Container Service
Category: Infrastructure
Node id: container-service
Manages the lifecycle of sidecar containers within a pipeline execution. Pulls, starts, health-checks, and exposes container endpoints to downstream nodes.
What it does
- Pull and start Docker containers as pipeline sidecar services
- TCP and HTTP health check with configurable timeout
- Automatic cleanup on execution completion
- Image allowlist enforcement for security
- Resource caps: 2048 MB memory, 2.0 CPU cores
- Max 3 containers per pipeline execution
- Pipeline inputs injected as HYDRA_INPUTS environment variable
- Interface resolution maps port names to full endpoint URLs
Use cases
- Run HAPI FHIR server as a pipeline-local service
- Spin up temporary databases for data transformation
- Launch ML model inference containers on-demand
- Run custom microservices as pipeline steps
Runtime
- Performance tier:
slow
Input ports
| Name | Type | Required | Description |
|---|---|---|---|
input | any | No | Optional data injected into the container as HYDRA_INPUTS env var |
Output ports
| Name | Type | Required | Description |
|---|---|---|---|
result | json | No | Container metadata: endpoints, interfaces, container_id, image, status |
success | boolean | No | True when container started and passed health check |
Settings
Basic
| Setting | Name | Type | Required | Default | Allowed values | Description |
|---|---|---|---|---|---|---|
| Container Image | image | string | Yes | Docker image to pull and run (must be in allowlist) | ||
| Port Mappings | ports | json | No | Array of port mappings: [{name, container_port, host_port, protocol}] | ||
| Environment Variables | env_vars | json | No | Key-value map of environment variables to inject | ||
| Command | cmd | json | No | Override container entrypoint with a command array |
Resources
| Setting | Name | Type | Required | Default | Allowed values | Description |
|---|---|---|---|---|---|---|
| Memory (MB) | memory_mb | number | No | 512 | min 64; max 2048 | Memory limit in megabytes (capped at 2048) |
| CPU Cores | cpu_cores | number | No | 1 | min 0.1; max 2 | CPU core limit (capped at 2.0) |
Health
| Setting | Name | Type | Required | Default | Allowed values | Description |
|---|---|---|---|---|---|---|
| Health Check | health_check | json | No | Health check config: {type, port, path, interval, retries}. Types: tcp, http | ||
| Health Check Type | health_check_type | enum | No | tcp | tcp, http | |
| Health Check Port | health_check_port | string | No | Port name or number to check | ||
| Health Check Path | health_check_path | string | No | / | HTTP path for health check (only for http type) | |
| Startup Timeout (seconds) | startup_timeout_seconds | number | No | 60 | min 5; max 300 | Maximum time to wait for the container to become healthy |
Auth
| Setting | Name | Type | Required | Default | Allowed values | Description |
|---|---|---|---|---|---|---|
| Registry Auth | registry_auth | json | No | Credentials for private registries: {server, username, password} |
Advanced
| Setting | Name | Type | Required | Default | Allowed values | Description |
|---|---|---|---|---|---|---|
| Pull Policy | pull_policy | enum | No | always | always, if-not-present, never | |
| Inject Pipeline Inputs | inject_inputs | enum | No | true | true, false | Whether to inject upstream data as HYDRA_INPUTS env var |
| Interfaces | interfaces | json | No | Map of interface names to port names for endpoint resolution | ||
| Volume Mounts | volumes | json | No | Array of volume mounts: [{name, type, host_path, container_path, read_only}] | ||
| Network Config | network | json | No | Network configuration: {mode, extra_hosts, dns} | ||
| Device Requests | devices | json | No | Array of device requests (e.g., GPU): [{type, count}] |
Errors
| Code | Description |
|---|---|
IMAGE_REQUIRED | image config field is missing or empty |
IMAGE_NOT_ALLOWED | image is not in the allowed registry list |
DOCKER_UNAVAILABLE | Docker daemon is not reachable |
PULL_FAILED | failed to pull the container image |
START_FAILED | container failed to start |
HEALTH_CHECK_FAILED | container did not become healthy within timeout |
MAX_CONTAINERS | execution already has the maximum number of containers |
Runtime notes
- result.endpoints maps port names to full URLs — wire these to downstream API nodes that need the container's address
- Containers are cleaned up automatically when the pipeline execution ends; max 3 containers per execution
- Image must match the allowlist (docker.io/library/, hapiproject/, ghcr.io/tiet-ai/, europe-west1-docker.pkg.dev/tietai-*) or the node will reject it
- Pipeline inputs are JSON-serialized into the HYDRA_INPUTS env var unless inject_inputs is set to 'false'
See Build any workflow for how to put nodes together and Ports, types and data flow for how port types connect.
Etwas unklar? Senden Sie Feedback zur Dokumentation per E-Mail..