Re: Checking backend downtime (Apache Users)

On Fri, Mar 1, 2019, 9:20 PM Paweł Cholewiński <pawelcholewinski@xxxxx wrote:

Hello,

I configured proxy with proxy loadbalancer to route traffic to backend application servers.
Balancer members do not detect the unavailability of the backend (downtime) and directs traffic to the broken node. As a result, users are waiting for a response from the broken backend servers.
I want the balancer member to change its state to Err when detect backend downtime. I read in the documentation (https://httpd.apache.org/docs/2.4/mod/mod_proxy.html#proxypass), that the unavailability of the backend should be examined using the ping parameter.

ping .. “This feature has been added to avoid problems with hung and busy backends.”..

When defining a balancer member, I use the ping option:
ProxyPass "/" "balancer://test-site-local-balancer/"
<Proxy "balancer://test-site-local-balancer">
BalancerMember "http://127.0.0.1:2010" ping=2
..

Pawel,

I hope you are not using sticky session. That usually cause this type of problem.

Another thing I can see that you have not set any RETRY value explicitly for the balancer member. Apache will continue to use the default value zero, which means continuous retry even the server is down. Please set RETRY value to 300sec or something like that and re-test. Later you can tune it as per your requirement.

The configuration does not work as I expected – loadbalancer still directs traffic to the unavailable node. I'm listening packets using tcpdump and I do not see additional packets to port 2010 which should be associated with ping HTTP/1.1 100-Continue requests.
I use Debian 10 with Server version: Apache/2.4.38 (Debian).
How to setup ping mechanism correctly to detects unavailable backend servers?

Thanks,
Paweł