feat: public pricing API + DaemonSet-aware node scoring#4
Merged
Conversation
Replace the AWS Pricing API (requires pricing:GetProducts IAM permission) with the public runs-on.com EC2 pricing API. No authentication required, includes both on-demand and spot prices, updated hourly. - Remove aws-sdk-go-v2/service/pricing dependency entirely - Simplify AWSProvider: only needs ec2:DescribeInstanceTypes permission - EnrichWithPricing called automatically in GetInstanceTypes - Update default instance families to current gen (m7i, c7i, r7i, etc.) - Fix "0 scenarios" log message (was hardcoded, never updated)
Per AWS docs: "fewer, larger instances are better, especially if you have a lot of DaemonSets" — each DS runs on every node, so more nodes means more wasted resources on DS replicas. - Refine resilience scoring: sweet spot at 5-15 nodes, progressive penalty above 30 nodes instead of flat 100 for 3-50 - Add DaemonSet overhead penalty: high DS count × high node count reduces the resilience score (up to -20 points) - Pass DaemonSetCount from orchestrator to scorer
Add cluster-wide P95 CPU/memory and observed min/max node count queries to capture HPA/autoscaler scaling peaks that per-pod instant snapshots miss. Enforce a configurable minimum node count (default 3) as an HA constraint in bin-packing. Compute scaling efficiency per candidate instance type and penalize poor trough utilization in scoring. - Add ClusterAggregateMetrics and ScalingEfficiency model types - Replace unused peak replica queries with 4 cluster aggregate PromQL - Add MinNodes to SimulationConfig (default 3) with BFD padding - Compute trough CPU utilization from scaling ratio - Penalize resilience score when trough utilization < 30% - Display cluster P95, node range, min nodes in report headers - Show [trough: XX%] warning in table/markdown notes - Update README with full documentation and correct AWS requirements
Per-pod effective sizing uses max(request, P95_usage) which inflates CPU when pods over-request relative to actual usage. This led to "compute-optimized" classification on a cluster that was actually memory-bound (0.8 vCPU, 9.4 GiB → 11.75 GiB/vCPU). Prefer cluster-level aggregate P95 CPU/memory (from the full metrics window) for classification when available. Fall back to per-pod totals when aggregate metrics are absent.
When auto-classifying to an extreme (compute or memory-optimized), also include M-series (general-purpose) families. Per-pod requests may skew the bin-packing constraint away from the aggregate classification — M-series provides a balanced middle ground that the scorer evaluates alongside the primary family. Fixes clusters where aggregate usage is memory-heavy but per-pod CPU requests are inflated, causing R-series to be CPU-saturated (97%) with wasted memory (32%).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
pricing:GetProducts) with public runs-on.com API — no IAM permissions neededaws-sdk-go-v2/service/pricingdependency entirelyScoring changes
Per AWS: "fewer, larger instances are better, especially with many DaemonSets"
DaemonSet penalty:
min(dsCount × nodeCount / 100 × 5, 20)— with 12 DS and 40 nodes, that's a -9.6 point penalty.Test plan
go test -race ./...— all passgolangci-lint run— 0 issues