devops

🚀 DevOps & SRE live

2026–2027 path from Linux kernel and eBPF to Bare-Metal Talos/Cilium BGP, vLLM GPU clusters, and Chaos Drills (Maps 01–20)

▸ Stage 1. Linux Kernel Internals, eBPF & Networking (Map 01) live

  • Process scheduler (CFS), states R/S/D (uninterruptible I/O sleep) and Z (zombies), PID 1 signal handling with tini
  • Namespaces (pid, net, mnt, user) and cgroups v2 (memory.max, cpu.max, OOM Killer scoring)
  • Filesystems, inodes (df -i), and fixing 100% disk usage from deleted open files via lsof +L1 and /proc/<PID>/fd
  • USE/RED triage in 60 seconds: uptime, vmstat 1, iostat -xz 1, ss -tulpan, and eBPF kernel tracing
  • TCP/IP handshake, TLS 1.3, DNS troubleshooting (dig +trace), nf_conntrack tuning, and hardened Nginx/Envoy

▸ Stage 2. Engineering in Python/Go & Container Internals (Map 02) live

  • Python automation with uv, httpx, boto3, official kubernetes client, Click/Typer CLIs, pytest, and ruff
  • Go for cloud-native engineers: goroutines, channels, context timeouts, Cobra CLIs, and static CGO_ENABLED=0 builds
  • OCI runtime internals: containerd -> containerd-shim -> runc, and OverlayFS lowerdir/upperdir/merged layers
  • Multi-stage distroless Dockerfiles running as nonroot (UID 65532) with readOnlyRootFilesystem

▸ Stage 3. CI/CD Pipelines, Dagger, OIDC & OpenTofu GitOps (Map 03) live

  • GitHub Actions and GitLab CI with rootless BuildKit/Kaniko instead of privileged Docker-in-Docker
  • Zero static cloud keys in CI via short-lived OIDC JWT federation (AssumeRoleWithWebIdentity)
  • Programmable pipelines as code in Go using the Dagger SDK for 100% local-to-CI parity
  • Modular OpenTofu/Terraform with remote S3 state locking, DRY Terragrunt wrappers, and idempotent Ansible roles
  • Pull-based GitOps delivery with ArgoCD (App-of-Apps, ApplicationSet) and Argo Rollouts Canary with Prometheus rollback

▸ Stage 4. Kubernetes Architecture, Cilium eBPF & Autoscaling (Map 04) live

  • Control Plane internals: kube-apiserver, etcd Raft quorum and snapshot restore, kube-scheduler, and kubelet
  • Workload resilience: startup/liveness/readiness probes, QoS classes (Guaranteed/Burstable), PDB, and TopologySpreadConstraints
  • Cilium eBPF CNI replacing kube-proxy iptables, L7 Envoy policies, and Kubernetes Gateway API (HTTPRoute)
  • Event-driven workload scaling with KEDA (Kafka lag, Prometheus queries) and bin-packing Spot node provisioning with Karpenter

▸ Stage 5. Observability Stack, eBPF Profiling & Supply Chain Security (Maps 05, 13, 14) live

  • Prometheus and VictoriaMetrics (PromQL rate/histogram_quantile) with cardinality discipline and Alertmanager routing
  • Structured JSON logging with Vector and Grafana Loki (LogQL) correlated with OpenTelemetry and Tempo traces via trace_id
  • Zero-code eBPF telemetry with Grafana Beyla and continuous CPU/memory FlameGraph profiling with Pyroscope
  • Secrets management with Vault/OpenBao + ExternalSecrets, SBOMs (Syft), keyless Cosign signatures, and Kyverno policies

▸ Stage 6. SRE Practices, Error Budgets & Incident Playbooks (Maps 06, 09, 10, 12) live

  • SLI/SLO/SLA math (99.9% = 43.2m monthly downtime), Error Budget burn-rate alerts, and Blameless Postmortems (5 Whys)
  • First-5-minute incident runbooks: CrashLoopBackOff exit codes (137/143/126/127), OOMKilled, and 502/504 storms
  • On-call DBA runbook: pg_stat_activity lock trees, lock_timeout DDL, autovacuum dead tuples, and PgBouncer transaction pooling
  • Continuous resilience verification with Chaos Mesh fault injection and Grafana k6 stress testing

▸ Stage 7. Bare-Metal Kubernetes: Talos Linux, Cilium BGP & Storage (Map 17) live

  • Immutable API-driven Talos Linux: zero SSH, zero shell, declarative upgrades and recovery via mTLS talosctl gRPC API
  • Cilium BGP Control Plane on bare metal: announcing LoadBalancer VIPs via ECMP to Top-of-Rack switches
  • Bare-metal CSI architecture: distributed Rook-Ceph (RWO/RWX/S3) vs local NVMe TopoLVM for replicated databases

▸ Stage 8. Platform Engineering: Go Operators, Backstage IDP & Multi-Cluster (Maps 16, 18) live

  • Custom Kubernetes Operators in Go with Kubebuilder (controller-runtime): CRDs, idempotent Reconcile, OwnerReferences, Finalizers
  • Internal Developer Portal (IDP) with Backstage Software Templates and Crossplane XRDs for 3-minute service onboarding
  • Fleet lifecycle management via Cluster API (CAPI), 5-second ephemeral PR clusters via vCluster, and Cilium Cluster Mesh

▸ Stage 9. AI/GPU Infrastructure & Highload Data Streaming (Maps 14, 19) live

  • GPU orchestration in Kubernetes (NVIDIA GPU Operator, DRA/MIG) and LLM serving with vLLM, KEDA, and LiteLLM/Envoy AI Gateway
  • ZooKeeper-less Apache Kafka in KRaft mode via Strimzi Operator with S3 Tiered Storage
  • Petabyte OLAP clusters on ClickHouse with ReplicatedMergeTree, C++ ClickHouse Keeper, and S3 Lakehouse tiers

▸ Stage 10. Chaos Drill Lab, System Design (10k+ RPS) & Career XYZ (Maps 07, 08, 11, 15, 20) live

  • Hands-on Chaos Drill Simulator (break-my-cluster.yaml + verify-fixes.sh): fixing readOnlyRootFilesystem, OOM/probes, selectors, and NetworkPolicy
  • Production Pet-Project Blueprint: Go/Python service + Postgres + Redis + Terraform + ArgoCD + Prometheus/Loki + k6
  • Senior/Staff System Design (10k+ RPS, Multi-AZ, RTO/RPO) and ATS-proof resume engineering with the Google XYZ formula

Source: roadmaps/devops.json · ETA: 2026-11