My understanding, and I could very well be wrong. Is that HAProxy is not SMP aware, it's single threaded and will not automatically take advantage of systems with multiple CPU's or cores. Other than the OS scheduler moving things around.
Running multiple instances allows you to peg each instance to a particular CPU, core. Is this correct? On Tue, Jul 1, 2014 at 12:54 PM, Justin Franks <justin.fra...@lithium.com> wrote: > Not sure why you would run multiple HAProxy in one node. I don't > understand what you want to do. But... > > Look into using Consul to help load balance/cluster your HAProxy > instances. That is what we are doing. Simple and works great. Think of > Consul as a Global load balancing service that works internally. > > Consul will look at all your instances (HAProxy or whatever) and > round-robin or weight load balance to all of them based on health checks > you plug in. I don't know if this would solve your prob though. > > > > ************************* > Justin Franks > Lead Operations Engineer > SaaS, Cloud, Data Centers & Infrastructure > Lithium Technologies, Inc > 225 Bush St., 15th Floor > San Francisco, CA 94104 > tel: +1 415 757 3100 x3219 > ------------------------------ > *From:* Xu (Simon) Chen <xche...@gmail.com> > *Sent:* Monday, June 30, 2014 7:38 AM > *To:* HAProxy > *Subject:* single or many haproxy instances > > Hi folks, > > I am writing a simple load balancer as a service to automate haproxy > configuration while providing a simple API to users, who only need to give > a few simple specifications of the load balancer they want. > > I am trying to decide whether to run multiple haproxy instances or a > single instance on a particular node. I currently use jinja2 template to > combine all services into a single haproxy configuration file and run a > single instance of haproxy. Every time, when a service spec is changed, I > run check config mode, and only reload the config if the test passes. But I > fear that a single incorrect service spec would prevent everyone else from > updating their services, unless I maintain some last-known good config for > every service. > > Managing one haproxy instance for every service solves this problem, but > I might end up with too many processes on a single box. > > Any recommendations on which way to go? Is there a recommended max number > of haproxy instances per node/core? > > Thanks. > -Simon > -- *Zachary Buckholz* E: zachary.buckh...@pearson.com T: 480-457-7789 PearsonAlways Learning Learn more at www.pearson.com
