[SPARK-38140] Desc column stats (min, max) for timestamp type is not consistent with the value due to time zone difference

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 3.1.2, 3.2.1
• Fix Version/s: 3.3.0
• Component/s: SQL
• Assignee: Zhenhua Wang
• Reporter: Zhenhua Wang
• Created: 08/Feb/22 12:48
• Updated: 21/Feb/22 07:40
• Resolved: 21/Feb/22 07:40

Description

Currently timestamp column's stats (min/max) are stored in UTC in metastore, and when desc its min/max column stats, they are also shown in UTC.

As a result, for users not in UTC, the column stats (shown to users) are not consistent with the actual value, which causes confusion.

For example:

spark-sql> create table tab_ts_master (ts timestamp) using parquet;

spark-sql> insert into tab_ts_master values make_timestamp(2022, 1, 1, 0, 0, 1.123456), make_timestamp(2022, 1, 3, 0, 0, 2.987654);

spark-sql> select * from tab_ts_master;
2022-01-01 00:00:01.123456
2022-01-03 00:00:02.987654

spark-sql> set spark.sql.session.timeZone;
spark.sql.session.timeZone	Asia/Shanghai

spark-sql> analyze table tab_ts_master compute statistics for all columns;

spark-sql> desc formatted tab_ts_master ts;
col_name	ts
data_type	timestamp
comment	NULL
min	2021-12-31 16:00:01.123456
max	2022-01-02 16:00:02.987654
num_nulls	0
distinct_count	2
avg_col_len	8
max_col_len	8
histogram	NULL

Issue Links

• links to: [Github] Pull Request #35440 (wzhfy)