Skip to main content
DevOps // SecOps

Know it's broken before your customers tweet about it.

Metrics, logs, and traces with alerts that mean something, plus the security hardening that keeps the alerts quiet in the first place.

~/anvexis/devops/secops-monitoring

$ anvexis tldr

Alert on symptoms users feel. Dashboard everything else. Patch on schedule.

$ anvexis facts

Metrics
Prometheus · Grafana
Logs
Loki · ELK
Security
Hardening · scanning

Alerting anti-patterns

Myth: Alert on CPU over 80%.

Reality: Alert on what users feel: errors, latency, and failed jobs. CPU belongs on a dashboard.

Myth: More alerts mean more safety.

Reality: Noisy alerts get ignored. Every page should be actionable.

Myth: Logs are enough.

Reality: Logs say what happened; metrics say how often; traces say where. You need all three.

Hardening baseline

  • SSH keys only, no password logins
  • Automatic OS security updates
  • Firewall rules with deny by default
  • TLS everywhere, HSTS at the edge
  • Dependency and image vulnerability scanning
  • Centralized, tamper-resistant audit logs

Why monitoring and security belong together

Most outages and many security incidents are discovered the same way: someone outside the company notices first. A customer can't log in, an integration stops syncing, or an unusual login appears in an audit long after the fact. Good monitoring and good security share one goal: find out what is happening in your systems before anyone else does.

Monitoring answers whether the system is healthy: are pages fast, are jobs running, is the disk filling up? Security monitoring answers whether anything suspicious is happening: failed logins, unexpected processes, changes to permissions. Both rely on the same foundations, which are metrics, logs, traces, and alerts that reach a person who can act.

We build that visibility in proportion to your size. A small team needs a handful of meaningful alerts and a clear dashboard, not a wall of graphs nobody reads.

What we watch

SignalExamplesWhy it matters
AvailabilityUptime checks, error rates, failed jobsKnow when users are affected
PerformanceResponse times, slow queries, worker saturationCatch slowdowns before complaints
CapacityDisk, memory, connections, queue lengthFix limits before they become outages
Security eventsFailed logins, privilege changes, new admin usersSpot misuse early
BackupsLast successful backup, restore test resultsKnow recovery is possible
Certificates and domainsExpiry datesAvoid self-inflicted outages

Tools and practices

Metrics and dashboards

Prometheus and Grafana, or your cloud's native tools, with dashboards built for decisions.

Central logging

Application, system, and access logs in one searchable place with retention rules.

Actionable alerts

Few alerts, each with a clear owner and a runbook link.

Hardening

Patching, firewall rules, SSH and admin access controls, and secure configuration baselines.

Access reviews

Regular checks of who has access to what, with removals tracked.

Vulnerability management

Scheduled scans of dependencies, images, and hosts, with fixes prioritized by risk.

Getting to good visibility

  1. Stage 01

    Baseline

    Week 1

    Review existing monitoring, logs, and recent incidents.

  2. Stage 02

    Instrument

    Week 1–2

    Add metrics, logs, and health checks where they are missing.

  3. Stage 03

    Alert

    Week 2

    Define a small set of alerts with owners and runbooks.

  4. Stage 04

    Harden

    Week 2–3

    Apply the security baseline and fix the highest-risk findings.

  5. Stage 05

    Practice

    Week 3–4

    Run an incident drill and refine alerts and runbooks.

Signs your alerting needs work

  • Alerts arrive that nobody acts on
  • Customers report problems before alerts do
  • Nobody is sure who responds at night
  • Logs are on each server and disappear when it does

Common pitfalls we help you avoid

The first pitfall is monitoring everything and alerting on nothing useful. Hundreds of graphs and noisy alerts train people to ignore them, so we start with the few signals that show users are affected.

The second is keeping logs only on the servers. When a server fails or is compromised, its local logs go with it, so logs are shipped to a central place with access controls.

The third is hardening once and forgetting. New packages, users, and services appear over time, so baselines are checked automatically and access is reviewed on a schedule.

