This article focuses on application setup for Rack applications, but can be expanded to all applications that connect to external resources and expect short response times.
This article is not specific to Unicorn, but exists to discourage the overuse of the built-in timeout directive in Unicorn.
Network reliability can never be guaranteed. Network failures cannot be detected reliably by the client (Rack application) in a reasonable timeframe, not even on a LAN.
Thus, application authors must configure timeouts when interacting with external resources.
Most database adapters allow configurable timeouts.
Net::HTTP and Net::SMTP in the Ruby standard library allow configurable timeouts.
Even for things as fast as memcached, dalli, memcached and memcache-client RubyGems all offer configurable timeouts.
Consult the relevant documentation for the libraries you use on how to configure these timeouts.
Use non-blocking I/O and IO.select with a timeout to wait on sockets.
Ruby offers a Timeout module in its standard library. It has several caveats and is not always reliable:
/Some/ Ruby C extensions are not interrupted/timed-out gracefully by this module (report these bugs to extension authors, please) but pure-Ruby components should be.
Long-running tasks may run inside `ensure' clauses after timeout fires, causing the timeout to be ineffective.
The Timeout module is a second-to-last-resort solution, timeouts using IO.select (or similar) are more reliable. If you depend on libraries that do not offer timeouts when connecting to external resources, kindly ask those library authors to provide configurable timeouts.
Most operations to regular files on POSIX filesystems are NOT interruptable. Thus, the “timeout” module in the Ruby standard library can not reliably timeout systems with massive amounts of iowait.
If your app relies on the filesystem, ensure all the data your application works with is small enough to fit in the kernel page cache. Otherwise increase the amount of physical memory you have to match, or employ a fast, low-latency storage system (solid state).
Volumes mounted over NFS (and thus a potentially unreliable network) must be mounted with timeouts and applications must be prepared to handle network/server failures.
The timeout mechanism in Unicorn is an extreme solution that should be avoided whenever possible. It will help catch bugs in your application where and when your application forgets to use timeouts, but it is expensive as it kills and respawns a worker process.