The issue here seems to not be the invite server but the droplets themselves. So droplets did not have a swap memory at all. So on some of the nodes it went over its minimal actual memory and thereby shut the vault off only in a handful of proxy nodes. On some of these it also caused the droplet to hang which then prevented the invite server from being able to edit the client whitelist.
Luckily since the network already has churn process integrated, it has allowed us to simply set some swap memory on the nodes and start the vaults on the affected nodes back up. We did not intend to be testing churn on TEST-17 but good thing it works and was tested in an unintentional way I guess.
Invite server should be back online now I think and we’ll keep an eye on the droplets to see if the hanging issue is reproduced later on as that part is still a question as to what caused some nodes to hang.