Documentation / Reference
Runtime groups and installed catalogs
Map logical workflow steps to installed runtime groups, memory slices, SQS queues and Lambda function ARNs.
Mapping phases to installed-groups.json
Workflows set a workflow-wide default_group (or defaultGroup) and optional per-phase group (or runtimeGroup) overrides. During validation and deployment, each group resolves against installed-groups.json.
json
{
"groups": {
"python-light": {
"runtime": "python3.13",
"workerMemoryMib": 512,
"handlerSliceMib": 256,
"timeoutSecs": 30,
"queueUrl": "https://sqs.us-east-1.amazonaws.com/123456789012/rippflow-prod-python-light",
"functionArn": "arn:aws:lambda:us-east-1:123456789012:function:rippflow-prod-python-light:4"
}
}
}Standard runtime group profiles
| Group family | Worker / slice memory | Timeout | Typical workload |
|---|---|---|---|
python-light, node-light, rust-light, java-light | 512 MiB worker (240–288 MiB slice) | 30s | Fast validation, routing, API transforms, and fan-out/fan-in steps. |
*-standard (Python, Node.js, Rust, Java) | 1,024–2,048 MiB worker (608–1,280 MiB slice) | 120s | Document parsing, batch aggregation, and moderate compute. |
*-heavy (Python, Node.js, Rust, Java) | 4,096 MiB worker (2,816–3,072 MiB slice) | 300s | Memory-intensive transforms; use external compute above 3 GiB. |
Admitted workflow versions pin their group memory and timeout immutably. Per-phase memory overrides (such as memoryMb) are rejected; profile locally with rippflow test --profile.
Verified sources
Checked 2026-11-05.
- Runtime group memory and slice contract 3a3be99 · main
- Installed runtime group catalog 7f5ea32 · main
- Runtime group pinning 04cb082 · main