Decision record ADR-007
One broker for every team. We propose a managed one.
Why we need to decide, the options we compared and the decision with its consequences.
Why decide now
Three brokers today, none of them owned
Today
Each team runs its own broker. Upgrades slip, and nobody is on call for them.
What we need
One supported broker for the order pipeline, with an owner and an SLO.
Options
A managed broker wins on effort and risk
Scored by the integration guild.
| Option | Operations effort | Throughput | Team skills | Lock-in | Cost |
|---|---|---|---|---|---|
| Self-hosted Kafka | Poor | Strong | Weak | Low | Medium |
| Managed Kafka | Good | Strong | Good | Medium | Medium |
| Cloud queue service | Good | Medium | Good | High | Low |
RecommendationManaged Kafka keeps our client libraries and removes the on-call burden.
Cost
Managed Kafka costs more to run, less to own
Monthly cost in k€, including people time.
| Self-hosted Kafka | 6 k€ | +9 Operations time |
|---|---|---|
| Managed Kafka | 11 k€ | +1.5 Operations time |
| Cloud queue service | 4 k€ | +2 Operations time |
Operations timeSource: vendor quotes and time tracking, Q3
ADR-007
Use managed Kafka for the order pipeline
Accepted · 2026-10-08
Context
Three self-run brokers, no owner and an event-driven order pipeline planned for Q1.
Decision
Adopt a managed Kafka service owned by the platform team; retire the other brokers by June.
Consequences
- +No broker upgrades or on-call for the teams
- +Same client libraries and tooling
- −Higher service bill than self-hosting
- −Topics and quotas to govern centrally
Next step
Approve ADR-007 today
The platform team starts the managed cluster on Monday and migrates the order topics first.
Sources
- 02Three brokers today, none of them ownedPlatform inventory, September
- 04Managed Kafka costs more to run, less to ownVendor quotes, Q3
- Team time tracking, Q3