Define SLO/SLI/SLA

Set measurable reliability targets with Service Level Objectives, track with indicators, manage error budgets.

When Use

• Define reliability targets for customer-facing services or APIs
• Set clear expectations between service providers and consumers
• Balance feature velocity with system reliability through error budgets
• Create objective criteria for incident severity and response
• Migrate from arbitrary uptime goals to data-driven reliability metrics
• Implement Site Reliability Engineering (SRE) practices
• Measure and improve service quality over time

Inputs

• Required: Service description and critical user journeys
• Required: Historical metrics data (request rates, latencies, error rates)
• Optional: Existing SLA commitments to customers
• Optional: Business requirements for service availability and performance
• Optional: Incident history and customer impact data

Steps

See Extended Examples for complete configuration files and templates.

Step 1: Grok SLI, SLO, SLA Hierarchy

Learn relationship and differences between three concepts.

Definitions:

SLI (Service Level Indicator)
- **What**: A quantitative measure of service behavior
- **Example**: Request success rate, request latency, system throughput
- **Measurement**: `successful_requests / total_requests * 100`

SLO (Service Level Objective)
- **What**: Target value or range for an SLI over a time window
- **Example**: 99.9% of requests succeed in 30-day window
- **Purpose**: Internal reliability target to guide operations

SLA (Service Level Agreement)
- **What**: Contractual commitment with consequences for missing SLO
- **Example**: 99.9% uptime SLA with refunds if breached
- **Purpose**: External promise to customers with penalties

Hierarchy:

SLA (99.9% uptime, customer refunds)
  ├─ SLO (99.95% success rate, internal target)
  │   └─ SLI (actual measured: 99.97% success rate)
  └─ Error Budget (0.05% failures allowed per month)

Key rule: SLO must be stricter than SLA. Gives buffer before customer impact.

Example:

• SLA: 99.9% availability (customer promise)
• SLO: 99.95% availability (internal target)
• Buffer: 0.05% cushion before SLA breach

Got: Team groks differences. Agreement on which metrics become SLIs. Alignment on SLO targets.

If fail:

• Review Google SRE book chapters on SLI/SLO/SLA
• Run workshop with stakeholders to align on definitions
• Start with simple success-rate SLI before complex latency SLOs

Step 2: Pick SLIs

Pick SLIs reflecting user experience and business impact.

Four Golden Signals (Google SRE):

1. Latency: Time to serve request

   # P95 latency
   histogram_quantile(0.95,
     sum(rate(http_request_duration_seconds_bucket[5m])) by (le)
   )
   

2. Traffic: Demand on system

   # Requests per second
   sum(rate(http_requests_total[5m]))
   

3. Errors: Rate of failed requests

   # Error rate percentage
   sum(rate(http_requests_total{status=~"5.."}[5m]))
   / sum(rate(http_requests_total[5m])) * 100
   

4. Saturation: How "full" system is

   # CPU saturation
   avg(rate(node_cpu_seconds_total{mode!="idle"}[5m]))
   

Common SLI patterns:

# Availability SLI
availability:
  description: "Percentage of successful requests"
  query: |
    sum(rate(http_requests_total{status!~"5.."}[5m]))
    / sum(rate(http_requests_total[5m]))
  good_threshold: 0.999  # 99.9%

# Latency SLI
latency:
  description: "P99 request latency under 500ms"
  query: |
    histogram_quantile(0.99,
      sum(rate(http_request_duration_seconds_bucket[5m])) by (le)
    ) < 0.5
  good_threshold: 0.95  # 95% of windows meet target

# Throughput SLI
throughput:
  description: "Requests processed per second"
  query: |
    sum(rate(http_requests_total[5m]))
  good_threshold: 1000  # Minimum 1000 req/s

# Data freshness SLI (for batch jobs)
freshness:
  description: "Data updated within last hour"
  query: |
    (time() - max(data_last_updated_timestamp)) < 3600
  good_threshold: 1  # Always fresh

SLI pick criteria:

• User-visible: Reflects actual user experience
• Measurable: Quantifiable from existing metrics
• Actionable: Team improves through engineering work
• Meaningful: Correlates with customer satisfaction
• Simple: Easy to grok and explain

Avoid:

• Internal system metrics not visible to users (CPU, memory)
• Vanity metrics not predicting customer impact
• Overly complex composite scores

Got: 2-4 SLIs picked per service. Covers availability and latency at minimum. Team agreement on measurement queries.

