Cloudera Docs

Support for Hive JDBC connector

Cloudera Data Warehouse offers support for a read-only Hive JDBC connector designed to facilitate SELECT operations on Hive sources. It optimizes query execution performance by pushing down filters, limits, expressions, and aggregates directly to the source database, taking full advantage of the underlying Hive engine's capabilities.

Key features and capabilities

Supported data types

The connector supports precise schema discovery and runtime query handling through type-safe mappings between the Trino and Hive data layers:

  • Numeric: TINYINT, SMALLINT, INTEGER, BIGINT, FLOAT, REAL, DOUBLE, and DECIMAL(p, s) or NUMERIC(p, s)
  • Character: CHAR, VARCHAR (bounded), and STRING (unbounded)
  • Date or Time: DATE and TIMESTAMP
  • Complex: ARRAY, MAP, and STRUCT (unsupported)
Constraints related to DATE and TIMESTAMP data types

The temporal data types have specific timezone and precision semantics that differ across engines, requiring the connector to explicitly use custom logic for all date and timestamp mappings. The core issue with legacy JDBC drivers is the silent application of the JVM's local timezone when calling ResultSet.getDate(). As a result, pushing down a condition directly to the source database can yield incorrect or inconsistent query results, shifting historical dates by a full day and causing mismatched data issues during predicate pushdowns.

Workaround (Date): The connector bypasses java.sql.Date entirely, opting to read the date as a raw String from Hive (example, 2026-06-11). It then applies a strict DateTimeFormatter and converts it directly to Trino epoch days, guaranteeing strict correctness and cross-engine uniformity during data extraction.

Workaround (Timestamp): The connector bypasses standard driver-level timestamp extraction, opting to read timestamps into LocalDateTime. It then manually rounds them to match Trino's MAX_SHORT_PRECISION, guaranteeing strict correctness and cross-engine uniformity during data extraction.

Query optimization
To maximize cluster execution efficiency and balance engine workloads, query operations are evaluated and selectively pushed down to the source database based on the verified Trino capabilities matrix:

Full push down:

The connector supports pushdown for a number of operations:

  • Predicate pushdown
  • Dynamic filter
  • Limit pushdown
  • Join pushdown

Aggregate pushdown for the following functions:

  • avg()
  • count(), also count(distinct x)
  • max()
  • min()
  • sum()

Additionally, pushdown is supported for advanced statistical metrics with the following functions:

  • stddev()
  • variance()
  • covariance()
  • correlation()
  • Regression

Not push down:

The connector does not support pushdown for the following:

  • Top-N pushdown
  • Join pushdown (IS DISTINCT FROM)
  • Dereference pushdown
Limitations and unsupported operations

This connector is strictly read-only. Attempting any of the following will trigger a NOT_SUPPORTED exception:

  • DML operations: INSERT UPDATE, DELETE and MERGE are blocked.
  • DDL operations: CREATE SCHEMA, CREATE TABLE, DROP SCHEMA, and RENAME SCHEMA are not permitted.
  • JDBC connection pooling: Configuration and reuse metrics for integrated connection pooling are currently unsupported for the Hive JDBC connector.