gistfile1.md

In this scenario, there is a single pool of combined workers/content servers, and each of these machines is authoritative for a given set of sites (probably determined by some sort of consistent hash). A dynamic reverse proxy is responsible for dispatching each incoming request to the correct worker machine based on the request hostname/path.

Pros

• simple
• exactly one machine is authoritative for each site, so no need for distributed locking (in-process lock is sufficient)

Cons

• migration could be tricky; need to co-ordinate update of dynamic proxy's routing table with migration of content between workers
• possible resource wastage - system anticipated to be read-heavy, yet we're running a worker pool on each machine

Pros

• worker pool and content server pool can be scaled independently
• no migration issues - all content servers capable of serving from the entire shared FS

Cons

• complicated
• need measures in place to deal with dangling locks
• it's quite complicated
• did I mention it's complicated