If fail:

• Map user journey. Find critical failure points
• Analyze incident history. Which metrics predicted customer impact?
• Validate SLI with A/B test. Degrade metric, measure customer complaints
• Start with simple availability SLI. Add complexity iteratively

Step 3: Set SLO Targets and Time Windows

Define realistic, achievable reliability targets.

SLO spec format:

service: user-api
slos:
  - name: availability
    objective: 99.9
    description: |
      99.9% of requests return non-5xx status codes
# ... (see EXAMPLES.md for complete configuration)

Time window selection:

Common windows:

• 30 days (monthly): Typical for external SLAs
• 7 days (weekly): Faster feedback for engineering teams
• 1 day (daily): High-frequency services needing rapid response

Example 30-day window error budget:

SLO: 99.9% availability over 30 days
Allowed failures: 0.1%
Total requests per month: 100M
Error budget: 100,000 failed requests
Daily budget: ~3,333 failed requests

Setting realistic targets:

1. Baseline current performance:

   # Check actual availability over past 90 days
   avg_over_time(
     (sum(rate(http_requests_total{status!~"5.."}[5m]))
     / sum(rate(http_requests_total[5m])))[90d:5m]
   )
   # Result: 99.95% → Set SLO at 99.9% (safer than current)
   

2. Calculate cost of nines:

   99%    → 7.2 hours downtime/month (low reliability)
   99.9%  → 43 minutes downtime/month (good)
   99.95% → 22 minutes downtime/month (very good)
   99.99% → 4.3 minutes downtime/month (expensive)
   99.999% → 26 seconds downtime/month (very expensive)
   

3. Balance user happiness vs engineering cost:

   • Too strict: Expensive. Slows feature development
   • Too loose: Poor user experience. Customer churn
   • Sweet spot: Slightly better than user expectations

Got: SLO targets set with business stakeholder buy-in. Documented with rationale. Error budget calculated.

If fail:

• Start with achievable target (e.g., 99% if current is 98.5%)
• Iterate SLO targets quarterly based on actual performance
• Get executive sponsorship for realistic targets vs "five nines" demands
• Document cost-benefit analysis for each additional nine

Step 4: Do SLO Monitoring with Sloth

Use Sloth to generate Prometheus recording rules and alerts from SLO specs.

Install Sloth:

# Binary installation
wget https://github.com/slok/sloth/releases/download/v0.11.0/sloth-linux-amd64
chmod +x sloth-linux-amd64
sudo mv sloth-linux-amd64 /usr/local/bin/sloth

# Or Docker
docker pull ghcr.io/slok/sloth:latest

Create Sloth SLO spec (slos/user-api.yml):

version: "prometheus/v1"
service: "user-api"
labels:
  team: "platform"
  tier: "1"
slos:
# ... (see EXAMPLES.md for complete configuration)

Generate Prometheus rules:

# Generate recording and alerting rules
sloth generate -i slos/user-api.yml -o prometheus/rules/user-api-slo.yml

# Validate generated rules
promtool check rules prometheus/rules/user-api-slo.yml

Generated recording rules (excerpt):

groups:
  - name: sloth-slo-sli-recordings-user-api-requests-availability
    interval: 30s
    rules:
      # SLI: Ratio of good events
      - record: slo:sli_error:ratio_rate5m
# ... (see EXAMPLES.md for complete configuration)

Generated alerts:

groups:
  - name: sloth-slo-alerts-user-api-requests-availability
    rules:
      # Fast burn: 2% budget consumed in 1 hour
      - alert: UserAPIHighErrorRate
        expr: |
# ... (see EXAMPLES.md for complete configuration)

Load rules into Prometheus:

# prometheus.yml
rule_files:
  - "rules/user-api-slo.yml"

Reload Prometheus:

curl -X POST http://localhost:9090/-/reload

Got: Sloth generates multi-window multi-burn-rate alerts. Recording rules evaluate successfully. Alerts fire during incidents.

If fail:

• Validate YAML syntax with yamllint slos/user-api.yml
• Check Sloth version compatibility (v0.11+ recommended)
• Verify Prometheus recording rule evaluation: curl http://localhost:9090/api/v1/rules
• Test with synthetic error injection to trigger alerts
• Check Sloth docs for SLI event query format

Step 5: Build Error Budget Dashboards

Visualize SLO compliance and error budget consumption in Grafana.

