Replies: 2 comments 1 reply
-
|
as experiment creates POD and work in similar way as in canary. you can set weight for the experiment . If 20% is set for the experiment and this will be deducted from total traffic. You can time the experiment and after specified duration, it will return to pre experiment state. Yes abort does the same as well |
Beta Was this translation helpful? Give feedback.
-
|
Thank you @harikrongali - that sounds promising!
this is probably the part that I am missing out: I do not see the traffic weights being applied via an experiment - i only see it when the experiment is used as part of a rollout. Is there a |
Beta Was this translation helpful? Give feedback.
-
|
https://argoproj.github.io/argo-rollouts/features/experiment/#integration-with-rollouts |
Beta Was this translation helpful? Give feedback.
-
Background
We usually do our wargames by having a dedicated service instance and push more traffic to it than all the other instances.
That way we can simulate how services will behave under higher load in general & see how they will fail.
We use this approach as our production traffic is very hard to artificially generate.
Question
How does the traffic distribution work when there is an experiment started in a rollout-scenario?
We use Argo Rollouts very successful with Traffic Management against AWS ALB & Istio.
How do experiments work?
Are the pods started and they get the same traffic percentage wise as regular pods?
Assuming 9 regular pods, every gets 1/9 of the traffic.
Adding 1 pod with an experiment, is now each pod getting 1/10 of the traffic?
Can we somehow force the experiment to take more of the traffic, preferably very manually via
kubectlor something? (Also with some kind ofkubectl experiment abortthat just gets us back to the pre-experiment-running state)Beta Was this translation helpful? Give feedback.
All reactions