Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
bsdcafe-technical-details [2024/08/05 06:42] – stefanobsdcafe-technical-details [2024/08/26 13:13] (current) – stefano
Line 13: Line 13:
   * Another jail runs KeyDB for the Mastodon cache   * Another jail runs KeyDB for the Mastodon cache
   * Another jail runs PostgreSQL, serving as the database - the memory of BSD Cafe.   * Another jail runs PostgreSQL, serving as the database - the memory of BSD Cafe.
-  * There's a separate jail designated for media storage, acting as the 'multimedia memory' of BSD Cafe. This jail resides on an external server with rotating disks behind Cloudflare. The goal is georeplicated caching of multimedia data to reduce bandwidth consumption.+  * There's a separate jail designated for media storage, acting as the 'multimedia memory' of BSD Cafe. This jail resides on an external server with rotating disks <del>behind Cloudflare</del>. There are [[:bsdcafemedia-reverse-proxies|reverse proxies]] located in various parts of the world that receive and cache requests, serving them from the nearest server. The goal is to use georeplicated caching of multimedia data to reduce bandwidth consumption. The selection of reverse proxies is handled at the DNS level through a PowerDNS LUA script. More details have been documented here: [[https://it-notes.dragas.net/2024/08/26/building-a-self-hosted-cdn-for-bsd-cafe-media/|Building a Self-Hosted CDN for BSD Cafe Media]]
   * Two jails host the Mastodon application itself, including sidekiq, puma, and streaming components. This is where all processing and connection management occur. They're running in different hosts and are both providing some sidekiq queues and one of the two is set as a backup puma and streaming server when the first one (more powerful) is down. This is useful when updating, to avoid downtime.   * Two jails host the Mastodon application itself, including sidekiq, puma, and streaming components. This is where all processing and connection management occur. They're running in different hosts and are both providing some sidekiq queues and one of the two is set as a backup puma and streaming server when the first one (more powerful) is down. This is useful when updating, to avoid downtime.