Как ГРУППИРОВАТЬ ПО последовательным данным (в данном случае по дате)

http://sqlfiddle.com/#!2/20108/1

Вот процедура хранения, которая выполняет эту работу

CREATE PROCEDURE myProc()
BEGIN
    -- Drop and create the temp table
    DROP TABLE IF EXISTS reached;
    CREATE TABLE reached (
    sku CHAR(32) PRIMARY KEY,
    record_date date,
    nb int,
    total int)
   ENGINE=HEAP;

-- Initial insert, the starting point is the MAX sales record_date of each product
INSERT INTO reached 
SELECT products.sku, max(sales.record_date), 0, 0
FROM products
join sales on sales.sku = products.sku
group by products.sku;

-- loop until there is no more updated rows
iterloop: LOOP
    -- Update the temptable with the values of the date - 1 row if found
    update reached
    join sales on sales.sku=reached.sku and sales.record_date=reached.record_date
    set reached.record_date = reached.record_date - INTERVAL 1 day, 
        reached.nb=reached.nb+1, 
        reached.total=reached.total + sales.items;

    -- If no more rows are updated it means we hit the most longest days_sold
    IF ROW_COUNT() = 0 THEN
        LEAVE iterloop;
    END IF;
END LOOP iterloop;

-- select the results of the temp table
SELECT products.sku, products.title, products.price, reached.total as sales, reached.nb as days_sold 
from reached
join products on products.sku=reached.sku;

END//

Тогда вам просто нужно сделать

call myProc()

Решение на чистом SQL без процедуры хранения: Fiddle

SELECT sku
     , COUNT(1) AS consecutive_days
     , SUM(items) AS items
FROM
(
  SELECT sku
       , items
       -- generate a new guid for each group of consecutive date
       -- ie : starting with day_before is null
       , @guid := IF(@sku = sku and day_before IS NULL, UUID(), @guid) AS uuid
       , @sku := sku AS dummy_sku
  FROM 
  (
    SELECT currents.sku
         , befores.record_date as day_before
         , currents.items
    FROM sales currents
      LEFT JOIN sales befores 
        ON currents.sku = befores.sku 
        AND currents.record_date = befores.record_date + INTERVAL 1 DAY
    ORDER BY currents.sku, currents.record_date
  )  AS main_join
    CROSS JOIN (SELECT @sku:=0) foo_sku
    CROSS JOIN (SELECT @guid:=UUID()) foo_guid
) AS result_to_group
GROUP BY uuid, sku

Запрос на самом деле не такой сложный. Объявите переменные через cross join (SELECT @type:=0) type. Затем в выборках вы можете устанавливать значения переменных построчно. Это необходимо для имитации функции ранга.