Krr since 1.5.X does not return recommendations

[agabrys]

Describe the bug
It is a continuation of #127. Unfortunately, I cannot reopen the issue.

Our Prometheus scrapes all requried metrics (#150):

• container_cpu_usage_seconds_total
• container_memory_working_set_bytes
• kube_replicaset_owner
• kube_pod_owner
• kube_pod_status_phase

When I execute a command as follows

krr simple --verbose --kubeconfig kubeconfig--perfx-k8s--agabrys.yaml --namespace monitor-operator --history_duration 1 --timeframe_duration 2

I get different results depending on the used version.

Version	CPU Recommendations	Memory Recommendations	Logs (Gist)
1.4.1	Yes	Yes	krr-simple-1.4.1
1.5.3	No	Yes	krr-simple-1.5.3
1.6.0	No	No	krr-simple-1.6.0
1.6.0 (34974bf9)	No	No	krr-simple-1.6.0-dev-34974bf9

The metrics were collected by the test environment for around 3 hours (should be fine because I set --history_duration to 1 hour).

To Reproduce

1. Setup test environment where Prometheus scrapes:

   • container_cpu_usage_seconds_total
   • container_memory_working_set_bytes
   • kube_replicaset_owner
   • kube_pod_owner
   • kube_pod_status_phase

2. Wait at least 1 hour, so historical data can be collected

3. Execute

   krr simple --verbose --namespace <your-namespace> --history_duration 1 --timeframe_duration 2

   with using different versions

Expected behavior

Recommendations should be calculated.

Are you interested in contributing a fix for this?

I can help with testing. I learned how to build the software.

Desktop

• Computer:
  • OS: Microsoft Windows 10.0.22621
  • Platform: Win32T
• Shell: PowerShell 7.4.0

[hofmann-dev]

I have exactly the same problem, and when I use version 1.4.1, it works like a charm.

[arikalon1]

thanks for reporting it @agabrys & @hofmann-dev
We're checking it

[LeaveMyYard]

So basically at some point we added a minimum amount of data points the simple strategy requires for calculation. This is done to not give recommendations on very small amounts of data. By default it is set on 100 points.

For your case , if you want 1 hour data with 2 minute timeframe (that's 30 points) you can set --points_required 15 (it will filter pods having only 30 minutes of data, as `points_required * timeframe_duration = 15 * 2 minutes = 30 minutes)

Also you can do --points_required 1 to use all available data

But I personally would not recommend using only 1 hour of data in general case, as from personal experience that is not enough.

[agabrys]

Oh, I did not take that into account. I ran KRR 1.6.0 on production today and everything seems fine 🥳
It means that adding the recommended metrics is sufficient 🥳

But I personally would not recommend using only 1 hour of data in general case, as from personal experience that is not enough.

That was my test environment. I use a longer time period in production. I will remember to set --points_required 1 for testing 👍