- Configuring Perlbal
- Load Balancing
- HTTP, SSL
- SEE ALSO
Perlbal::FAQ - Frequently Asked Questions about Perlbal
This document aims at listing several Frequently Asked Questions regarding Perlbal.
Yes, you can find one under
debian/perlbal.init. It implements
restart/force-reload. Make sure you adjust it to your particular taste and/or needs.
No, there is not. But typically, if you're making changes, you can just make them on the management console, which doesn't require any restart whatsoever.
Still, restarting is probably easy. The trick to it is to simulate a graceful restart.
Here's a sample script that will allow you to perform a graceful restart:
$ cat restart-perlbal.sh echo "shutdown graceful" | nc localhost 60000 /usr/local/bin/perlbal --conf=/etc/perlbal.conf
The idea is that you tell the old Perlbal to do a graceful shutdown; that immediately closes all of the listening sockets, so new connections are not accepted. As soon as that's done (which is instant) you can start up a new Perlbal.
This gives you a minimum of downtime that can be measured on the order of milliseconds (the time it takes for the new Perlbal to start up).
Remember that you need to have a
management service listening on port 60000 for this example to work. See Perlbal::Manual::Management.
Currently, Perlbal supports only one balancing method:
SET pool balance_method = 'random'
With this mode, Perlbal selects one of the nodes within the pool randomly for each request received. It prefers reusing existing idle backend connections if backend_persist is enabled, which is faster than waiting for a new connection to open each time.
Yes. When you set the plugins for your service they get to register their hooks in order.
SET plugins = AccessControl HighPri
These hooks are pushed into an array, which means that they preserve the order of the plugins.
Perlbal for the most part only speaks HTTP/1.0 both to clients and to backend webservers. It happily takes requests advertising HTTP/1.1 and downgrading them to HTTP/1.0 when speaking to backend serves.
It knows all about persistent connections (in both 1.0 and 1.1) and will reply with HTTP/1.0 Connection: keep-alive the request was only implicitly keep-alive with HTTP/1.1. etc.
Perlbal is now also starting to speak more of 1.1. For instance, Perlbal does support receiving transfer-encoding "chunked" requests from clients (a feature of HTTP/1.1), will send a
100 Continue in response to
Expect: 100-continue, and will parse the chunked requests, writing the request-of-unknown-length to disk (only if
buffered_uploads is enabled), and then will send an HTTP/1.0 request to the backends, with the actual
Content-Length (now known) filled in.
When more of 1.1 is supported, it will become an option, and later become the default. However, after several years of usage, there just hasn't been that much of a reason. The chunked requests (common from mobile phones uploading large images) has been the most annoying shortcoming but now that it's solved, it's questionable whether or not more of HTTP/1.1 will be supported.
Yes. To use SSL mode you'll need IO::Socket::SSL
You can do SSL either on
selector modes, but not on a vhost-based
selector service, because SSL and vhosts aren't compatible.
See the configuration file ssl.conf under conf/ for an example.