Resource Footprint
Resource requirements
Section titled “Resource requirements”The agent typically requires ~2 CPU cores and 4 GB memory across all components. Log collection adds ~100m CPU and 128Mi memory per node in your cluster.
Storage: ~20 GB persistent (25 GB with repository caching enabled).
Network: under 150 MB/month outbound to the PulseStream control plane. All traffic is HTTPS (TLS 1.2+). No inbound connections into your cluster are required.
What scales with your cluster
Section titled “What scales with your cluster”The agent’s resource footprint is mostly fixed regardless of cluster size. The one exception is log collection, which runs once per node:
| Cluster size | Additional overhead |
|---|---|
| 3 nodes | +300m CPU, +384Mi memory |
| 10 nodes | +1,000m CPU, +1.28Gi memory |
| 25 nodes | +2,500m CPU, +3.2Gi memory |
If your cluster has dedicated infrastructure nodes that don’t run application workloads, you can scope log collection to application nodes only via a node affinity rule in your Helm values.
Node sizing
Section titled “Node sizing”The agent shares nodes with your existing workloads. You do not need dedicated nodes. Ensure nodes have at least 20 GB of ephemeral storage headroom, as the agent uses up to 5 GB of temporary disk during active incident analysis.
| Scenario | Recommended node size |
|---|---|
| Shared node (dev/staging) | 4 vCPU / 8 GB RAM |
| Shared node with headroom | 4 vCPU / 16 GB RAM |
| Dedicated agent node | 2 vCPU / 8 GB RAM |
On EKS, t3.xlarge or m5.large are common choices. On GKE, e2-standard-4 is comparable.
Estimated infrastructure cost
Section titled “Estimated infrastructure cost”On EKS or GKE with on-demand pricing and shared nodes, the agent typically adds $25-35/month in incremental compute plus $2-5/month in persistent storage.
Log collection adds roughly $0.75/month per node in your cluster.
These estimates assume on-demand pricing in a US region on shared nodes. Reserved instances or committed-use discounts can reduce compute costs by 30-40%.