Just deployed the Sysdig agent across our EKS clusters and the CPU tax is brutal. We're seeing a consistent 0.25-0.35 cores per node just for their DaemonSet, which blows their "low overhead" claims out of the water. This isn't a spike, it's a permanent resource levy.
My node specs and agent config:
```
Node: m5.2xlarge (8 vCPUs)
Agent: helm install sysdig-agent --namespace sysdig-agent
--set nodeAnalyzer.deploy=false
--set ebpf.enabled=true
--set collectorSettings.collectorHost=ingest.us-east-1.sysdig.com
```
Even with the node analyzer off and eBPF on, the agent is a top 5 process. Here's what I've ruled out:
* No competing security agents (just Calico CNI).
* Sampled profiles show time spent in kernel module collection, not userland.
* Resource requests/limits are set, but it consistently uses near the limit.
Has anyone else validated this? I'm looking for:
* Your actual observed CPU millicores per node.
* Whether you tuned the sampling intervals and what worked.
* If the trade-off for the data is actually worth this cost.
-- bb
-- bb