Finding continuous ranges in a set of numbers

Question

I have a reasonably large set of phone numbers (approximately 2 million) in a database table. These numbers have been inserted in blocks, so there are many continuous ranges of numbers, anything from 10 numbers to 10 thousand in a range. Some of these numbers are in use and are therefore marked as unavailable, the rest are available. Given a particular number I need a way to find continuous ranges of numbers, both above and below that number. The range should continue until it finds an unavailable number, or encounters the boundary of two ranges.

For example given the following set:

1000
1001
1002
1010
1011
1012
1013
1020
1021
1022

Doing a search using 1012 as the parameter should return 1010, 1011, 1012, 1013.

What is a good way of forming a query to find these ranges? We use NHibernate on top of SQL server, a solution using either is fine.

Answer 1

Theoretically the items in a set have no particular value, so I'm assuming you also have some continuous ID column that defines the order of the numbers. Something like this:

ID  Number
1   1000
2   1001
3   1002
4   1010
5   1011
6   1012
7   1013
8   1020
9   1021
10  1022

You could create an extra column that contains the result of Number - ID:

ID  Number  Diff
1   1000    999
2   1001    999
3   1002    999
4   1010    1006
5   1011    1006
6   1012    1006
7   1013    1006
8   1020    1012
9   1021    1012
10  1022    1012

Numbers in the same range will have the same result in the Diff column.

Answer 2

If you use SQL server you should be able to make a recursive query that will join on root.number = leaf.number + 1

If you select the number from the root and from the last recursion, and the level of the recursion you should have a working query.

I would first test performance of that, and then if not satisfactory turn to cursor/row based approach (which in this case would do a job with a single full scan, where recursion can fail by reaching max recursion depth).

Otherwise your options is to store data differently and maintain a list of min, max numbers associated with a table.

This could actually be implemented in triggers with not such a high penalty on single row updates (updates on the single row of the base table would either update, delete or split a row in the min-max table; this can be determined by querying the 'previous' and 'next' row only).

Answer 3

SQL can't really do this in a single query (except there are native SQL enhancements I don't know about), because SQL can't access the row 'before' or 'after'.

You need to go through the sequence in a loop.

You may try NHibernates Enumerable, which doesn't load the entities into memory, but only creates proxies of them. Actually I don't think that it is a good idea, because it will create proxies for the whole 2 million numbers.

Plan B, use paging. Roughly, it looks like this:

List<PhoneNumber> result = new List<PhoneNumber>();

int input = 1012;
int pageSize = 100;
int currentPage = 0;
int expectedNumber = input;

bool carryOn = true;

while(carryOn)
{
  var numbers = session
    .CreateQuery("from PhoneNumber pn where pn.Number > :input")
    .SetInt("input", input)
    .SetFirstResult(currentPage * pageSize)
    .SetMaxResult(pageSize)
    .List<PhoneNumbers>();

  foreach(var number in numbers)
  {
    expectNumber++;
    if (number.Number != expectedNumber) 
    {
      carryOn = false;
      break;
    }
    result.Add(number);
  }

  currentPage++;
}

And the same for the range before in the other direction.

Answer 4

Use an auxiliary table of all possible sequential values or materialize one in a CTE e.g.

WITH
-- materialize a table of sequential integers
l0 AS (SELECT 0 AS c UNION ALL SELECT 0),
l1 AS (SELECT 0 AS c FROM l0 AS a, l0 AS b),
l2 AS (SELECT 0 AS c FROM l1 AS a, l1 AS b),
l3 AS (SELECT 0 AS c FROM l2 AS a, l2 AS b),
l4 AS (SELECT 0 AS c FROM l2 AS a, l3 AS b),
l5 AS (SELECT 0 AS c FROM l2 AS a, l4 AS b),
nums AS (SELECT row_number() OVER(ORDER BY c) AS n FROM l5), 
-- materialize sample table
MyTable (ID) AS 
(
 SELECT 1000
 UNION ALL 
 SELECT 1001
 UNION ALL 
 SELECT 1002
 UNION ALL 
 SELECT 1010
 UNION ALL 
 SELECT 1011
 UNION ALL 
 SELECT 1012
 UNION ALL 
 SELECT 1013
 UNION ALL 
 SELECT 1020
 UNION ALL 
 SELECT 1021
 UNION ALL 
 SELECT 1022
), 
-- materialize parameter table
params (param) AS (SELECT 1012)
SELECT MIN(N1.n) - 1 AS last_in_sequence
  FROM nums AS N1 
       CROSS JOIN params AS P1
 WHERE N1.n > P1.param
       AND NOT EXISTS 
       (
        SELECT * 
          FROM MyTable AS T1
         WHERE N1.n = T1.ID
       );