Grafana dashboard JSON (excerpt):

{
  "dashboard": {
    "title": "SLO Dashboard - User API",
    "panels": [
      {
        "type": "stat",
# ... (see EXAMPLES.md for complete configuration)

Key metrics to visualize:

• SLO target vs current SLI
• Error budget remaining (percentage and absolute)
• Burn rate (how fast budget depletes)
• Historical SLI trends (30-day rolling window)
• Time to exhaustion (if current burn rate continues)

Error budget policy dashboard (markdown panel):

## Error Budget Policy

**Current Status**: 78% budget remaining

### If Error Budget > 50%
- ✅ Full speed ahead on new features
# ... (see EXAMPLES.md for complete configuration)

Got: Dashboards show real-time SLO compliance. Error budget depletion visible. Team makes informed decisions about feature velocity.

If fail:

• Verify recording rules exist: curl http://localhost:9090/api/v1/rules | jq '.data.groups[].rules[] | select(.name | contains("slo:"))'
• Check Prometheus datasource in Grafana has correct URL
• Validate query results in Explore view before adding to dashboard
• Set time range to appropriate window (e.g., 30d for monthly SLOs)

Step 6: Set Error Budget Policy

Define org process for managing error budgets.

Error budget policy template:

service: user-api
slo:
  availability: 99.9%
  latency_p99: 200ms
  window: 30 days

# ... (see EXAMPLES.md for complete configuration)

Automate policy enforcement:

# Example: Deployment gate script
import requests
import sys

def check_error_budget(service):
    # Query Prometheus for error budget
# ... (see EXAMPLES.md for complete configuration)

Integrate into CI/CD pipeline:

# .github/workflows/deploy.yml
jobs:
  check-error-budget:
    runs-on: ubuntu-latest
    steps:
      - name: Check SLO Error Budget
        run: |
          python scripts/check_error_budget.py user-api
      - name: Deploy
        if: success()
        run: |
          kubectl apply -f deploy/

Got: Clear policy documented. Automated gates prevent risky deployments during budget depletion. Team aligned on reliability priorities.

If fail:

• Start with manual policy enforcement (Slack reminders)
• Gradually automate with soft gates (warnings, not blocks)
• Get executive buy-in before hard gates (blocking deployments)
• Review policy effectiveness quarterly, adjust thresholds as needed

Checks

• SLIs picked reflect user experience and business impact
• SLO targets set with stakeholder agreement and documented rationale
• Prometheus recording rules generate SLI metrics successfully
• Multi-burn-rate alerts configured and tested with synthetic errors
• Grafana dashboards show real-time SLO compliance and error budget
• Error budget policy documented and communicated to team
• Automated gates prevent risky deployments during budget depletion
• Weekly/monthly SLO review meetings scheduled
• Incident retrospectives include SLO impact analysis
• SLO compliance reports shared with stakeholders

Pitfalls

• Overly strict SLOs: "Five nines" without cost analysis leads to burnout and slowed feature velocity. Start achievable, iterate up.
• Too many SLIs: Tracking 10+ indicators creates confusion. Focus on 2-4 critical user-facing metrics.
• SLO without SLA buffer: SLO equal to SLA leaves no margin before customer impact. Keep 0.05-0.1% buffer.
• Ignoring error budget: Tracking SLOs but not acting on budget depletion defeats purpose. Enforce error budget policy.
• Vanity metrics as SLIs: Internal metrics (CPU, memory) instead of user-visible metrics (latency, errors) misaligns priorities.
• No stakeholder buy-in: Engineering-only SLOs without product/business agreement leads to conflicts. Get executive sponsorship.
• Static SLOs: Never reviewing or adjusting targets as system evolves. Revisit quarterly based on actual performance and user feedback.

See Also

• setup-prometheus-monitoring - Configure Prometheus to collect metrics for SLI calculation
• configure-alerting-rules - Integrate SLO burn rate alerts with Alertmanager for on-call notifications
• build-grafana-dashboards - Visualize SLO compliance and error budget consumption
• write-incident-runbook - Include SLO impact in runbooks for prioritizing incident response