Date & Time

Overview

DATE keeps only calendar values, TIMESTAMP stores UTC internally but renders through the current session timezone, and TIMESTAMP_TZ preserves the original offset for auditing or replication scenarios.

Examples

DATE

CREATE TABLE events (event_date DATE);
INSERT INTO events VALUES ('2024-01-15'), ('2024-12-31');
SELECT * FROM events;

Result:

┌────────────┐
│ event_date │
├────────────┤
│ 2024-01-15 │
│ 2024-12-31 │
└────────────┘

TIMESTAMP

CREATE TABLE meetings (
  meeting_id INT,
  meeting_time TIMESTAMP
);

INSERT INTO meetings VALUES (1, '2024-01-15 14:00:00+08:00');

SETTINGS (timezone = 'UTC')
SELECT meeting_id, meeting_time FROM meetings;

SETTINGS (timezone = 'America/New_York')
SELECT meeting_id, meeting_time FROM meetings;

Result (timezone = 'UTC'):

┌────────────┬──────────────────────┐
│ meeting_id │ meeting_time         │
├────────────┼──────────────────────┤
│          1 │ 2024-01-15T06:00:00 │
└────────────┴──────────────────────┘

Result (timezone = 'America/New_York'):

┌────────────┬──────────────────────┐
│ meeting_id │ meeting_time         │
├────────────┼──────────────────────┤
│          1 │ 2024-01-15T01:00:00 │
└────────────┴──────────────────────┘

TIMESTAMP_TZ

CREATE TABLE system_logs (
  log_id INT,
  log_time TIMESTAMP_TZ
);

INSERT INTO system_logs VALUES
  (1, '2024-01-15 14:00:00+08:00'),
  (2, '2024-01-15 06:00:00+00:00'),
  (3, '2024-01-15 01:00:00-05:00');

SETTINGS (timezone = 'UTC')
SELECT log_id, TO_STRING(log_time) AS log_time FROM system_logs;

SETTINGS (timezone = 'Asia/Shanghai')
SELECT log_id, TO_STRING(log_time) AS log_time FROM system_logs;

Result (timezone = 'UTC'):

┌────────┬────────────────────────────────────────────┐
│ log_id │ log_time                                   │
├────────┼────────────────────────────────────────────┤
│      1 │ 2024-01-15 14:00:00.000000 +0800           │
│      2 │ 2024-01-15 06:00:00.000000 +0000           │
│      3 │ 2024-01-15 01:00:00.000000 -0500           │
└────────┴────────────────────────────────────────────┘

Result (timezone = 'Asia/Shanghai'):

┌────────┬────────────────────────────────────────────┐
│ log_id │ log_time                                   │
├────────┼────────────────────────────────────────────┤
│      1 │ 2024-01-15 14:00:00.000000 +0800           │
│      2 │ 2024-01-15 06:00:00.000000 +0000           │
│      3 │ 2024-01-15 01:00:00.000000 -0500           │
└────────┴────────────────────────────────────────────┘

The offset is part of the stored value, so the display never changes.

Choosing the Right Type

• Use DATE for calendar values without time of day.
• Use TIMESTAMP when different sessions should display the same moment in their local timezone.
• Use TIMESTAMP_TZ when you must keep the input offset for compliance or debugging.

Daylight Saving Time Adjustments

Enable enable_dst_hour_fix to make TiDB Cloud Lake automatically roll missing hours forward when daylight saving time skips part of the day.

SET enable_dst_hour_fix = 1;

SETTINGS (timezone = 'America/Toronto')
SELECT to_datetime('2024-03-10 02:01:00');

Result:

┌────────────────────────────────────┐
│ to_datetime('2024-03-10 02:01:00') │
├────────────────────────────────────┤
│ 2024-03-10T03:01:00                │
└────────────────────────────────────┘

Use SET enable_dst_hour_fix = 0 to return to the default behavior if you would rather raise errors for missing hours.

Handling Invalid Values

Dates outside the supported range automatically clamp to their minimum values.

SELECT
  ADD_DAYS(TO_DATE('9999-12-31'), 1)         AS overflow_date,
  SUBTRACT_MINUTES(TO_DATE('1000-01-01'), 1) AS underflow_timestamp;

Result:

┌───────────────┬──────────────────────────┐
│ overflow_date │ underflow_timestamp      │
├───────────────┼──────────────────────────┤
│ 0001-01-01    │ 0999-12-31T18:41:28      │
└───────────────┴──────────────────────────┘

The values wrap to the minimum representable date or timestamp instead of raising an error.

Formatting Date and Time

Functions such as TO_DATE and TO_TIMESTAMP accept explicit format strings. Control how they parse or render values by adjusting date_format_style and week_start.

Date Format Styles

Use date_format_style to switch between two format vocabularies:

• MySQL (default) uses specifiers like %Y, %m, %d.
• Oracle uses specifiers like YYYY, MM, DD to match ANSI-style masks.

