Retirement of old discovery servers

General
1

I’m retiring the old global discovery servers that are unused in 0.14.44+. For those still on 0.14.43 and earlier you may see a note in GUI about discovery being “4/8” or something like that. It’s essentially harmless - the new / remaining discovery servers are much faster and better and you’re already using them.

The old system consisted of three separate servers, all of which took each announce and query over both IPv4 and IPv6. In total this looks to the client like six separate servers, and every announce or query was repeated six times.

  • discovery-v4-2, discovery-v6-2 in New York
  • discovery-v4-3, discovery-v6-3 in Singapore
  • discovery-v4-4, discovery-v6-4 in Amsterdam

As each server got the full load of all clients, on both v4 and v6, each was essentially permanently overloaded.

The new system consists of a cluster with internal synchronization and round robin DNS. Announces and queries only need to happen once per cluster instead of once per server. Announces are still made over both IPv4 and IPv6, but a single query covers both. Each of the servers is also more powerful than the previous instances.

  • discovery-1 in Paris
  • discovery-2 in Amsterdam

The cluster presents the same certificate as the old discovery-v[46]-4 instances, and are in fact what old clients already talk to under those names. I am now simply decommissioning the New York and Singapore instances.

7 Likes
2

What kind of impact does having discovery servers exclusively in Europe cause for users in Asia, Australia, NZ, South/North America?

I presume it is fractions of a percent if you are storing multiple gigabytes in a cluster of ~2-10 devices?

3

None at all. Previously all clients announced to all discovery servers, now clients announce to any discovery server and they sync between each other.

The previous implementation potentially increased lookup times, but it’s still negligible in the grand scheme of things.

1 Like