Hot & cold tiers

The hot/cold split is how OrcheStack fits a full data stack inside a single SME-scale host without running services that aren't being used.

Definition

Every OrcheStack-managed service falls into one of three classes:

Assignment rules

The principle is simple: services that need to respond to external traffic at any moment are hot; services that respond to triggered work are cold.

Cold-service lifecycle

Cold services transition through four states: STOPPEDSTARTINGRUNNINGSTOPPING. Transitions are driven by three triggers:

  1. Airflow DAG schedule — e.g. the nightly ETL at 02:00 starts Airbyte + workers + dbt + Elementary in sequence.
  2. Dashboard manual button — e.g. "Run dbt now" or "Open pgAdmin".
  3. Upstream dependency — e.g. when dbt finishes, Elementary auto-starts to record test results.

A cold service only stops when its reference-counted session count hits zero AND the idle timeout has elapsed. See Service sessions for the multi-user safety this provides.

Resource comparison

StateServices runningEst. RAM
Base installProxy, Front, Dashboard, PostgreSQL, Orchestrator~2 GB
Configured (idle)Base + hot tier (MinIO, Metabase, Airflow scheduler)~4–5 GB
Active pipelineConfigured + cold tier during a run~7–10 GB peak