Suppose we have several identical nodes which are the application servers of some n-tier service. And suppose we use Apache ZooKeeper to keep all the config's of our distributed application. Plus we have an nginx as a load balancer and reverse proxy in front of this application.
So let's say we perform a command which changes data only on node1, and for some period of time node2 differs from node1. And we want proxy to redirect all that special requests (which need that specific data) to node1 until all the infomation has migrated to node2 and node2 has the same data as node1.
Is there any way to make nginx (or other proxy) read its config from Apache ZooKeeper? Or more broader: is there any way to effectively switch proxy configuration on fly? And of course it should be done without (or with minimal) downtime of the whole system - so restarting nginx is not the option.
Nginx has two methods of changing configuration:
HUP
signal to the master process results in "reload". Nginx starts a bunch of new workers and lets the old workers to shutdown gracefully, i.e. they finish existing requests. There is no interruption of service. This method of configuration change is very lightweight and quick, but has few limitations: you cannot change cache zones or re-compile Perl scripts.USR2
signal, thenWINCH
and thenQUIT
to the master process result in "executable upgrade" and this sequence lets completely re-read whole configuration and even upgrade the Nginx executable. It reloads disk caches as well (which maybe time consuming). This method results in no interruption of service too.
Please try Nginx-Clojure. We can use a clojure/java/groovy rewrite handler to access zookeeper then update some nginx variables to dynamically change proxy target. e.g.
In nginx.conf
set $mytarget "";
location / {
rewrite_handler_type java;
## We will change $mytarget in MyRewriteHandler
rewrite_handler_name my.MyRewriteHandler;
proxy_pass $mytarget;
}
In MyRewriteHandler.java
public static class MyRewriteHandler implements NginxJavaRingHandler {
@Override
public Object[] invoke(Map<String, Object> request) {
//access zookeeper
...............
//change nginx variable mytarget
((NginxJavaRequest)request).setVaraible("mytarget", "http://some-host-or-url");
}
As an update:Hipache stores its host configuration in redis, which can easily be manipulated at runtime. It's also based on node.js and node-http-proxy.
There is an interesting project using nginx Lua to allow dynamic configuration of nginx and doing exactly what you want (https://github.com/samalba/hipache-nginx)
It is written by the guys behind Hipache.
This may be late but if you have the money. Nginx plus is exactly for you. It uses a simple url call to get new configurations on the fly.
It is possible using HAProxy and its UNIX domain socket interface: http://cbonte.github.io/haproxy-dconv/configuration-1.5.html#9.2.
It supports switching a server or an entire front-end from down to up and back again on the fly. With a configuration file that defines two sets of front-ends, each configured for one specific state, you would be able to achieve what you want.
From Docs:
nginx -s reload
-s
is for 'signal', where signal can be 'quit', 'reload', 'reopen', or 'stop'.
Not sure that it is possible to dynamicaly change nginx
configuration without restarting a server.
If I had a same requirement I'd probably dug into nodejs
and zookeeper
integration.
There are several interesting opensource projects:
node-zookeeper integrates nodejs
with zookeeper
;
node-http-proxy proxy http server that can be used for load balancing.
Of course they lack maturity but they might be interesting for you.
来源:https://stackoverflow.com/questions/8982717/is-there-any-way-to-configure-nginx-or-other-quick-reverse-proxy-dynamically