Backup Verification and Restore Testing#

An untested backup is not a backup. It is a file that might contain your data and might be restorable. Teams discover the difference during an actual incident, when the database backup turns out to be corrupted, the restore takes 6 hours instead of the expected 30 minutes, or the backup process silently stopped running three weeks ago.

Backup verification is the practice of regularly proving that your backups contain valid data and can be restored within your required RTO.

Decision-first: macOS and Linux node_exporter expose different metric names — write per-OS memory/disk expressions. The stock node dashboard hides Darwin on purpose. Scrape external hosts via ScrapeConfig + relabel job/instance. On a GB10, there are no GPU framebuffer or profiling metrics — read model footprint from system RAM.

Scope & freshness: kube-prometheus-stack + node_exporter + DCGM, macOS + Linux/GB10, as of 2026-05-25. Re-check the GB10 DCGM gaps after a DCGM/driver bump.

An agent is dispatched on a task it cannot complete. The spec is broken. The dependency is missing. The credentials are wrong. What happens next determines whether you have an autonomous fleet or a fleet that quietly fails.

The most common answer — instructing the agent in its prompt to “ask for help if stuck” — does not survive contact with production. Agents either keep grinding and produce broken work, or output text that looks like a question but never reaches a human, or politely “complete” the task by writing nothing and reporting success. None of these failure modes are visible from the outside until the dashboards have been lying for hours.

A backlog item flips to status='completed' in the database. The dashboard ticks up. The agent posts “PR ready for review” and walks away. Three hours later, a different agent notices the fleet is running yesterday’s binary. The PR was never reviewed. CI was red on main. No image got built. Nothing actually shipped.

This is the closed-loop problem. When an autonomous agent declares work complete, what does “complete” mean? In most agent fleets, it means the agent called the last tool in its own workflow — typically open_pr or its equivalent. That is not the same as “the change is live for users”, and the gap between the two is where state-of-record systematically lies.

Routing Tree#

Alertmanager receives alerts from Prometheus and decides where to send them based on a routing tree. Every alert enters at the root route and travels down the tree until it matches a child route. If no child matches, the root route’s receiver handles it.

# alertmanager.yml
global:
  resolve_timeout: 5m
  slack_api_url: "https://hooks.slack.com/services/T00/B00/xxx"
  pagerduty_url: "https://events.pagerduty.com/v2/enqueue"

route:
  receiver: "default-slack"
  group_by: ["alertname", "namespace"]
  group_wait: 30s
  group_interval: 5m
  repeat_interval: 4h
  routes:
    - match:
        severity: critical
      receiver: "pagerduty-oncall"
      group_wait: 10s
      repeat_interval: 1h
      routes:
        - match:
            team: database
          receiver: "pagerduty-dba"
    - match:
        severity: warning
      receiver: "team-slack"
      repeat_interval: 12h
    - match_re:
        namespace: "staging|dev"
      receiver: "dev-slack"
      repeat_interval: 24h

Timing parameters matter. group_wait is how long Alertmanager waits after receiving the first alert in a new group before sending the notification – this lets it batch related alerts together. group_interval is the minimum time before sending updates about a group that already fired. repeat_interval controls how often an unchanged active alert is re-sent.

Why Break Things on Purpose#

Production systems fail in ways that testing environments never reveal. A database connection pool exhaustion under load, a cascading timeout across three services, a DNS cache that masks a routing change until it expires – these failures only surface when real conditions collide in ways nobody predicted. Chaos engineering is the discipline of deliberately injecting failures into a system to discover weaknesses before they cause outages.

Choosing a Monitoring Stack#

Monitoring is not optional. Without metrics, you are guessing. The question is not whether to monitor but which stack to use. The right choice depends on your cost tolerance, operational capacity, retention requirements, and how much you value control versus convenience.

Decision Criteria#

Before comparing tools, clarify what matters to your organization:

• Cost model: Are you optimizing for infrastructure spend or engineering time? Self-managed tools cost less in licensing but more in operational hours. SaaS tools cost more in subscription fees but less in engineering effort.
• Operational burden: Who manages the monitoring system? Do you have an infrastructure team, or are developers responsible for everything?
• Data retention: Do you need metrics for 15 days, 90 days, or years? Long retention changes the equation significantly.
• Query capability: Does your team know PromQL? Do they need ad-hoc analysis or mostly pre-built dashboards?
• Alerting requirements: Simple threshold alerts, or complex multi-signal alerts with routing and escalation?
• Team expertise: An organization fluent in Prometheus wastes that investment by switching to Datadog. An organization with no Prometheus experience faces a learning curve.

Options at a Glance#

Prometheus + Grafana (Self-Managed)#

Prometheus is the de facto standard for Kubernetes metrics. It uses a pull-based model, scraping metrics from endpoints at configurable intervals, and stores time series data on local disk. Grafana provides visualization. Alertmanager handles alert routing.

When an Alert Should Fire but Does Not#

Silent alerts are the most dangerous failure mode in monitoring. The system appears healthy because no one is being paged, but the condition you intended to catch is actively occurring. Work through this checklist in order.

Step 1: Verify the Expression Returns Results#

Open the Prometheus UI at /graph and run the alert expression directly. If the expression returns empty, the alert cannot fire regardless of anything else.

From Empty Cluster to Production-Ready#

This is the definitive operational plan for taking a fresh Kubernetes cluster and making it production-ready. Each phase builds on the previous one, with verification steps between phases and rollback notes where applicable. An agent should be able to follow this sequence end-to-end.

Estimated timeline: 5 days for a single operator. Phases 1-2 are blocking prerequisites. Phases 3-6 can partially overlap.

Phase 1 – Foundation (Day 1)#

Everything else depends on a healthy cluster with proper namespacing and storage. Do not proceed until every verification step passes.

GPU and ML Workloads on Kubernetes#

Running GPU workloads on Kubernetes requires hardware-aware scheduling that the default scheduler does not provide out of the box. GPUs are expensive – an NVIDIA A100 node costs $3-12/hour on cloud providers – so efficient utilization matters far more than with CPU workloads. This article covers the full stack from device plugin installation through GPU sharing and monitoring.

The NVIDIA Device Plugin#

Kubernetes has no native understanding of GPUs. The NVIDIA device plugin bridges that gap by exposing GPUs as a schedulable resource (nvidia.com/gpu). Without it, the scheduler has no idea which nodes have GPUs or how many are available.