What we measure

MeasureWhy it matters
Time to detectHow quickly problems are noticed
Time to resolveHow quickly they are fixed
Alert noiseShare of alerts that needed no action
Open vulnerabilitiesKnown issues by severity and age

Who does what

We handle

  • Monitoring, logging, and alerting setup
  • Security baseline and scans
  • Runbooks and incident drills

Your team handles

  • Owning alerts during agreed hours
  • Approving access and policy changes
  • Communicating with users during incidents

A worked example

Picture an online business that hears about outages from customers on social media. Uptime checks, error-rate alerts, and a dashboard for checkout performance now go to an on-call channel, and logs from every service flow to one place.

When a payment provider slows down, the team sees it in minutes, follows the runbook, and posts a status update before customers notice. A monthly review removes noisy alerts and adds checks for anything that slipped through.

Before you start

  • A list of critical user journeys
  • Who responds to alerts, and when
  • Where logs are stored today
  • Recent incidents and what caused them
  • Compliance requirements for log retention

What success looks like

Success is hearing about problems from alerts rather than customers, with a clear owner and a runbook for each one. Alert noise falls over time, known vulnerabilities are fixed in order of risk, and access reviews happen on schedule rather than after an incident.

Monitoring principles from Google's SRE book

Google's Site Reliability Engineering book names four golden signals of monitoring: latency, traffic, errors, and saturationSource [1]. It also argues that every page should be actionable and that it's better to spend effort catching symptoms users feel than every possible causeSource [1]. Alert fatigue comes from ignoring both.

Security monitoring extends to how software is built. GitHub's hardening guidance, for example, warns against plain-text secrets in workflows and recommends minimum token permissions and pinned actionsSource [2]. A compromised pipeline is a direct route to production, so we watch it like any other critical system.

Frequently asked questions

Which monitoring stack do you use?
Usually Prometheus and Grafana, or your cloud's native tools if you prefer.
Can you monitor Odoo specifically?
Yes: worker usage, request latency, cron jobs, and database health.
Do you run penetration tests?
We harden and scan. Formal penetration tests are best done by an independent specialist.
Who receives alerts?
Your team, ours under a support plan, or both, routed by severity.
What is FinOps monitoring?
Tracking cloud spend alongside usage so cost spikes get noticed like errors do.
Do you provide 24/7 monitoring?
We set up monitoring and alerting, and can respond to alerts under a support plan. Coverage hours are agreed in the plan.
Which tools do you use?
Commonly Prometheus, Grafana, Loki, and OpenTelemetry, or your cloud provider's tools. We fit into what you have where possible.
How do you avoid alert fatigue?
Every alert must be actionable and owned. Noisy alerts are tuned or removed during regular reviews.
Can monitoring cover external services we depend on?
Yes: payment gateways, email providers, and APIs can be checked from outside and included in dashboards.
Do you write incident runbooks?
Yes, for the alerts we set up, with clear first steps, escalation paths, and contacts.
Can you help after a security incident?
We help contain, recover, and harden. For forensic investigation we work alongside specialist responders.
Can you integrate alerts with our chat tools?
Yes: Slack, Microsoft Teams, email, or paging tools, with routing by severity and time of day.
How long are logs kept?
As long as your policy and regulations require, with cheaper storage tiers for older logs.
Can you monitor servers we don't host with you?
Yes. Lightweight agents or exporters send metrics and logs from any server you control, on-premise or in another cloud, into the same dashboards and alerts.
Do you check SSL certificates and domains?
Yes. Certificate and domain expiry dates are monitored, with alerts weeks before anything lapses.
Can we get a monthly security and health report?
Yes. A short monthly report can summarize incidents, alerts, patching status, open vulnerabilities, and access changes, with recommended next steps.
Do you monitor scheduled jobs and backups?
Yes. Missed or failed cron jobs and backups raise alerts, so silent failures are caught the same day.
Why do we get so many useless alerts?
Usually alerts on causes rather than user-facing symptoms, and alerts nobody can act on. The SRE guidance is to make every page actionableSource [1].
What should we monitor first?
The four golden signals for each user-facing service: latency, traffic, errors, and saturationSource [1].