A network profile quietly ignored its own settings

While rolling the isolated cluster network off ordinary switch ports, one machine lost its primary cluster link.

What happened. The replacement network profile listed the cluster network among the ones it was supposed to carry, but a single setting on that profile meant the switch honoured only the untagged network and ignored that list entirely. The profile was therefore identical to the plain access profile, and silently dropped the traffic it was written to carry.

Why it took a while to surface. The profile had already been applied to three switches without incident — but only ever to ports that had no cluster traffic on them, so nothing had ever exercised it. The first port that genuinely needed it was the one that failed.

Impact. None to any service. The cluster runs a second, independent link for exactly this case, and it carried the cluster throughout; the machine kept working and no workload moved or stopped. The fault was diagnosed within minutes because two machines on the same switch, still using the old profile, kept working — which isolated the profile as the only variable.

Fix. One setting on the profile. The link returned immediately, and the rollout then completed across all five switches.