I've been running Prisma Cloud on a mid-sized EKS cluster (~150 nodes, 2000+ pods) for about six months. The compute defender and console are configured per their docs, but the scan latency is killing our CI/CD pipeline. A full vulnerability scan on a new image push can take upwards of 25 minutes, which is a huge bottleneck.
I've tried tweaking the collector resources, but it feels like the architecture itself might not be optimized for scale. Here's a snippet of our current defender config for the daemonset:
```yaml
resources:
limits:
cpu: "1000m"
memory: "2Gi"
requests:
cpu: "500m"
memory: "1Gi"
```
Has anyone else hit a wall with performance on large k8s workloads? I'm curious about:
* Whether you moved the defender to a sidecar model instead of daemonset for certain namespaces.
* If pruning the audit logs or adjusting the `resources.profiles` scope made a tangible difference.
* Any API tuning parameters that aren't well-documented.
The data is valuable, but the speed is becoming a deal-breaker for our dev teams who are used to near-instant feedback from other tools in their stack. I'm wondering if this is a common experience or if our setup is just missing a key optimization.