Reduce memory consumption

[r2k1]

Opencost stores Kubernetes objects (Pod, ReplicaSet, Deployment, DaemonSet, etc.) with all their fields in memory.

For every request to the /metrics endpoint, a copy of these objects is created, which can lead to significant memory usage, particularly for large clusters.

I suspect this is to prevent accidental modifications, but how likely is that? I think there should be no reason to modify the k8s objects.

I ran some tests on a single-node cluster, creating 300 deployments. Max memory usage dropped from 125MB to 105MB after avoiding the extra copies. Taking generic overhead into account it may produce greater benefits for the large clusters.

Related issue #2637

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Updated (UTC)
opencost	✅ Ready (Inspect)	Visit Preview	May 1, 2024 10:56pm

[sonarcloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarCloud

[r2k1]

Additional Notes:

kube-state-metrics employs certain strategies to optimize memory usage:

• Instead of Informers, it uses Reflector instead of Informer, allowing a direct transformation from a Kubernetes object to a Prometheus string format, retaining the object in memory only momentarily during unmarshalling.
• It has a custom method to serve metrics since the Prometheus Gatherer can be inefficient.
  More details can be found here.

[AjayTripathy]

Thanks @r2k1 . We'll get this reviewed.