[question] high availability support? about headscale

[SuperPauly]

Did anyone eventually work on this? Or is there another HA solution for Headscale?

[SilverBut]

Not sure about if the database transactions are well managed. If so, HA might be simple within same region: just use a shared MySQL database, if the service itself is not stateful.

But I think this is not what you mean (and also not mine), since deploying service in different regions is still not possible (for example, two servers running in Russia and Japan). Maybe we should consider either support some distributed database (like TiDB) so state can be synced via database, or we should use something like raft or paxos to build a cluster.

[juanfont]

We need to think a bit about it. It is not trivial with the current architecture, as a TCP connection is opened and kept from the clients to the server.

This connection is used for keepalives and sending network map updates to the client. Should we have more than one server instance, we would need a mechanism to have cross-headscale communication to notify the peers polling in different instances - which requires some changes in our side.

On the other hand, having the control server down is not great, but not immediately terrible. Everything keeps working, but slowly decaying (Tailscale.com has a KB article on this https://tailscale.com/kb/1091/what-happens-if-the-coordination-server-is-down/)

New users and devices cannot be added to the network.
Keys cannot be refreshed and exchanged, meaning that existing devices will gradually lose access to each other.
Firewall rules cannot be updated.
Existing users cannot have their keys revoked.

Hope this helps...

[jsiebens]

hi @juanfont, that clarifies a lot, thanks for the feedback!