Balancer_by_lua performance issues

paul.berthaux · 2018-03-20T18:08:03+00:00

I tried buffering the response in a shared cache, but the decision has to be made over a dynamic response over the same request. Can't seem to be able to do...

Balancer_by_lua performance issues

paul.berthaux

Hi,

I'm currently working on an openresty based loadbalancer with redis as dynamic backend, and I noticed some performances issues when comparing to our static nginx based current solution.

I stripped down most of of the openresty configuration in order to isolate the part causing most of latency variations.
The performance issue seems to be related to the balancer by lua bloc.

When I replace in my legacy nginx static config :

upstream upstream-server-group {

server :443;

keepalive 64;

}

with :

upstream upstream-server-group {

server 127.0.0.1;

balancer_by_lua_block

{

local balancer = require "ngx.balancer"

local ok, err = balancer.set_current_peer("", 443)

if not ok then

ngx.log(ngx.ERR, "failed to set the current peer: ", err)

return ngx.exit(500)

end

}

keepalive 64;

}

the results of my tests goes from :

Latencies [mean, 50, 95, 99, max] 2.011923ms, 1.972201ms, 2.068322ms, 2.757951ms, 37.345576ms

Latencies [mean, 50, 95, 99, max] 3.164499ms, 2.011021ms, 2.171413ms, 43.643579ms, 214.515421ms

Is this a normal behaviour ? is there something special to do to tune balancer_by_lua_block ?

Regards,
Paul

peter_booth

The values that you report sound very high. How are you taking your latency measurements and what is your test workload? What are the best case (minimum ) latencies?

Are your two test scenarios both stable systems or are they in an overload state?

Sent from my iPad

On Mar 20, 2018, at 6:08 AM, 'Paul Berthaux' via openresty-en <openresty-en@googlegroups.com> wrote:

Hi,

I'm currently working on an openresty based loadbalancer with redis as dynamic backend, and I noticed some performances issues when comparing to our static nginx based current solution.

I stripped down most of of the openresty configuration in order to isolate the part causing most of latency variations.
The performance issue seems to be related to the balancer by lua bloc.

When I replace in my legacy nginx static config :

upstream upstream-server-group {
server :443;
keepalive 64;
}

with :

upstream upstream-server-group {
server 127.0.0.1;
balancer_by_lua_block
{
local balancer = require "ngx.balancer"

local ok, err = balancer.set_current_peer("", 443)
if not ok then
ngx.log(ngx.ERR, "failed to set the current peer: ", err)
return ngx.exit(500)
end
}
keepalive 64;
}

the results of my tests goes from :

Latencies [mean, 50, 95, 99, max] 2.011923ms, 1.972201ms, 2.068322ms, 2.757951ms, 37.345576ms

to

Latencies [mean, 50, 95, 99, max] 3.164499ms, 2.011021ms, 2.171413ms, 43.643579ms, 214.515421ms

Is this a normal behaviour ? is there something special to do to tune balancer_by_lua_block ?

Regards,
Paul

.