan Kooman <stefan@bit.nl>:
On 9/16/25 18:34, Boris wrote:
Hi,
I am currently debugging an issue with the ceph config checks. We have some random hosts that alert
"HOSTNAME does not have an interface on any public network"
but they have. It is IPv6, static configured and, because we don't have a cluster_network, OSDs are bound to that IP in the specific network.
I went through the netplan config and they are basically the same on all hosts. And after rebooting the hosts some of them resolved and some didn't.
How can I dig deeper to figure out what is going on.
Can you find some log output related to these events (docu here [1])?
All services are ceph-orch podman containers All hosts are ubuntu 22.04 with latest HWE kernel (6.8.0-79-generic)
What version of Ceph are you running? We have ubuntu 22.04 clusters configured exactly like this (IPv6 only), so I'm really curious. We haven't seen this behavior yet (18.2.4).
Gr. Stefan
[1]:
https://docs.ceph.com/en/latest/cephadm/operations/#watching-cephadm-log-mes...
-- Die Selbsthilfegruppe "UTF-8-Probleme" trifft sich diesmal abweichend im groüen Saal.