Platform SLIs and SLOs

Define platform SLIs: API latency, scaffold-to-deploy time, environment provisioning time, CI queue time. Error budgets for platform.

8 min readPlatform Engineering & Developer ExperienceMaster

Why Platform SLIs?

Platform teams often measure infrastructure health (cluster up, API server latency) but miss developer experience. Platform SLIs measure the platform from the developer's perspective: how fast can I scaffold? How long until my code deploys? How often does CI queue block me?

Infrastructure SLI	Platform SLI	Why It Matters
API server latency (p99)	Platform API latency (p99)	Developer waits for every CLI command
Cluster CPU utilization	Scaffold-to-deploy time (p50)	Measures end-to-end developer flow
Node count	Environment provisioning time (p95)	Preview env speed = iteration speed
etcd latency	CI queue time (p95)	Queue time = wasted developer time
Pod restart rate	Golden path adoption rate	Adoption = platform value delivery
Disk pressure	Developer NPS (quarterly)	Subjective but critical

Key Platform SLIs

SLI	Definition	Target	Measurement
Scaffold Success Rate	% of `platform scaffold` commands completing without error	≥99.5%	CLI telemetry + API logs
Scaffold-to-Deploy Time (p50)	Time from successful scaffold to first production deploy	<15 min	Git webhook → ArgoCD sync → deployment
Scaffold-to-Deploy Time (p95)	Same, 95th percentile	<45 min	Same
Environment Provision Time (p95)	Time from PR open to preview URL ready	<3 min	GitHub webhook → namespace + ArgoCD + deps
CI Queue Time (p95)	Time from push to CI start	<5 min	GitHub Actions / GitLab CI queue metrics
Platform API Latency (p99)	p99 latency for mutating API calls	<500ms	API gateway / load balancer metrics
Platform API Availability	% of time API returns 2xx/4xx (not 5xx)	99.9%	Uptime monitor / synthetic checks
Golden Path Adoption	% of new services using golden path templates	>80%	Catalog: services with `platform/golden-path` label
Template Drift Detection Time	Time to detect golden path repo divergence from template	<24h	Scheduled template diff job
Developer NPS	Quarterly survey: 'How likely to recommend platform?'	>30	Survey tool (Delighted, Typeform)

Defining SLOs

SLO = SLI target over a time window. Platform SLOs use rolling windows (28 days for most, 7 days for latency). Error budget = 100% - SLO target.

Platform SLO Definitions


apiVersion: platform.example.com/v1
kind: PlatformSLO
metadata:
  name: platform-slos
spec:
  slos:
    - name: scaffold-success-rate
      sli: scaffold-success-rate
      target: 99.5
      window: 28d
      description: "99.5% of scaffold commands succeed"
    - name: scaffold-to-deploy-p50
      sli: scaffold-to-deploy-time
      target: 900  # 15 minutes in seconds
      percentile: 50
      window: 28d
      description: "50% of services deploy to prod within 15 min of scaffold"
    - name: scaffold-to-deploy-p95
      sli: scaffold-to-deploy-time
      target: 2700  # 45 minutes
      percentile: 95
      window: 28d
    - name: preview-env-provision-p95
      sli: environment-provision-time
      target: 180  # 3 minutes
      percentile: 95
      window: 28d
    - name: ci-queue-time-p95
      sli: ci-queue-time
      target: 300  # 5 minutes
      percentile: 95
      window: 7d
    - name: api-latency-p99
      sli: platform-api-latency
      target: 500  # milliseconds
      percentile: 99
      window: 7d
    - name: api-availability
      sli: platform-api-availability
      target: 99.9
      window: 28d
    - name: golden-path-adoption
      sli: golden-path-adoption-rate
      target: 80
      window: 28d
    - name: template-drift-detection
      sli: template-drift-detection-time
      target: 24  # hours
      window: 28d

SLOs as code: versioned, reviewable, deployable. Each references an SLI definition with target, window, percentile.

Error Budgets

Error budget = (1 - SLO target) × time window. Platform team spends error budget on: new features, tech debt, migrations. When budget exhausted → freeze non-critical changes, focus on reliability.

SLO	Target	Error Budget (28d)	Budget Exhausted Action
Scaffold Success	99.5%	3.36 hours of failures	Freeze new template features, fix flaky scaffold steps
Scaffold-to-Deploy p50	15 min	N/A (latency budget)	Investigate CI/CD bottlenecks, optimize ArgoCD sync
Preview Env p95	3 min	N/A	Scale preview node pool, optimize dependency provisioning
CI Queue p95	5 min	N/A	Add CI runners, optimize test parallelization
API Latency p99	500ms	N/A	Profile API, add caching, scale replicas
API Availability	99.9%	40.3 min downtime	Incident response, postmortem, reliability sprint
Golden Path Adoption	80%	N/A (rate budget)	Improve templates, reduce escape hatches, better DX

Measurement & Implementation

CLI Telemetry: OpenTelemetry in CLI, send spans to platform OTel collector
API Metrics: Prometheus metrics on API gateway (latency, errors, rate)
GitHub/GitLab API: Webhooks for PR events, workflow runs, deployment status
ArgoCD Metrics: Application sync status, sync duration, health
Kubernetes Metrics: kube-state-metrics + custom metrics for platform resources
Survey Tool: Quarterly NPS via Delighted/Typeform, linked to user identity

Alerting on SLOs

Alert on error budget burn rate, not absolute thresholds. Multi-window, multi-burn-rate alerting (Google SRE workbook).

PrometheusRule: Platform SLO Burn Rate Alerts


apiVersion: monitoring.coreos.com/v1
kind: PrometheusRule
metadata:
  name: platform-slo-burn-rate
  namespace: monitoring
spec:
  groups:
    - name: platform-slo-burn-rate
      rules:
        # Fast burn: 2% budget in 1h (14.4x normal rate)
        - alert: PlatformSLOFastBurn
          expr: |
            (
              (1 - platform_scaffold_success_rate_5m) / (1 - 0.995)
            ) > 14.4
          for: 2m
          labels:
            severity: critical
            team: platform
          annotations:
            summary: "Platform scaffold success rate burning error budget fast"
            description: "Burn rate > 14.4x normal. Error budget exhausted in <1h at current rate."
            runbook_url: "https://runbooks.platform.example.com/scaffold-failures"

        # Slow burn: 10% budget in 6h (4x normal rate)
        - alert: PlatformSLOSlowBurn
          expr: |
            (
              (1 - platform_scaffold_success_rate_1h) / (1 - 0.995)
            ) > 4
          for: 15m
          labels:
            severity: warning
            team: platform
          annotations:
            summary: "Platform scaffold success rate burning error budget"
            description: "Burn rate > 4x normal. Error budget exhausted in <6h at current rate."

        # API latency burn rate
        - alert: PlatformAPILatencyBurn
          expr: |
            histogram_quantile(0.99, rate(platform_api_request_duration_seconds_bucket[5m])) > 0.5
          for: 5m
          labels:
            severity: warning
            team: platform
          annotations:
            summary: "Platform API p99 latency > 500ms"
            description: "API latency SLO at risk. Check API gateway, upstream dependencies."

Multi-window burn rate alerts: fast burn (2% budget in 1h) = critical, slow burn (10% in 6h) = warning. Based on Google SRE workbook.

Mark this lesson complete to store local progress and unlock a cleaner resume path the next time you visit.

Stage 7 · Master

Observability & Developer Experience Metrics