Willy Tarreau wrote:
> On Fri, Dec 19, 2008 at 01:19:26AM -0800, Will Buckner wrote:
>> One of the big problems that we were having under high load was that
>> HAProxy would take servers out of the pool and add them back in many
>> times a minute, and this flapping ended up causing more load on the
>> servers that weren't flapping as much. I was seeing a few machines
>> effectively serving zero requests due to the flapping. My rise is set to
>> 2 and fall set to 5. SSL seems to be flapping more than HTTP, but both
>> were flapping. Any ideas on this?
> if you are losing a lot of packets on the interface, health checks will
> fail under traffic. From your description, it is also possible that it's
> the servers' interfaces which fail, so as soon as haproxy excludes a server
> from the pool, the traffic goes down and the checks succeed again.
> Try to run full-speed transfers between the server and the LB, in each
> direction. If you can't reach perfectly stable bandwidth of about 115-118
> megabytes per second, you have a problem somewhere.
Doing speed tests between the HAProxy and webserver machines, I was getting over 120Mb/s. I'm also seeing *no* packet loss/network errors/etc on either machine or the switch. Is there anything else it could be?
Will Received on 2008/12/19 21:42
This archive was generated by hypermail 2.2.0 : 2008/12/19 21:47 CET