Runtime Environments

The Problem

Traditional approaches to multi-environment deployment require either:

• Separate models per environment - duplication, drift, maintenance burden
• Configuration files - error-prone, not type-safe, easy to deploy wrong config
• Hard-coded environment logic - brittle, unmaintainable, scattered throughout code

None of these approaches scale. They create risk, slow down deployments, and make onboarding new environments painful.

The Solution: RuntimeEnvironments

RuntimeEnvironments (RTEs) use logical references in your model that are resolved at deployment time via EnvironmentMaps. One model, multiple environments, zero duplication.

How It Works

1. Define EnvironmentMaps

Teams define mappings from logical names to physical resources per environment:

team.add(EnvironmentMap(
    keyword="prod",
    dataContainers={
        frozenset(["customer_db"]): PostgresDatabase(
            "CustomerDB_Prod",
            hostPort=HostPortPair("prod-db.internal", 5432),
            databaseName="customer_production",
            productionStatus=ProductionStatus.PRODUCTION
        )
    },
    dtReleaseSelectors={
        "transformer_version": VersionPatternReleaseSelector(
            "V*.*.*-prod", ReleaseType.STABLE_ONLY
        )
    }
))

team.add(EnvironmentMap(
    keyword="dev",
    dataContainers={
        frozenset(["customer_db"]): PostgresDatabase(
            "CustomerDB_Dev",
            hostPort=HostPortPair("dev-db.internal", 5432),
            databaseName="customer_dev",
            productionStatus=ProductionStatus.NOT_PRODUCTION
        )
    },
    dtReleaseSelectors={
        "transformer_version": VersionPatternReleaseSelector(
            "V*.*.*-dev", ReleaseType.STABLE_ONLY
        )
    }
))

2. Use Logical References

Your model uses logical references - no environment-specific values:

Datastore(
    "CustomerData",
    capture_metadata=SQLSnapshotIngestion(
        EnvRefDataContainer("customer_db"),  # Logical reference
        CronTrigger("Hourly", "0 * * * *"),
        Credential("db_reader", CredentialType.USER_PASSWORD)
    ),
    datasets=[...]
)

DataTransformer(
    name="MaskCustomerPII",
    code=PythonCodeArtifact(
        VersionedRepository(
            GitHubRepository("myorg/transformers", "main"),
            EnvRefReleaseSelector("transformer_version")  # Logical reference
        )
    ),
    ...
)

3. Resolution at Deployment

When each RTE deploys, logical references resolve to environment-specific values:

Safety Features

The linter validates your model at PR time, preventing common mistakes:

Deployment Workflow

1. Develop - Teams develop and test with dev RTE (watches for -dev tags)
2. Tag for QA - Tag model with V1.2.3-qa for QA environment testing
3. QA validates - QA RTE picks up tag, deploys with qa EnvironmentMaps
4. Tag for Prod - After validation, tag V1.2.3-prod for production
5. Prod deploys - Prod RTE picks up tag, deploys with prod EnvironmentMaps

Each RTE's infrastructure updates independently within 5-15 minutes of tagging.

Benefits

Real-World Impact

• No environment-specific branches - single codebase, single PR workflow
• No manual config edits - all configuration in type-safe Python
• No accidental prod/dev mixing - validated at PR time
• Faster deployments - tag and go, no manual coordination