[IMPALA-10491] Impala parquet scanner should use writer.time.zone when converting Hive timestamps

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: Impala 3.4.0
• Component/s: Backend
• Labels: parquet
• Target Version: Product Backlog
• Epic Color: ghx-label-9
• Assignee: Unassigned
• Reporter: Tim Armstrong
• Created: 09/Feb/21 04:35
• Updated: 09/Feb/21 04:36

Description

IMPALA-8721 reports some issues with Hive 3 and timezone conversion.

HIVE-21290 fixed some of the issues, and also sets writer.time.zone in the Parquet metadata, which provides a better way to determine how the time zone was written. E.g.

tarmstrong@tarmstrong-Precision-7540:~/impala/impala$ hadoop jar ~/repos/parquet-mr/parquet-tools/target/parquet-tools-1.12.0-SNAPSHOT.jar meta /test-warehouse/asdfgh/000000_0
21/02/08 20:26:44 INFO hadoop.ParquetFileReader: Initiating action with parallelism: 5
21/02/08 20:26:44 INFO hadoop.ParquetFileReader: reading another 1 footers
21/02/08 20:26:44 INFO hadoop.ParquetFileReader: Initiating action with parallelism: 5
file:        hdfs://localhost:20500/test-warehouse/asdfgh/000000_0
creator:     parquet-mr version 1.10.99.7.2.7.0-44 (build 27344fd5fdaa371e364c604f471b340f8bcf8936)
extra:       writer.date.proleptic = false
extra:       writer.time.zone = America/Los_Angeles
extra:       writer.model.name = 3.1.3000.7.2.7.0-44

We should use this timezone when converting timestamps, I think either always or when convert_legacy_hive_parquet_utc_timestamps=true.

CC boroknagyz csringhofer

Issue Links

• relates to: HIVE-21290 Restore historical way of handling timestamps in Parquet while keeping the new semantics at the same time
• relates to: IMPALA-8721 Wrong result when Impala reads a Hive written parquet TimeStamp column