Skip to content

Two-Layer Deployment Model

floe deployment separates two concerns:

LayerOwnerK8s ResourceLifecycle
Platform ServicesPlatform TeamDeployment, StatefulSetLong-lived, upgraded
Pipeline JobsData Team (execution)JobRun-to-completion
+---------------------------------------------------------------------------+
| PLATFORM SERVICES (Layer 3 - Long-lived) |
| Deployed by: floe platform deploy |
| |
| +---------------------------------------------------------------------+ |
| | Orchestrator (Dagster): Catalog (Polaris): Semantic (Cube): | |
| | * webserver (Deployment) * server (Deployment) * server (Deploy) | |
| | * daemon (Deployment) * PostgreSQL * Redis | |
| | * PostgreSQL (StatefulSet) | |
| +---------------------------------------------------------------------+ |
| |
| +---------------------------------------------------------------------+ |
| | Observability: Storage: | |
| | * OTLP Collector * MinIO / S3 / GCS / ADLS | |
| | * Prometheus (optional) * OCI Registry (platform artifacts) | |
| | * Grafana (optional) | |
| +---------------------------------------------------------------------+ |
+----------------------------------+-----------------------------------------+
| Triggers
v
+---------------------------------------------------------------------------+
| PIPELINE JOBS (Layer 4 - Ephemeral) |
| Triggered by: Orchestrator on schedule/sensor |
| |
| +---------------------------------------------------------------------+ |
| | dbt run pods (K8s Job) Quality check pods Ingestion pods | |
| | * Run-to-completion (K8s Job) (K8s Job) | |
| | * Inherit platform config * Run-to-completion * dlt pipelines | |
| +---------------------------------------------------------------------+ |
+---------------------------------------------------------------------------+

Platform services run continuously and are managed by Platform Team.

CharacteristicValue
K8s ResourceDeployment, StatefulSet
LifecycleLong-lived, upgraded in place
StateStateful (databases, caches)
ScalingFixed replicas or HPA
Deploymentfloe platform deploy
UpgradesRolling updates, blue-green
OwnerPlatform Team

Pipeline jobs run to completion and are triggered by the orchestrator.

CharacteristicValue
K8s ResourceJob
LifecycleRun-to-completion
StateStateless
ScalingOne pod per execution
DeploymentTriggered by orchestrator
RetriesHandled by orchestrator
OwnerData Team (execution), Platform Team (infrastructure)