The Apache Pinot Native Data Connector allows you to connect to a Pinot cluster (self-hosted, or StarTree Cloud) and get an instant GraphQL API on top of your Pinot data.
This connector is built on the Rust Data Connector SDK from the Data Connector Hub and implements the Data Connector Spec.
In order to use this connector you will need:
- Access to an Apache Pinot cluster's Controller and Broker REST APIs (self-hosted, or StarTree Cloud)
- A Hasura CLI session
This connector should be used via the Hasura DDN CLI.
See the configuration instructions for details on configuration.json, environment variables, and native queries.
See the development instructions for how to build, test, and run this connector locally.
See which Hasura OpenDD Spec features are supported by the connector here.
View other documentation for the Pinot connector here.
This connector is intentionally narrow in scope for v1. The limitations below are deliberate design decisions, not gaps to be closed opportunistically — read the reasoning, not just the "no":
- No relationships / joins. Apache Pinot's SQL engine has limited, non-correlated
JOINsupport and no equivalent of the correlated subqueries a single-statement join resolution strategy would need to resolve NDC relationships inside one query. Rather than emulate relationships with fragile N+1 query stitching, this connector declares no relationship capability. Model cross-collection joins as Hasura engine-level remote joins across separate connector queries instead. - No mutations. Pinot is a read-only OLAP query engine from this connector's perspective. Data gets into Pinot tables through ingestion pipelines (batch or streaming), not through SQL
INSERT/UPDATE/DELETE— so this connector declares no mutation capability. - No nested-field filtering, ordering, or aggregation. Selecting, filtering, sorting, or aggregating by fields nested inside a struct/object column isn't supported. Pinot's own complex types (
STRUCT,MAP,LIST) are surfaced as opaqueJSONscalars rather than structured object types, so there's nothing to nest into in the first place. TIMESTAMPcolumns are opaque strings. Pinot serializesTIMESTAMPvalues in query results asyyyy-MM-dd HH:mm:ss.S-style strings by default, not ISO-8601. To avoid implying a formatting guarantee this connector can't back,TIMESTAMPcolumns are exposed with a plainStringtype representation rather than NDC'sTimestamprepresentation.BOOLEANcolumns are coerced back to real booleans. Pinot's query result serialization has no native JSON boolean forBOOLEANcolumns — it returns the strings"true"/"false". This connector coerces those strings back into real JSONtrue/falsevalues so callers see the type they declared, rather than passing the string through as-is.- Native queries are a lightweight escape hatch, not a SQL dialect. Native queries are plain
.sqlfiles with{{argument_name}}placeholders that get substituted with escaped SQL literals at query time. There's no query parsing or type inference — you declare the argument types and the result columns yourself inconfiguration.json.
See docs/usage.md for the full feature-support checklist.
Licensed under the Apache License, Version 2.0.