serving
serving copied to clipboard
Activator LB doesn't work well
What version of Knative?
0.9.x 0.10.x 0.11.x Output of
git describe --dirty
0.25.2
Expected Behavior
Settings:
- 2 activator
- no istio sidecar injection
- 5 replica server
- CC: 1
- TBC : -1
- no istio sidecar injection
With CC: 1 and no istio sidecar injection, I understood that firstAvailableLBPolicy
is used.
I expected that all 5 pod get requests evenly through Activator.
Actual Behavior
Four pod got requests evenly, but one pod didn't even get a single request at all. ( I sent around 20000 requests)





Steps to Reproduce the Problem
I made a simple FastAPI webserver with an API that waits for 0.1 seconds.
Sent requests using hey in local.
hey -t 0 -n 10000 -c 100 ${URL}
Similar issue : https://github.com/knative/serving/issues/12593
if the requests are done quickly enough, the same pod can handle them, then the activator might not need to send the any requests to the last pod
@nader-ziada
if the requests are done quickly enough, the same pod can handle them, then the activator might not need to send the any requests to the last pod
Even if all pods were busy, the remaining requests were not sent to some pods.
In the above test, there are 5 pods, and each pod processes 10 rps. Total traffic that can be handled is 50 rps.
On client side, I sent 100 rps using hey
.
This issue is stale because it has been open for 90 days with no
activity. It will automatically close after 30 more days of
inactivity. Reopen the issue with /reopen
. Mark the issue as
fresh by adding the comment /remove-lifecycle stale
.
/reopen
/triage accepted /milestone v1.11.0
Want to priortize an investigation in v1.11.0 timeframe
Hi @dprotaso,
May I work on this issue? /assign
The bug has been reproduced on Kind environment. I will investigate on it.