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 corrupted, while the parent ids are still intact. It would require tons of queries to repair it in the application layer. Is there a good way to put the burden on the database and have it recalculate the left/right values using only SQL?

---

Just to clarify, I need to recalculate the numeric lft/rght values of a nested set, not the ids of neighboring records. http://dev.mysql.com/tech-resources/articles/hierarchical-data.html

Answer 1

Using SQL Server, following script seems to be working for me.

Output testscript

category_id name                 parent      lft         rgt         lftcalc     rgtcalc
----------- -------------------- ----------- ----------- ----------- ----------- -----------
1           ELECTRONICS          NULL        1           20          1           20
2           TELEVISIONS          1           2           9           2           9
3           TUBE                 2           3           4           3           4
4           LCD                  2           5           6           5           6
5           PLASMA               2           7           8           7           8
6           PORTABLE ELECTRONICS 1           10          19          10          19
7           MP3 PLAYERS          6           11          14          11          14
8           FLASH                7           12          13          12          13
9           CD PLAYERS           6           15          16          15          16
10          2 WAY RADIOS         6           17          18          17          18

Script

SET NOCOUNT ON
GO

DECLARE @nested_category TABLE (
 category_id INT PRIMARY KEY,
 name VARCHAR(20) NOT NULL,
 parent INT,
 lft INT,
 rgt INT
);

DECLARE @current_Category_ID INTEGER
DECLARE @current_parent INTEGER
DECLARE @SafeGuard INTEGER
DECLARE @myLeft INTEGER
SET @SafeGuard = 100

INSERT INTO @nested_category 
SELECT           1,'ELECTRONICS',NULL,NULL,NULL
UNION ALL SELECT 2,'TELEVISIONS',1,NULL,NULL
UNION ALL SELECT 3,'TUBE',2,NULL,NULL
UNION ALL SELECT 4,'LCD',2,NULL,NULL
UNION ALL SELECT 5,'PLASMA',2,NULL,NULL
UNION ALL SELECT 6,'PORTABLE ELECTRONICS',1,NULL,NULL
UNION ALL SELECT 7,'MP3 PLAYERS',6,NULL,NULL
UNION ALL SELECT 8,'FLASH',7,NULL,NULL
UNION ALL SELECT 9,'CD PLAYERS',6,NULL,NULL
UNION ALL SELECT 10,'2 WAY RADIOS',6,NULL,NULL

/* Initialize */
UPDATE  @nested_category 
SET     lft = 1
        , rgt = 2
WHERE   parent IS NULL

UPDATE  @nested_category 
SET     lft = NULL
        , rgt = NULL
WHERE   parent IS NOT NULL

WHILE EXISTS (SELECT * FROM @nested_category WHERE lft IS NULL) AND @SafeGuard > 0
BEGIN
  SELECT  @current_Category_ID = MAX(nc.category_id)
  FROM    @nested_category nc
          INNER JOIN @nested_category nc2 ON nc2.category_id = nc.parent
  WHERE   nc.lft IS NULL
          AND nc2.lft IS NOT NULL

  SELECT  @current_parent = parent
  FROM    @nested_category
  WHERE   category_id = @current_category_id

  SELECT  @myLeft = lft
  FROM    @nested_category
  WHERE   category_id = @current_parent

  UPDATE @nested_category SET rgt = rgt + 2 WHERE rgt > @myLeft;
  UPDATE @nested_category SET lft = lft + 2 WHERE lft > @myLeft;
  UPDATE @nested_category SET lft = @myLeft + 1, rgt = @myLeft + 2 WHERE category_id = @current_category_id

  SET @SafeGuard = @SafeGuard - 1
END

SELECT * FROM @nested_category ORDER BY category_id

SELECT  COUNT(node.name), node.name, MIN(node.lft)
FROM    @nested_category AS node,
        @nested_category AS parent
WHERE   node.lft BETWEEN parent.lft AND parent.rgt
GROUP BY 
        node.name
ORDER BY
        3, 1

Testscript ##

SET NOCOUNT ON
GO

DECLARE @nested_category TABLE (
 category_id INT PRIMARY KEY,
 name VARCHAR(20) NOT NULL,
 parent INT,
 lft INT,
 rgt INT, 
 lftcalc INT,
 rgtcalc INT
);

INSERT INTO @nested_category 
SELECT           1,'ELECTRONICS',NULL,1,20,NULL,NULL
UNION ALL SELECT 2,'TELEVISIONS',1,2,9,NULL,NULL
UNION ALL SELECT 3,'TUBE',2,3,4,NULL,NULL
UNION ALL SELECT 4,'LCD',2,5,6,NULL,NULL
UNION ALL SELECT 5,'PLASMA',2,7,8,NULL,NULL
UNION ALL SELECT 6,'PORTABLE ELECTRONICS',1,10,19,NULL,NULL
UNION ALL SELECT 7,'MP3 PLAYERS',6,11,14,NULL,NULL
UNION ALL SELECT 8,'FLASH',7,12,13,NULL,NULL
UNION ALL SELECT 9,'CD PLAYERS',6,15,16,NULL,NULL
UNION ALL SELECT 10,'2 WAY RADIOS',6,17,18,NULL,NULL

/* Initialize */
UPDATE  @nested_category 
SET     lftcalc = 1
        , rgtcalc = 2
WHERE   parent IS NULL

