serving.yaml

Declares how and where the output should be accessible. The developer states intent; the platform routes to the correct store via Dagster IO Managers.

S3 object storage (Ceph RGW, via Iceberg/Lakekeeper) is always written to regardless of endpoint configuration. It is the source of truth with full time-travel capability. The endpoints below are additional projections optimized for specific access patterns.

Example

apiVersion: akili/v1
kind: Serving

endpoints:
  - type: lookup
    description: Point lookups by outlet_id for the Portal
    config:
      index_columns:
        - outlet_id
        - sale_date

  - type: analytics
    description: Territory-level aggregations for executive dashboards

  - type: realtime
    description: Today's running totals, refreshed every 15 minutes
    config:
      key_template: "outlet:{outlet_id}:daily:{sale_date}"
      ttl: 24h
      include_columns:
        - outlet_id
        - sale_date
        - total_revenue
        - transaction_count

visualization:
  enabled: true
  dashboard_template: daily-sales-overview
  refresh_interval: 15m

Store Routing Table

Developer Declares	Platform Routes To	Mechanism	When To Use
`type: lookup`	PostgreSQL	IO Manager upserts via COPY, tenant RLS	Point lookups, Portal API, key-value access
`type: timeseries`	TimescaleDB	IO Manager inserts, auto continuous aggregates	Time-range queries, trend analysis
`type: analytics`	StarRocks	No-op — StarRocks federates to Iceberg on S3 directly	OLAP, GROUP BY, bulk scans, dashboards
`type: realtime`	Redis	IO Manager SET with TTL, tenant-prefixed keys	Sub-10ms reads, streaming hot path
(no endpoints)	S3 object storage only	Iceberg via Lakekeeper	Cold storage, time travel, source of truth

type: analytics is a no-op write. StarRocks queries Iceberg tables on S3 object storage directly via its External Catalog feature. The platform simply registers the table in StarRocks’ catalog. No data movement occurs.

Field Reference — Endpoint

Field	Type	Required	Default	Description
`type`	enum	Yes	—	`lookup`, `timeseries`, `analytics`, `realtime`
`description`	string	No	—	Documents intent for Portal and catalog
`config`	object	Varies	—	Type-specific configuration (see below)

Type-Specific Configuration

lookup

Field	Type	Required	Default	Description
`index_columns`	string[]	No	Primary key columns	Columns to index in PostgreSQL
`history_mode`	enum	No	`current`	`current` = latest state only (upsert). `type_2` = SCD Type 2 with effective_from/effective_to/is_current. `type_3` = previous-value columns.
`tracked_columns`	string[]	If history_mode is not `current`	—	Columns whose changes trigger a new historical record

timeseries

Field	Type	Required	Default	Description
`time_column`	string	No	First `date`/`timestamp` column	Column for hypertable partitioning
`chunk_interval`	string	No	`1 day`	TimescaleDB chunk size
`continuous_aggregates`	list	No	none	List of aggregate definitions (auto-created)

analytics

No configuration needed. The platform registers the Iceberg table in the StarRocks External Catalog automatically.

realtime

Field	Type	Required	Default	Description
`key_template`	string	Yes	—	Redis key pattern. Use `{column_name}` placeholders.
`ttl`	duration	No	`24h`	Time-to-live for Redis keys. Format: `1h`, `24h`, `7d`.
`include_columns`	string[]	No	All columns	Subset of columns to store (reduces memory)

Visualization

Optional Superset integration. When enabled, the platform creates or updates a Superset dataset, applies RLS filters, and optionally provisions a dashboard.

Field	Type	Required	Default	Description
`visualization.enabled`	bool	No	`false`	Enable Superset integration
`visualization.dashboard_template`	string	No	—	Platform-provided template name
`visualization.refresh_interval`	duration	No	—	Superset cache refresh interval