Hey there, two updates on this.
* 2.6.22 works better when I add some jitter on the requests, it seems data was colliding; * Just re-read the doc and found that "The pushed values overwrite remote ones without aggregation.”, which means that 2.6 is misbehaving, and 3.2 is working as expected. :( Wondering how do you folks use to aggregate values. I could maintain one table per instance on every instance and sum all of its metrics, but wondering if there is an easier and more straightforward way out there. --jm > On 1 Aug 2025, at 09:16, Joao Morais <l...@joaomorais.com.br> wrote: > > > Hey folks, I’m doing some experiments with stick-table and peers. This is my > configuration: > > h1.cfg .. h9.cfg > > global > localpeer h1 # <<--- changing to h2..h9 on h2.cfg .. h9.cfg > log stdout format raw local0 > listen l > timeout server 1m > timeout client 1m > timeout connect 5s > log global > option httplog > mode http > bind :28081 # <<--- changing to 28082 to 28089 > stick-table type string size 1 peers p store gpc(1),conn_rate(10s) > http-request track-sc0 str(1) > http-request return hdr x-gpc %[sc_inc_gpc(0,0)] hdr x-rate > %[sc_conn_rate(0)] > peers p > bind :29001 # <<--- changing to 29002 to 29009 > log stdout format raw local0 > server h1 # <<--- adding h1 endpoint on others > server h2 127.0.0.1:29002 # <<--- endpoint removed only on h2 > server h3 127.0.0.1:29003 # <<--- endpoint removed only on h3 > ... until h9 and :29009 > > Running as `haproxy -f h[1..9].cfg`. > > And these are my findings so far: > > 1. when rotating requests manually between ports 28081 to 28089, I can see: > * on 2.6.22, metrics being shared - gpc updates accordingly, rate seems > coherent with the speed I send the requests; > * on 3.2.3, it seems like the instances are isolated, when rotating the > requests, I cannot see the next instance updated based on data coming from > the former ones. > > 2. when running parallel calls (via vegeta), some instances receiving 10rps, > others receiving 15rps, during 60s: > * 2.6.22 seems to receive metrics from one other peer only, like, rate in the > 10s window about 300 and total gpc about 1800 when the test finishes; > * 3.2.3 seems to continue isolated > > I’ll continue my tests, but wanted to share some partial results, so you > folks can share if I’m having some misunderstanding on how peers works or if > I’m missing some configuration. > > --jm > > >
