Hi,
We confirmed monitoring system is measuring correctly. Turning off sticky
sessions did the trick and requests are being routed almost equally. We made
our app work without sticky sessions.
Vamsee
On Thursday, January 26, 2017 2:57 AM, Luca Toscano
<toscano.l...@gmail.com> wrote:
Hi,
2017-01-25 22:05 GMT+01:00 Vamsee Lakamsani <vam...@yahoo.com.invalid>:
Hi,
We are using mod_proxy_balancer (mod_lbmethod_byrequests) with Apache httpd 2.4
on Amazon Linux 4.4.41-35.53.amzn1.x86_ 64 (2016.09). We are using the docs
from here:
mod_proxy_balancer - Apache HTTP Server Version 2.4
|
| |
mod_proxy_balancer - Apache HTTP Server Version 2.4
| |
|
Here is our balancer setup:
ProxyHCExpr ok234 {%{REQUEST_STATUS} =~ /^[234]/}
Header add Set-Cookie "ROUTEID=.%{BALANCER_WORKER_ ROUTE}e; path=/"
env=BALANCER_ROUTE_CHANGED
<Proxy "balancer://mainBalancer">
BalancerMember "http://api1:3000"; route=1 loadfactor=1 hcmethod=GET
hcuri=/onkore/api/v1/time hcexpr=ok234 hcinterval=10 hcpasses=2 hcfails=3
BalancerMember "http://api2:3000"; route=2 loadfactor=1 hcmethod=GET
hcuri=/onkore/api/v1/time hcexpr=ok234 hcinterval=10 hcpasses=2 hcfails=3
BalancerMember "http://api3:3000"; route=3 loadfactor=1 hcmethod=GET
hcuri=/onkore/api/v1/time hcexpr=ok234 hcinterval=10 hcpasses=2 hcfails=3
BalancerMember "http://api4:3000"; route=4 loadfactor=1 hcmethod=GET
hcuri=/onkore/api/v1/time hcexpr=ok234 hcinterval=10 hcpasses=2 hcfails=3
BalancerMember "http://api5:3000"; route=5 loadfactor=1 hcmethod=GET
hcuri=/onkore/api/v1/time hcexpr=ok234 hcinterval=10 hcpasses=2 hcfails=3
BalancerMember "http://api6:3000"; route=6 loadfactor=1 hcmethod=GET
hcuri=/onkore/api/v1/time hcexpr=ok234 hcinterval=10 hcpasses=2 hcfails=3
BalancerMember "http://api7:3000"; route=7 loadfactor=1 hcmethod=GET
hcuri=/onkore/api/v1/time hcexpr=ok234 hcinterval=10 hcpasses=2 hcfails=3
BalancerMember "http://api8:3000"; route=8 loadfactor=1 hcmethod=GET
hcuri=/onkore/api/v1/time hcexpr=ok234 hcinterval=10 hcpasses=2 hcfails=3
BalancerMember "http://api9:3000"; route=9 loadfactor=1 hcmethod=GET
hcuri=/onkore/api/v1/time hcexpr=ok234 hcinterval=10 hcpasses=2 hcfails=3
BalancerMember "http://api10:3000"; route=10 loadfactor=1 hcmethod=GET
hcuri=/onkore/api/v1/time hcexpr=ok234 hcinterval=10 hcpasses=2 hcfails=3
ProxySet lbmethod=byrequests stickysession=ROUTEID nofailover=Off
</Proxy>We are seeing (as measure by New Relic our monitoring service) that
http request are not being distributed equally. The first server (api1) is
getting more than the second one and so on with the last one getting the least
amount of request. Attached is a screen shot that shows this.
Quick sanity check: have you verified from the httpd access logs that the
monitoring service is reporting correct metrics?
We would appreciate any feedback on what is causing this and how to fix. A
related question is on the health checks. Is there any way to avoid
copy/pasting the health check info on all the member lines?
https://httpd.apache.org/docs/2.4/mod/mod_proxy_balancer.html#stickyness
I would check what is the effect of the configured sticky session to the load
balancer behavior (for example do some tests without it if possible).
Hope that helps!
Luca
