Observability¶

Logs, metrics, traces. The three pillars that turn "production is broken" into "production is broken because the order-service p99 spiked at 14:32 after the deploy at 14:30."

flowchart LR
  APP[Apps + infra] -- OpenTelemetry / agents --> COL[Collectors]
  COL --> L[(Logs:<br/>structured, indexed)]
  COL --> M[(Metrics:<br/>time-series, low cardinality)]
  COL --> T[(Traces:<br/>request spans across services)]
  L --> DASH[Dashboards + queries]
  M --> DASH
  T --> DASH
  M --> ALERT[Alerts on SLO burn]
  ALERT --> ONCALL[On-call / incident manager]

Learn¶

• Observability basics - the three pillars, golden signals, when to use which
• Eventual consistency - the failure modes you'll see in dashboards

Compare¶

• Observability and monitoring services - CloudWatch vs Azure Monitor vs Cloud Operations, Datadog, New Relic
• LLM observability - LangSmith, Langfuse, Helicone, Phoenix, Braintrust

Reference¶

• Architecture pattern: Chaos engineering
• Troubleshooting: AWS
• Troubleshooting: Azure
• Troubleshooting: GCP
• Troubleshooting: Kubernetes

Build¶

• Set up a monitoring stack - Prometheus, Grafana, alerting end-to-end
• Set up an eval harness - LLM-side observability and regression detection

Certify¶

Operations and SRE - AWS CloudOps Engineer (SOA-C03) - AWS DevOps Engineer Professional (DOP-C02) - Azure DevOps Engineer Expert (AZ-400) - GCP Cloud DevOps Engineer - IBM Cloud Site Reliability Engineer

Kubernetes ops - Prometheus Certified Associate (PCA) - CKA - Certified Kubernetes Administrator

Platform / Linux - Linux Foundation LFCS - Red Hat RHCSA (EX200)

LLM observability appears in: - Anthropic Claude Developer Foundations cert - NVIDIA AI Operations Professional

Roadmap¶

The operational view lives in DevOps/SRE roadmap and Platform Engineer roadmap.