On 12.10.2010 17:46, David Birdsong wrote:
Haproxy sounds like the perfect tool for it. I know of many companies
that use haproxy simply because they're fronting an ever changing set
of EC2 instances that go up and down all the time.
Haproxy's reload is very clean; it does a very good job of
transitioning ownership of the bound socket between old and new PID's
without dropping incoming requests while existing requests continue to
serve out of the old PID. At my last company, I had a job that wrote
new config files and reloaded haproxy automatically at regular
intervals.
There are some problems I see with reloading:
1. Stats/ counters are reset.
2. Already dead backends start alive again and so the first requests -
until the backend is marked dead again - get errors.
Or are there good ways to handle both drawbacks/ problems? :)
Corin
