MDEV-15621 Auto add RANGE COLUMNS partitions by interval

[mariadb-YuchenPei]

/gemini review

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[gemini-code-assist]

Code Review

This pull request introduces Range Interval Partitioning for RANGE COLUMNS tables, allowing for the automatic creation of partitions based on a specified temporal interval during DML operations. The implementation involves new logic for calculating and adding partitions during the table opening process, along with parser updates to support the INTERVAL syntax. Critical feedback was provided regarding a potential infinite loop if the interval is zero, a null pointer dereference risk when encountering MAXVALUE partitions, and the omission of error checking for partition count calculations. Furthermore, the reviewer suggested using session timezones for temporal calculations and identified risky bitmap manipulation that should be replaced with a safer approach.

[mariadb-YuchenPei]

/gemini review

[gemini-code-assist]

Code Review

This pull request introduces range interval partitioning for RANGE COLUMNS tables, enabling the automatic creation of new partitions based on a specified time interval during DML operations such as INSERT, UPDATE, and LOAD DATA. The implementation includes parser support for the INTERVAL syntax, logic to calculate and add new partitions dynamically, and comprehensive test cases. Review feedback identifies a critical issue where newly added partitions are not correctly marked with the PART_TO_BE_ADDED state, which would cause them to be skipped during constant initialization. Additionally, several improvements were suggested regarding type safety—specifically avoiding unsafe pointer casts for warning variables—and ensuring the consistent use of boolean literals instead of integers for boolean types.

[mariadb-YuchenPei]

/gemini review

[gemini-code-assist]

Code Review

This pull request introduces automatic partition creation for RANGE COLUMNS partitioning using an INTERVAL clause. The implementation includes parser updates to support the new syntax, logic within the table opening process to trigger partition creation during DML operations, and internal ALTER TABLE mechanics to append new partitions. The review feedback identifies a critical bug where partitioning fields are not correctly initialized during the auto-creation process, a design limitation where partitions are only created up to the current system time, and the need to support legacy DATE and DATETIME types for broader compatibility.

[gemini-code-assist]

This TODO should be addressed to ensure that the thread state is correctly initialized and doesn't trigger MSAN warnings during the internal partition creation process.

[mariadb-YuchenPei]

/gemini review

[gemini-code-assist]

Code Review

This pull request implements automatic range interval partitioning (similar to Oracle's INTERVAL partitioning) for DATE, DATETIME, and TIMESTAMP columns. It introduces new syntax parsing for INTERVAL with functions like NUMTODSINTERVAL and NUMTOYMINTERVAL, and adds support for automatically creating new partitions during DML operations (such as INSERT, LOAD DATA, and UPDATE) when data falls outside existing partition ranges. The reviewer identified several critical and high-severity issues in the implementation. These include an early return bug in check_range_interval_constants that leaves the table's write_set corrupted, missing properties when copying part_info during auto-partitioning, memory leaks from standard new allocations instead of using mem_root, and unrestored thread state (thd->work_part_info) on function exit. Additionally, the reviewer suggested adding defensive checks for empty partition lists, guarding against infinite loops, ensuring exact string length matches for interval units, checking the return value of date_add_interval, preventing potential null pointer dereferences in the parser, and optimizing partition suffix naming to avoid O(N^2) scans.

[gemini-code-assist]

Efficiency optimization: suffix is initialized to 0 for range interval partitions, which causes an $O(N^2)$ scan of all existing partitions when adding multiple partitions. Initializing it to table->part_info->partitions.elements is much more efficient.

  uint32 suffix= part_type == VERSIONING_PARTITION ? num_parts - 1 : table->part_info->partitions.elements;
  DBUG_ASSERT(suffix > 0 || part_type != VERSIONING_PARTITION);

[mariadb-YuchenPei]

Initialise to table->part_info->partitions.elements does not help because we don't know what are the used suffixes yet. Added a TODO to consider using a hash

[mariadb-YuchenPei]

/gemini review

[gemini-code-assist]

Code Review

This pull request implements automatic range interval partitioning for DATE, DATETIME, and TIMESTAMP columns, allowing partitions to be dynamically created during DML operations. It also adds support for Oracle-compatible interval functions like NUMTODSINTERVAL and NUMTOYMINTERVAL. The review feedback highlights critical safety improvements, including preventing potential null pointer dereferences when resolving partition expressions or field arrays, avoiding unsigned underflow in partition count calculations, and refactoring an empty conditional statement to improve code readability and maintainability.

[mariadb-YuchenPei]

/gemini review

[gemini-code-assist]

Code Review

This pull request implements range interval partitioning for RANGE COLUMNS (supporting DATE, DATETIME, and TIMESTAMP columns), enabling automatic partition creation during DML operations when inserted values exceed existing ranges. It also adds Oracle-style interval syntax support. The review feedback identifies several critical issues, including a missing interval copy during partition auto-creation that would result in incorrect ranges, missing error reporting on date overflow failures, and unsafe usage of atoi on non-null-terminated LEX_CSTRING tokens in the parser. Additionally, minor improvements are suggested to optimize redundant checks and correct misleading comments.

[gemini-code-assist]

If date_add_interval fails (e.g., due to date overflow), the function returns true but does not set any error in the diagnostics area. This can lead to a silent failure or an assertion failure in open_table because an error is expected when returning failure. We should explicitly report ER_DATA_OUT_OF_RANGE when date_add_interval fails.

    if (date_add_interval(thd, &cur_time, int_type, interval))
    {
      my_error(ER_DATA_OUT_OF_RANGE, MYF(0),
               part_field_array[0]->type_handler()->name().ptr(),
               "INTERVAL");
      return true;
    }