Premature failover to backup upstream

[adam]

Hello!

I am using Nginx 1.0.5 to reverse proxy to a pool of three upstreams. 
One of them is configured as a backup upstream.  From time to time, very
rarely, a request will be proxied to the backup upstream when it should
not be.  The two normal upstreams are completely healthy, and at
extremely low load.

Why might this be happening?

I see in the documentation
(http://wiki.nginx.org/HttpUpstreamModule#server) that the upstream
module will shunt to backup if the other upstreams are "down or busy." 
What is the definition of "busy" in this context?

Looking in the debug log, I see many many instances of "http upstream
recv(): -1 (11: Resource temporarily unavailable)".  Is this related?

One thing I haven't tried is to adjust the max_fails parameter, but
perhaps I should.

Posted at Nginx Forum: http://forum.nginx.org/read.php?2,219326,219326#msg-219326