Network & cloud operations platform: one place to see, act on and report every operational signal
Case study by Hasan-Uz-Zaman Ashik: a Go and React platform giving NOC and cloud operations teams one place for topology, alarms, Alertmanager alerts, OpenStack and server health, IP tracking, ticketing and reporting.
Before and after
- Before: Status spread across separate NMS screens, Grafana boards and device CLIs. After: One portal with topology, alarms, cloud and server health side by side.
- Before: Inventory, IP allocations and certificates tracked in spreadsheets. After: A shared source of truth with a history of every change and who made it.
- Before: Alerts arriving as repeated notifications, with no record of who acted. After: One record per alert, with acknowledge, clear and comment trails.
- Before: Requests and incidents handled in chat and email, easy to lose. After: Tickets with clear ownership, assignment, attachments and analytics.
- Before: Hours at month end rebuilding the performance report by hand. After: A one-click Excel export in the team's own report layout.
Modules
Network topology & link utilization
Problem: During an outage, engineers had to piece together which devices connect where from CLIs and diagrams that were often out of date, and congested links went unnoticed.
What it does: Builds a map per region from discovered neighbor data and colors each link by live utilization from time-series telemetry, weathermap style, with history per link.
What the team gets: Engineers see the blast radius of a failure and congested paths at a glance instead of tracing them by hand.
Device & link alarms
Problem: Down devices and links were visible only in separate tools, and there was no shared record of who had acknowledged or cleared what.
What it does: Per-site up/down summaries, an active alarm list with acknowledge, clear and comments, full device and link alarm history, and Excel export.
What the team gets: Shift handovers start from the same picture, and reports for management no longer need to be compiled manually.
Cloud alert management
Problem: Prometheus alerts arrived as a stream of notifications. The same alert repeated, and its history was hard to follow after the fact.
What it does: Receives Alertmanager webhooks, merges repeats into one alert by fingerprint with an event timeline, and filters by environment, cluster and severity.
What the team gets: One alert means one record, so noise drops and post-incident reviews have a complete timeline.
Server health & utilization
Problem: Threshold alarms fired for servers under planned maintenance, while servers that silently stopped reporting were easy to miss.
What it does: CPU, memory and temperature alarms with an acknowledge lifecycle, explicit 'no data' detection, and maintenance exclusions that automatically skip related alarms, with an audit trail.
What the team gets: Fewer false alarms, and nothing goes quiet without someone noticing.
OpenStack & VM placement
Problem: Before hardware maintenance it was slow to answer which VMs and projects run on a host, or where a VM had moved.
What it does: Overview of projects, public IPs and external networks, per-host VM and project counts by availability zone, and VM movement history.
What the team gets: The team can assess impact before maintenance and notify only the tenants who are affected.
IP address tracker
Problem: Public IP allocations lived in spreadsheets, which led to stale records and the risk of double allocation.
What it does: Prefixes and pools with a status and usage type for every address, plus a change history showing who changed what and when.
What the team gets: A single source of truth for addresses, with an audit trail instead of guesswork.
SSL certificate monitor
Problem: An expired certificate can take a service down, and expiry dates were easy to lose track of.
What it does: A central list of certificates with expiry status, so renewals are visible well ahead of time.
What the team gets: Certificates get renewed before they become an outage.
Inventory
Problem: Sites, devices, power supplies and services were recorded in different files maintained by different people.
What it does: A searchable inventory with add and update workflows, asset distribution views and Excel export.
What the team gets: Everyone works from the same asset data, and other modules build on it.
Ticketing
Problem: Requests and incidents came through chat and email, so ownership was unclear and work was hard to measure.
What it does: Tickets with a shared pool and assignment, severity, updates, secure attachments, role-based permissions, edit locks so two engineers can't overwrite each other, a full audit trail and analytics.
What the team gets: Clear ownership and accountability, and visibility of workload across the team.
Work log & monthly report
Problem: Engineers spent time at the end of every month reconstructing what they had worked on for the performance report.
What it does: A to-do list with topics and start/pause timers, a timesheet view, and a one-click Excel export that matches the team's report format.
What the team gets: The monthly report is generated from day-to-day work instead of being rewritten from memory.
Engineering decisions
- Transactional outbox for ticket events: Ticket changes and their notifications are written in the same transaction, and a background worker sends them using FOR UPDATE SKIP LOCKED. Events aren't lost if a send fails, and several workers never pick up the same one.
- Partitioned audit tables: Ticket updates and audit events grow without limit. Partitioning keeps queries fast and old data easy to archive, and partitions are created automatically at startup.
- Edit locks with expiry and version checks: Two engineers working the same incident can't silently overwrite each other, and a lock left by a closed browser expires on its own.
- Single sign-on with tokens verified locally: Users log in once through the company identity provider. The API verifies tokens against the provider's published keys and enforces roles on every request.
- Batching queries and per-request timeouts: Topology loads in one query instead of one per device, and every request carries a timeout so a slow database call can't tie up the server.
- Signed, short-lived links for attachments: Files live in object storage and are downloaded through temporary signed URLs, so attachments are never publicly reachable.
Stack
Go, Gin, GORM, PostgreSQL, InfluxDB, S3 object storage, Prometheus Alertmanager, OpenStack, Keycloak / OIDC, RBAC, React, TypeScript, TanStack Query, Zustand, React Hook Form + Zod, vis-network, JointJS, Recharts, Chakra UI, Excel export