photinus alpha
Fireflies flash in unison. Nobody conducts.
Decentralized mesh monitoring for servers. Every node watches. No node is in charge.
sync 0.00
0/0 lanterns
quorum ok
Click any light to take that server down. The mesh keeps flashing.
- Mesh
- Every server watches every other server. There is no central collector to lose, and no dashboard host you have to keep alive in order to find out that something died.
- Gossip
- Nodes tell their neighbours what they see, and those neighbours tell theirs. State crosses the whole mesh in seconds without anyone coordinating it.
- Quorum
- An alert fires when the mesh agrees, not when one machine has an opinion. A single node losing sight of a host is data. Half of them losing sight of it is an outage.
- Heartbeat
- Some failures are things that did not happen. A backup cron ends by pinging any lantern it can reach; when the pings stop past their window, the mesh agrees your backup is dead and pages you once.
Alpha. Running for real, not packaged yet.
A live fleet has been watching itself with photinus for a while now: tcp, http, TLS certificates, disk, cpu, memory, swap, uptime, network rate, clock skew, heartbeats, and the mesh watching its own members. The source is public and you can build and run it today; there is no installer yet, the config file is landing, and releases still break things. If you run servers and want this, write me and tell me what breaks in your current setup. That is what I am building against.
how the swarm works, with pictures
github.com/christianmeichtry/photinus