"Elf-Disclosure" for June 2023

Here's what we achieved in June 2023 (not an exhaustive list, there's only so much space!):

• [Prod website][elfhosted] based on (you guessed it!) mkdocs-material (look familiar?)
• HA, fault-tolerant K3s Kubernetes cluster (3 servers, 3 agents, 3 ceph nodes)
• Dedicated CI environment for pre-testing infrastructure changes
• BYO storage / VPN fully self-service when purchasing via the store[elfhosted/store]
• All previously supported Seedplicity [apps][elfhosted/apps] available ⁵
• CLI tool (ElfBot[elfhosted/elfbot]) for self-service app restarts, backups, resets

Here's what's on the short-list for prioritization next:

• Bring BYOVPN config to Deluge and ruTorrent
• Add new apps to support Premiumize
• Migrate to prod store, make bundles easier to manage
• Regular daily maintenance period for app updates / maintenance
• Load test with more users!

Most apps consume almost no CPU while idle - the larger consumers are streamers doing transcoding, and download clients doing download/unpack operations:

This graph represents memory usage across the entire cluster. By far the largest consumers of RAM are the storage platforms (longhorn and ceph):

I'm not sure these stats are accurate, they've likely overly high because pods on the host network (like metallb, ceph, etc) will end up counting all traffic on each host, rather than the pod itself. This is an outstanding issue to fix!

These are the traffic stats for egress from Hetzner. They exclude any traffic to/from Hetzner Storageboxes:

Longhorn provides RWX volumes for /config, and for some infrastructure components like Prometheus, Chartmuseum, etc.

Ceph provides optional storage ("ElfStorage"), typically used for long-term slow storage and seeding: