3 Unicorn performance is generally as good as a (mostly) Ruby web server
4 can provide. Most often the performance bottleneck is in the web
5 application running on Unicorn rather than Unicorn itself.
7 == Unicorn Configuration
9 See Unicorn::Configurator for details on the config file format.
11 * Setting a very low value for the :backlog parameter in "listen"
12 directives can allow failover to happen more quickly if your
13 cluster is configured for it.
15 * :rcvbuf and :sndbuf parameters generally do not need to be set for TCP
16 listeners under Linux 2.6 because auto-tuning is enabled. UNIX domain
17 sockets do not have auto-tuning buffer sizes; so increasing those will
18 allow syscalls and task switches to be saved for larger requests
21 * Setting "preload_app true" can allow copy-on-write-friendly GC to
22 be used to save memory. It will probably not work out of the box with
23 applications that open sockets or perform random I/O on files.
24 Databases like TokyoCabinet use concurrency-safe pread()/pwrite()
25 functions for safe sharing of database file descriptors across
28 * On POSIX-compliant filesystems, it is safe for multiple threads or
29 processes to append to one log file as long as all the processes are
30 have them unbuffered (File#sync = true) or they are
31 record(line)-buffered in userspace.
33 * worker_processes should be scaled to the number of processes your
34 backend system(s) can support. DO NOT scale it to the number of
35 external network clients your application expects to be serving.
36 Unicorn is NOT for serving slow clients, that is the job of nginx.
38 == Kernel Parameters (Linux sysctl)
40 WARNING: Do not change system parameters unless you know what you're doing!
42 * net.core.rmem_max and net.core.wmem_max can increase the allowed
43 size of :rcvbuf and :sndbuf respectively. This is mostly only useful
44 for UNIX domain sockets which do not have auto-tuning buffer sizes.
46 * For load testing/benchmarking with UNIX domain sockets, you should
47 consider increasing net.core.somaxconn or else nginx will start
48 failing to connect under heavy load.
50 * If you're running out of local ports, consider lowering
51 net.ipv4.tcp_fin_timeout to 20-30 (default: 60 seconds). Also
52 consider widening the usable port range by changing
53 net.ipv4.ip_local_port_range.
55 * Setting net.ipv4.tcp_timestamps=1 will also allow setting
56 net.ipv4.tcp_tw_reuse=1 and net.ipv4.tcp_tw_recycle=1, which along
57 with the above settings can slow down port exhaustion. Not all
58 networks are compatible with these settings, check with your friendly
59 network administrator before changing these.
61 * Increasing the MTU size can reduce framing overhead for larger
62 transfers. One often-overlooked detail is that the loopback
63 device (usually "lo") can have its MTU increased, too.