Nordlicht Logistics GmbH

Situation

Nordlicht runs a warehouse-management platform that must remain available across three shifts. Prior to engagement, alerting was limited to a hosting-provider dashboard, PostgreSQL backups were untested, and a disk-full event on a Saturday night halted goods-out for 3 hours 40 minutes. The internal IT team of four could not staff nights or weekends.

Environment

Region and placement

eu-central-1 (Frankfurt) with a warm standby in eu-west-1

Hosts

8 production VMs (application, PostgreSQL primary/replica, Redis, jump host, reporting)

Operating systems and runtime

Ubuntu 22.04 LTS; PostgreSQL 15; Redis 7

Data stores

2.4 TB warehouse and customs data; 380 GB nightly change volume

Connectivity and access

Site-to-site IPsec from the Hamburg DC to the cloud VPC; no public database endpoints

Agreed scope

• 24/7 uptime monitoring on all eight nodes and the VPN tunnels
• Daily encrypted backups of PostgreSQL with quarterly restore tests
• Query and index review of the goods-out and inventory schemas
• Security baseline: SSH key-only access, fail2ban, unattended-upgrades, UFW

Work performed

• Deployed Prometheus exporters and Alertmanager with PagerDuty routing to Veltis on-call.
• Implemented pgBackRest to object storage with AES-256, 14-day retention, and a documented restore runbook.
• Added missing indexes on shipment_events and stock_moves; reduced p95 goods-out query time from 1.8 s to 210 ms.
• Closed inbound 5432/6379 from the public internet; restricted admin access to the jump host and a named engineer group.
• Ran a full restore drill on 12 March 2026; recovered the primary to a staging host in 47 minutes.

Measured results