-- Oracle-style mask
SETTINGS (date_format_style = 'Oracle')
SELECT to_string('2024-04-05'::DATE, 'YYYY-MM-DD');

Result (Oracle):

┌──────────────────────────────────────┐
│ to_string('2024-04-05'::DATE, 'YYYY-MM-DD') │
├──────────────────────────────────────┤
│ 2024-04-05                           │
└──────────────────────────────────────┘

-- Back to MySQL-style mask
SETTINGS (date_format_style = 'MySQL')
SELECT to_string('2024-04-05'::DATE, '%Y-%m-%d');

Result (MySQL):

┌──────────────────────────────────────┐
│ to_string('2024-04-05'::DATE, '%Y-%m-%d') │
├──────────────────────────────────────┤
│ 2024-04-05                           │
└──────────────────────────────────────┘

Week Start Configuration

week_start defines which day begins the week for functions such as DATE_TRUNC or TRUNC when using WEEK precision.

SETTINGS (week_start = 0) SELECT DATE_TRUNC(WEEK, to_date('2024-04-05')); -- Sunday
SETTINGS (week_start = 1) SELECT DATE_TRUNC(WEEK, to_date('2024-04-05')); -- Monday

Result (week_start = 0):

┌────────────────────────────────┐
│ DATE_TRUNC(WEEK, TO_DATE('2024-04-05')) │
├────────────────────────────────┤
│ 2024-03-31                     │
└────────────────────────────────┘

Result (week_start = 1):

┌────────────────────────────────┐
│ DATE_TRUNC(WEEK, TO_DATE('2024-04-05')) │
├────────────────────────────────┤
│ 2024-04-01                     │
└────────────────────────────────┘

MySQL Format Specifiers

To handle date and time formatting, TiDB Cloud Lake makes use of the chrono::format::strftime module, which is a standard module provided by the chrono library in Rust. This module enables precise control over the formatting of dates and times. The following content is excerpted from https://docs.rs/chrono/latest/chrono/format/strftime/index.html:

It is possible to override the default padding behavior of numeric specifiers %?. This is not allowed for other specifiers and will result in the BAD_FORMAT error.

• %C, %y: This is floor division, so 100 BCE (year number -99) will print -1 and 99 respectively.

• %U: Week 1 starts with the first Sunday in that year. It is possible to have week 0 for days before the first Sunday.

• %G, %g, %V: Week 1 is the first week with at least 4 days in that year. Week 0 does not exist, so this should be used with %G or %g.

• %S: It accounts for leap seconds, so 60 is possible.

• %f, %.f, %.3f, %.6f, %.9f, %3f, %6f, %9f:

  The default %f is right-aligned and always zero-padded to 9 digits for the compatibility with glibc and others, so it always counts the number of nanoseconds since the last whole second. E.g. 7ms after the last second will print 007000000, and parsing 7000000 will yield the same.

  The variant %.f is left-aligned and print 0, 3, 6 or 9 fractional digits according to the precision. E.g. 70ms after the last second under %.f will print .070 (note: not .07), and parsing .07, .070000 etc. will yield the same. Note that they can print or read nothing if the fractional part is zero or the next character is not ..

  The variant %.3f, %.6f and %.9f are left-aligned and print 3, 6 or 9 fractional digits according to the number preceding f. E.g. 70ms after the last second under %.3f will print .070 (note: not .07), and parsing .07, .070000 etc. will yield the same. Note that they can read nothing if the fractional part is zero or the next character is not . however will print with the specified length.

  The variant %3f, %6f and %9f are left-aligned and print 3, 6 or 9 fractional digits according to the number preceding f, but without the leading dot. E.g. 70ms after the last second under %3f will print 070 (note: not 07), and parsing 07, 070000 etc. will yield the same. Note that they can read nothing if the fractional part is zero.

• %Z: Offset will not be populated from the parsed data, nor will it be validated. Timezone is completely ignored. Similar to the glibc strptime treatment of this format code.

  It is not possible to reliably convert from an abbreviation to an offset, for example CDT can mean either Central Daylight Time (North America) or China Daylight Time.

• %+: Same as %Y-%m-%dT%H:%M:%S%.f%:z, i.e. 0, 3, 6 or 9 fractional digits for seconds and colons in the time zone offset.

  This format also supports having a Z or UTC in place of %:z. They are equivalent to +00:00.

  Note that all T, Z, and UTC are parsed case-insensitively.

  The typical strftime implementations have different (and locale-dependent) formats for this specifier. While Chrono's format for %+ is far more stable, it is best to avoid this specifier if you want to control the exact output.

• %s: This is not padded and can be negative. For the purpose of Chrono, it only accounts for non-leap seconds so it slightly differs from ISO C strftime behavior.

Oracle Format Specifiers

When date_format_style is set to 'Oracle', the following format specifiers are supported:

Examples comparing MySQL and Oracle format styles with the same data:

-- MySQL format style (default)
SELECT to_string('2022-12-25'::DATE, '%m/%d/%Y');

┌────────────────────────────────┐
│ to_string('2022-12-25', '%m/%d/%Y') │
├────────────────────────────────┤
│ 12/25/2022                     │
└────────────────────────────────┘

-- Oracle format style (same data as MySQL example above)
SETTINGS (date_format_style = 'Oracle')
SELECT to_string('2022-12-25'::DATE, 'MM/DD/YYYY');

┌────────────────────────────────┐
│ to_string('2022-12-25', 'MM/DD/YYYY') │
├────────────────────────────────┤
│ 12/25/2022                     │
└────────────────────────────────┘