Luca Padovan
EN IT
Back to projects

PROJECT CASE STUDY / qawk

GO · POSTGRESQL · OTA

Coordinating updates across a device fleet.

An OTA server that combines hawkBit compatibility with release channels, staged delivery, and multi-device orchestration.

01 / The problem

An update can succeed on one device and fail on another. For systems made of multiple connected components, release management needs to consider the whole system, not just individual downloads.

02 / My contribution

I developed qawk’s OTA server, including hawkBit-compatible APIs, release progression and multi-device update orchestration.

03 / The approach

qawk implements the Eclipse hawkBit 1.1.0 device and management APIs, then adds release channels, promotion gates, rollout waves, and coordinated rollback for multi-device systems.

04 / Engineering decisions

Engineering decisions

01

Verify compatibility with a real reference

Contract tests are replayed against a real hawkBit server to check API behavior, covering 16 device operations and 153 management operations.

02

Control how releases progress

Channels separate release stages; promotion gates and approvals control progression. Rollout waves halt when error thresholds are exceeded.

03

Treat connected components as a system

Orchestrated updates can roll back the whole multi-device system when a component fails. Prometheus metrics and OpenTelemetry tracing help operators observe delivery.

05 / Trade-offs & constraints

API compatibility adds a contract that must be tested against the reference server. Staged delivery and coordinated rollback add state and operational complexity in exchange for greater control over releases.

06 / What it delivers

What it delivers

  • hawkBit-compatible APIs with contract-test coverage.
  • Controlled promotion, rollout waves, and coordinated rollback.
  • Mutual TLS and Kubernetes manifests for deployment.

Reported load-test result: 10,000 devices polling every 30 seconds, 333 requests/s, and 2 ms p99. This describes the benchmark in my CV; production performance depends on infrastructure and workload.