uwsgi¶

uwsgi 是一种协议， uWSGI是一个纯c写的web容器。 uWSGI实现了wsgi协议和uwsgi协议, 并且它的速度很快. uwsgi 默认是在当前python 下编译的，如果你有多个python版本环境，有两种方式可以解决:

在每个python下编译安装各自的uwsgi
编译uwsgi用 Bonus: multiple Python versions for the same uWSGI binary, 不绑定uwsgi到具体的python版本

nginx 和 uwsgi 应用交互¶

在nginx中配置uwsgi时，会有这么一行:

include     uwsgi_params;

这里面其实定义了很多变量:

uwsgi_param  QUERY_STRING       $query_string;
uwsgi_param  REQUEST_METHOD     $request_method;
uwsgi_param  CONTENT_TYPE       $content_type;
uwsgi_param  CONTENT_LENGTH     $content_length;
uwsgi_param  REQUEST_URI        $request_uri;
uwsgi_param  PATH_INFO          $document_uri;
uwsgi_param  DOCUMENT_ROOT      $document_root;
uwsgi_param  SERVER_PROTOCOL    $server_protocol;
uwsgi_param  HTTPS              $https if_not_empty;
uwsgi_param  REMOTE_ADDR        $remote_addr;
uwsgi_param  REMOTE_PORT        $remote_port;

也许就是利用环境变量传递这些值给uwsgi 应用的， uwsgi 会把这些变量转换成http header, 放在request里, 所以: 如果有些值， nginx 层有，但是django层看不到，那么可以利用nginx 的uwsgi_param指令，把值传到django里

uwsgi插件¶

uwsgi 默认编译了很多插件在里面，可以用 uwsgi --plugins-list 命令来查看有哪些可以使用的插件

route-cache¶

conf:

[uwsgi]
chdir = .
# plugins = router_cache
module = mysite.wsgi:application
master = True
processes = 1
stats = /tmp/test_stat.sock
http = :9090
harakiri = 8

cpu_affinity = 1
log-micros = true
cache2 = name=mycache,items=300,hash=murmur2
route = ^/feedback cache:key=${REQUEST_METHOD}${REQUEST_URI}${QUERY_STRING},name=mycache,content_type=application/json
route = ^/feedback cachestore:key=${REQUEST_METHOD}${REQUEST_URI}${QUERY_STRING},name=mycache,expires=10
route = ^/feedback log:hey i am printing $1


# route = ^/$ cache:key=myhome_for_${cookie[PHPSESSID]},name=mycache
# ; store each successfull request (200 http status code) for '/'
# route = ^/$ cachestore:key=myhome_for_${cookie[PHPSESSID]},name=mycache
autoload = true

alarm¶

uwsgi 可以设置报警，就是if then, example:

alarm-backlog = mailme cmd:mail -s 'uWSGI alarm' -a 'From: foobar@example.com' admin@example.com

有用的alarm条件有这几个:

log-alarm: 用regex监控log
larm-backlog: 如果backlog表满了
alarm-segfault: uwsgi 段错误了

Warning

注意，uwsgi缓存只会以文本缓存response的内容, 所以查询缓存并返回给client的时候，它默认是以text/html返回的。可以用 content_type指令来指定

uwsgi 的缓存可以监控状态:: https://pypi.python.org/pypi/uwsgicachetop

其他插件¶

Legion(军团系统)

uwsgi的集群，当不适用前端web server时， uwsgi也可以配高可用，就是这套集群系统。但是还是nginx的upstream集比较安全可靠.

cheaper

根据uwsgi 负载, 动态资源分配(workprocess 数量内存占用...)

交互式启动uwsgi¶

command:

uwsgi --chdir . --file mysite/wsgi.py --callable application  --master --processes 1 --http :9090 --harakiri 10

常见问题¶

当设置了 max-requests 后，每个worker会在收到那么多请求后自动重启，会在uwsgi log 里写入这么几行:

Fire in the hole
Sat Jul 11 13:48:19 2015 - Fire in the hole !!! (60 seconds to detonation)
Gracefully killing worker 1 (pid: 22813)...
Sat Jul 11 13:48:19 2015 - stopping gevent signals watchers for worker 1 (pid: 22813)...
Sat Jul 11 13:48:19 2015 - stopping gevent sockets watchers for worker 1 (pid: 22813)...
Sat Jul 11 13:48:19 2015 - main gevent watchers stopped for worker 1 (pid: 22813)...
Sat Jul 11 13:48:19 2015 - worker 1 (pid: 22813) core 93 is managing "GET /"
Sat Jul 11 13:48:19 2015 - worker 1 (pid: 22813) core 95 is managing "GET /"

如果同时使用 gevent 和 cpu_affinity 时，有可能导致worker进程不能被有效回收，导致可工作的worker越来越少，最终服务挂掉. 这个可以从uwsgi log里看到， respawn的worker id 越来越固定

未解决¶

调优: If you are running only a single application you can disable multiple interpreters. 为什么???

进阶¶

减负线程: https://uwsgi-docs.readthedocs.org/en/latest/OffloadSubsystem.html

Harakiri 模式¶

Harakiri有两种模式:

每个worker在接受每个request前，调用alarm，这种方式是不可靠的。因为handler里面可能调用alarm来取消当前设置的超时
master 进程创建一个shared memory zone, 每个worker处理请求之前，在pool中记录start时间，结束后删除。另外有一个monitor进程，不停去检测pool 中超时的worker, 找到后kill

标准配置¶

example:

[uwsgi]
chdir = .
chmod-socket=664
module = mysite.wsgi:application
master = True
processes = %(%k * 4)
pidfile = /tmp/%n.pid
uwsgi-socket = /tmp/%n.sock
# http 监控(返回json)
# stats = 127.0.0.1:1717
stats = /tmp/%n_stat.sock
# uwsgi 退出的时候，清理环境, 如unix文件/ pid文件h
vacuum = True
harakiri = 50
harakiri-verbose = True
single-interpreter = True
memory-report = True
max-requests = 10000
listen = 8192

log-truncate = True
disable-logging = True
# logformat = i am a logline reporting "%(method) %(uri) %(proto)" returning with status %(status)
daemonize = /home/log/uwsgi/%n_uwsgi.log
procname-master = uwsgi --ini %n.ini (master)
procname = uwsgi --ini %n.ini

need-app = True

# where n is the number of cpus to dedicate per-worker
# 和gevent配合使用，可能会导致worker不被回收， 可用的worker会慢慢减少，减到1个，server就会挂掉
# cpu_affinity = 1

# 只对tcp有效
# so-keepalive = True

# 设置worker一定时间自动重启(second), 默认不自动重启
# max-worker-lifetime = 600