← Back to projects

June 2026

Deployed on AWS EC2

CloudOps Observability Platform

A containerized observability platform built using Prometheus, Grafana, Loki, Promtail, Alertmanager, and Node Exporter to monitor, visualize, log, and alert on a Flask application. The complete monitoring stack is deployed and validated on AWS EC2 using Docker Compose.

DockerPrometheusGrafanaLokiAlertmanagerNode ExporterAWS EC2Bash
CloudOps Observability Platform cover

Deploying software to a cloud instance is only half the battle; knowing exactly how that software behaves under load, when it begins failing, and why it failed is what distinguishes hobby projects from production-grade operations.

The CloudOps Observability Platform is a containerized telemetry and log analysis platform engineered with Docker Compose and deployed on an AWS EC2 instance. It provides end-to-end operational visibility across system metrics, container states, centralized logs, and automated incident routing.


Architecture & Core Components

The stack is composed of decoupled micro-services coordinated through a unified Docker network:

  1. Prometheus: Time-series database scraping operational telemetry at 15-second intervals from infrastructure endpoints.
  2. Node Exporter: Host-level telemetry collector exposing CPU utilization, load averages, memory distribution, disk I/O, and network throughput directly to Prometheus.
  3. Grafana: Central visualization layer with dashboards providing at-a-glance insight into host health, container consumption, and network saturation.
  4. Loki & Promtail: Distributed log aggregation system. Promtail streams container and system /var/log/* files directly to Loki, allowing rapid log-to-metric correlation without leaving Grafana.
  5. Alertmanager: Deduplicates, groups, and routes alert events based on severity levels (Critical, Warning, Info) directly to incident channels.
terminal / snippet
       +---------------------------------------------+
       |                  AWS EC2                    |
       |                                             |
       |  +---------------+       +---------------+  |
       |  | Node Exporter |       |   Promtail    |  |
       |  +-------+-------+       +-------+-------+  |
       |          |                       |          |
       |          v                       v          |
       |  +---------------+       +---------------+  |
       |  |  Prometheus   |       |  Grafana Loki |  |
       |  +-------+-------+       +-------+-------+  |
       |          |                       |          |
       |          +-----------+-----------+          |
       |                      |                      |
       |                      v                      |
       |               +--------------+              |
       |               |   Grafana    |              |
       |               +--------------+              |
       |                      |                      |
       |  +-------------------+-------------------+  |
       |  | Alertmanager                          |  |
       |  +-------------------+-------------------+  |
       +----------------------|----------------------+
                              v
                   [ Alert Notifications ]

Failure Injection & Alert Validation

An alert that has never fired in a simulated failure is an alert that cannot be trusted in production. As part of this platform's hardening, I established a deliberate failure testing procedure:

1. Memory Exhaustion Simulation

Simulated high memory pressure using dummy buffer allocations to force host RAM beyond 80%:

terminal / snippet
- alert: HighMemoryUsage
  expr: (node_memory_MemTotal_bytes - node_memory_MemAvailable_bytes) / node_memory_MemTotal_bytes * 100 > 80
  for: 2m
  labels:
    severity: warning
  annotations:
    summary: "Host RAM utilization exceeded 80%"
    description: "Instance {{ $labels.instance }} has exceeded threshold for > 2m."

2. Disk Space Saturation

Generated synthetic disk saturation on temporary partitions to confirm that Alertmanager triggered critical disk warnings before root volume read-only triggers locked the system.

3. Log Correlation Verification

When deliberate failures were injected, Promtail and Loki allowed instantaneous transition from Grafana metric spikes directly to the exact stdout/stderr logs produced by the offending container.


Operational Outcomes

  • Single-Command Bootstrap: The entire environment spins up reproducibly via docker compose up -d with predefined data persistence volumes.
  • Faster Incident Triage: Metric-to-log jumping cut theoretical Mean Time to Detect (MTTD) and Root Cause Analysis (RCA) time significantly.
  • Zero-Guesswork Infrastructure: System administrators have definitive real-time data on resource saturation rather than relying on delayed user reports.