Workflow engine landscape — the worksheet
The two axes (answer these before any feature matrix)
- Author axis — who must read or author the flow? Business, compliance, or ops reads it → model-first (BPMN family). Only engineers read it → code-first (durable execution) is simpler and honest.
- State axis — where may state live? State must stay in your existing RDBMS (audit, ops, data residency) → Flowable / Camunda 7 class. A separate platform is acceptable → Camunda 8 / Temporal enter the shortlist.
The five-way table
| Flow authored as | State lives in | Human tasks | Decisions | Ops model | |
|---|---|---|---|---|---|
| Flowable OSS | BPMN/CMMN models | your RDBMS (embedded or standalone) | native (Phase 3) | DMN native | you run it; it’s a library or one service |
| Camunda 7 | BPMN models | your RDBMS, embedded | native | DMN native | same class; EOL path — legacy only |
| Camunda 8 | BPMN models | Zeebe cluster (SaaS/self-hosted) | native | DMN native | separate stateful platform |
| Temporal | code (Java/Go/TS/Py) | Temporal cluster/cloud | build your own | build your own | separate stateful platform |
| DIY | status columns + code | your tables | build your own | code | none today, all of it later |
Shortlist logic
- Model-first + your-DB → Flowable OSS (this course’s cell).
- Model-first + platform-OK + very high throughput → Camunda 8.
- Code-first + platform-OK → Temporal.
- Neither axis matters (tiny, stable, automated) → DIY status+queue, guilt-free.
Questions that cut through vendor decks
- Ask for the audit answer: “which rules version decided case X?” (Phase 8’s sentence.) Who produces it, and from which store?
- What is the second stateful system you’d now operate, and what is its backup/DR story?
- Where do human tasks, forms, and candidate groups come from — native, or DIY?
- What happens to in-flight instances on deploy? If the answer isn’t Phase 8’s three rules, dig deeper.
- Which parts are open source today, and what moved editions in the last three years?