How to repair a corrupted MPTT tree (nested set) in the database using SQL?

Question

I have an MPTT tree of over 100,000 records stored in MySQL using lft, rght and parent_id columns. Now the left/right values became co

Answer 1

You are rescue me!!! I use mixed tree model, so when the day is coming, my tree (30000+) was corrupted. I learn from both of your tech, but is not recovery, just fully rebuild with lost all of sorting and reverse tree... I think, need to keep in mind older cat_left.... Just for possible integrity. So, it maybe looks like...

DROP PROCEDURE IF EXISTS tree_recover;

DELIMITER |

CREATE PROCEDURE tree_recover ()
MODIFIES SQL DATA
BEGIN

    DECLARE currentId, currentParentId  CHAR(36);
    DECLARE currentLeft                 INT;
    DECLARE startId                     INT DEFAULT 1;

    # Determines the max size for MEMORY tables.
    SET max_heap_table_size = 1024 * 1024 * 512;

    START TRANSACTION;

    # Temporary MEMORY table to do all the heavy lifting in,
    # otherwise performance is simply abysmal.
    DROP TABLE IF EXISTS `tmp_cat`;
    CREATE TABLE `tmp_cat` (
        `cat_id`        char(36) NOT NULL DEFAULT '',
        `cat_parent` char(36)          DEFAULT NULL,
        `cat_left`       int(11)  unsigned DEFAULT NULL,
        `cat_right`      int(11)  unsigned DEFAULT NULL,
        `cat_left_old`   int(11)  unsigned DEFAULT NULL,
        PRIMARY KEY      (`cat_id`),
        INDEX USING HASH (`cat_parent`),
        INDEX USING HASH (`cat_left`),
        INDEX USING HASH (`cat_right`),
    INDEX USING HASH (`cat_left_old`)
    ) ENGINE = MEMORY
    SELECT `cat_id`,
           `cat_parent`,
           `cat_left`,
           `cat_right`,
       `cat_left` as cat_left_old
    FROM   `catalog`;

    # Leveling the playing field.
    UPDATE  `tmp_cat`
    SET     `cat_left`  = NULL,
            `cat_right` = NULL;

    # Establishing starting numbers for all root elements.
    WHILE EXISTS (SELECT * FROM `tmp_cat` WHERE `cat_parent` IS NULL AND `cat_left` IS NULL AND `cat_right` IS NULL ORDER BY cat_left_old LIMIT 1) DO

        UPDATE `tmp_cat`
        SET    `cat_left`  = startId,
               `cat_right` = startId + 1
        WHERE  `cat_parent` IS NULL
          AND  `cat_left`       IS NULL
          AND  `cat_right`      IS NULL
        LIMIT  1;

        SET startId = startId + 2;

    END WHILE;

    # Switching the indexes for the cat_left/rght columns to B-Trees to speed up the next section, which uses range queries.
    DROP INDEX `cat_left`  ON `tmp_cat`;
    DROP INDEX `cat_right` ON `tmp_cat`;
    DROP INDEX `cat_left_old` ON `tmp_cat`;

    CREATE INDEX `cat_left`  USING BTREE ON `tmp_cat` (`cat_left`);
    CREATE INDEX `cat_right` USING BTREE ON `tmp_cat` (`cat_right`);
    CREATE INDEX `cat_left_old` USING BTREE ON `tmp_cat` (`cat_left_old`);

    # Numbering all child elements
    WHILE EXISTS (SELECT * FROM `tmp_cat` WHERE `cat_left` IS NULL ORDER BY cat_left_old LIMIT 1) DO

        # Picking an unprocessed element which has a processed parent.
        SELECT     `tmp_cat`.`cat_id`
          INTO     currentId
        FROM       `tmp_cat`
        INNER JOIN `tmp_cat` AS `parents`
                ON `tmp_cat`.`cat_parent` = `parents`.`cat_id`
        WHERE      `tmp_cat`.`cat_left` IS NULL
          AND      `parents`.`cat_left`  IS NOT NULL
    ORDER BY `tmp_cat`.cat_left_old DESC
        LIMIT      1;

        # Finding the element's parent.
        SELECT  `cat_parent`
          INTO  currentParentId
        FROM    `tmp_cat`
        WHERE   `cat_id` = currentId;

        # Finding the parent's cat_left value.
        SELECT  `cat_left`
          INTO  currentLeft
        FROM    `tmp_cat`
        WHERE   `cat_id` = currentParentId;

        # Shifting all elements to the right of the current element 2 to the right.
        UPDATE `tmp_cat`
        SET    `cat_right` = `cat_right` + 2
        WHERE  `cat_right` > currentLeft;

        UPDATE `tmp_cat`
        SET    `cat_left` = `cat_left` + 2
        WHERE  `cat_left` > currentLeft;

        # Setting cat_left and rght values for current element.
        UPDATE `tmp_cat`
        SET    `cat_left`  = currentLeft + 1,
               `cat_right` = currentLeft + 2
        WHERE  `cat_id`   = currentId;

    END WHILE;

    # Writing calculated values back to physical table.
    UPDATE `catalog`, `tmp_cat`
    SET    `catalog`.`cat_left`  = `tmp_cat`.`cat_left`,
           `catalog`.`cat_right` = `tmp_cat`.`cat_right`
    WHERE  `catalog`.`cat_id`   = `tmp_cat`.`cat_id`;

    COMMIT;

    DROP TABLE IF EXISTS `tmp_cat`;

END|