teaglebuilt/sre-team

Cilium datapath debugging for the homelab. Use for pod-to-pod connectivity failures and policy drops — endpoint health, identities, IP cache, BPF maps, FQDN cache, encryption state and packet capture. Carries the layer-by-layer method from cheapest to most expensive, the context-blowout warning, and the destructive tools that need confirmation.

Cilium installation and configuration for the homelab — version and daemon status, ClusterMesh peering between the mlops and application clusters, BGP peers and routes, Hubble, LB-IPAM, and service and feature configuration. Use for cross-cluster reachability questions and "is this feature even on". Carries the mesh topology, the declarative-peering rule, and the tools that must never be run casually.

Grafana's own configuration for the homelab — dashboards, datasources, alert rules, Sift investigations, incidents and on-call. Use to find or edit a dashboard, explain where a panel's number comes from, read alert configuration, or manage an incident. Not for running PromQL, LogQL or TraceQL. Carries the provisioning model that makes API edits non-durable.

Kubernetes operations and troubleshooting for the homelab's two Talos clusters. Use for any question about the state, shape or health of a Kubernetes resource — pods, deployments, nodes, events, RBAC, CRDs, manifests, scaling and rollouts — and before any mutating kubectl-equivalent action. Carries the diagnostic order, the Talos-specific constraints, the patch-type rules for CRDs, and the GitOps safety rules for this estate.

Gateway API and kgateway ingress routing for the homelab. Use for north-south traffic — how a request from outside reaches a Service: Gateway and HTTPRoute status, listener and hostname conflicts, backend resolution, cross-namespace ReferenceGrants, TLS listeners and kgateway policies. Not for pod-to-pod connectivity. Carries the three GatewayClasses, the condition-reading order, and the quiet failures.

Loki logs and LogQL for the homelab. Use to find out why something failed, to search log text, to measure log volume or error rates, to extract structured fields, and to recover a trace_id for handoff to Tempo. Carries this deployment's label space, the writers that feed it, the retention and query limits that cause failures, and LogQL construction rules.

OpenTelemetry collector pipelines across the homelab's two clusters. Use when telemetry is missing, delayed or malformed — which is a different problem from a backend being unhealthy. Carries the two-collector topology, the ordered diagnostic path, the collector's own self-reported metrics and how to read them, and the naming constraint that breaks cross-cluster forwarding.

Prometheus metrics and PromQL for the homelab. Use when the question is numerical or about a threshold over time — request rates, error ratios, latency percentiles, saturation, capacity trends, restart counts, scrape health. Carries the two query paths and their endpoints, the data-model rules that make PromQL correct, and the query patterns for this estate.

The incident and triage playbook for the SRE lead: how to work a signal across specialist subagents, in what order, and how to correlate their findings into one answer. Use at the start of any incident or multi-signal investigation, and whenever a question needs more than one specialist. Carries the estate facts, the signal order, the correlation handoffs and the answer format.

Tempo distributed traces and TraceQL for the homelab. Use to find slow or failing requests, inspect spans, map service dependencies, expand a known trace_id, or answer where in a request path time is spent. Carries the datasource-proxy access pattern (there is no Tempo tool), TraceQL syntax, and the span-metrics shortcut in Prometheus.