DECLARE @current_Category_ID INTEGER
DECLARE @current_parent INTEGER
DECLARE @SafeGuard INTEGER
DECLARE @myRight INTEGER
DECLARE @myLeft INTEGER
SET @SafeGuard = 100
WHILE EXISTS (SELECT * FROM @nested_category WHERE lftcalc IS NULL) AND @SafeGuard > 0
BEGIN
  SELECT  @current_Category_ID = MAX(nc.category_id)
  FROM    @nested_category nc
          INNER JOIN @nested_category nc2 ON nc2.category_id = nc.parent
  WHERE   nc.lftcalc IS NULL
          AND nc2.lftcalc IS NOT NULL

  SELECT  @current_parent = parent
  FROM    @nested_category
  WHERE   category_id = @current_category_id

  SELECT  @myLeft = lftcalc
  FROM    @nested_category
  WHERE   category_id = @current_parent

  UPDATE @nested_category SET rgtcalc = rgtcalc + 2 WHERE rgtcalc > @myLeft;
  UPDATE @nested_category SET lftcalc = lftcalc + 2 WHERE lftcalc > @myLeft;
  UPDATE @nested_category SET lftcalc = @myLeft + 1, rgtcalc = @myLeft + 2 WHERE category_id = @current_category_id

  SELECT * FROM @nested_category WHERE category_id = @current_parent
  SELECT * FROM @nested_category ORDER BY category_id
  SET @SafeGuard = @SafeGuard - 1
END

SELECT * FROM @nested_category ORDER BY category_id

SELECT  COUNT(node.name), node.name, MIN(node.lft)
FROM    @nested_category AS node,
        @nested_category AS parent
WHERE   node.lft BETWEEN parent.lft AND parent.rgt
GROUP BY 
        node.name
ORDER BY
        3, 1

Answer 2

Here's what I have adapted from @Lieven's answer, incorporating feedback from here for better performance:

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.
    CREATE TABLE `tmp_tree` (
        `id`        char(36) NOT NULL DEFAULT '',
        `parent_id` char(36)          DEFAULT NULL,
        `lft`       int(11)  unsigned DEFAULT NULL,
        `rght`      int(11)  unsigned DEFAULT NULL,
        PRIMARY KEY      (`id`),
        INDEX USING HASH (`parent_id`),
        INDEX USING HASH (`lft`),
        INDEX USING HASH (`rght`)
    ) ENGINE = MEMORY
    SELECT `id`,
           `parent_id`,
           `lft`,
           `rght`
    FROM   `tree`;

    # Leveling the playing field.
    UPDATE  `tmp_tree`
    SET     `lft`  = NULL,
            `rght` = NULL;

    # Establishing starting numbers for all root elements.
    WHILE EXISTS (SELECT * FROM `tmp_tree` WHERE `parent_id` IS NULL AND `lft` IS NULL AND `rght` IS NULL LIMIT 1) DO

        UPDATE `tmp_tree`
        SET    `lft`  = startId,
               `rght` = startId + 1
        WHERE  `parent_id` IS NULL
          AND  `lft`       IS NULL
          AND  `rght`      IS NULL
        LIMIT  1;

        SET startId = startId + 2;

    END WHILE;

    # Switching the indexes for the lft/rght columns to B-Trees to speed up the next section, which uses range queries.
    DROP INDEX `lft`  ON `tmp_tree`;
    DROP INDEX `rght` ON `tmp_tree`;
    CREATE INDEX `lft`  USING BTREE ON `tmp_tree` (`lft`);
    CREATE INDEX `rght` USING BTREE ON `tmp_tree` (`rght`);

    # Numbering all child elements
    WHILE EXISTS (SELECT * FROM `tmp_tree` WHERE `lft` IS NULL LIMIT 1) DO

        # Picking an unprocessed element which has a processed parent.
        SELECT     `tmp_tree`.`id`
          INTO     currentId
        FROM       `tmp_tree`
        INNER JOIN `tmp_tree` AS `parents`
                ON `tmp_tree`.`parent_id` = `parents`.`id`
        WHERE      `tmp_tree`.`lft` IS NULL
          AND      `parents`.`lft`  IS NOT NULL
        LIMIT      1;

        # Finding the element's parent.
        SELECT  `parent_id`
          INTO  currentParentId
        FROM    `tmp_tree`
        WHERE   `id` = currentId;

        # Finding the parent's lft value.
        SELECT  `lft`
          INTO  currentLeft
        FROM    `tmp_tree`
        WHERE   `id` = currentParentId;

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

        UPDATE `tmp_tree`
        SET    `lft` = `lft` + 2
        WHERE  `lft` > currentLeft;

        # Setting lft and rght values for current element.
        UPDATE `tmp_tree`
        SET    `lft`  = currentLeft + 1,
               `rght` = currentLeft + 2
        WHERE  `id`   = currentId;

    END WHILE;

    # Writing calculated values back to physical table.
    UPDATE `tree`, `tmp_tree`
    SET    `tree`.`lft`  = `tmp_tree`.`lft`,
           `tree`.`rght` = `tmp_tree`.`rght`
    WHERE  `tree`.`id`   = `tmp_tree`.`id`;

    COMMIT;

    DROP TABLE `tmp_tree`;

END//

DELIMITER ;

Worked well with some test data, but it's still running on my 100,000 records tree, so I can't give any final judgement yet. The naïve script working directly on the physical table has abysmal performance, running for at least hours, more likely days. Switching to a temporary MEMORY table brought this time down to about an hour, choosing the right indexes cut it down to 10 mins.