Generating random strings with T-SQL

Question

If you wanted to generate a pseudorandom alphanumeric string using T-SQL, how would you do it? How would you exclude characters like dollar signs, dashes, and slashes from

Answer 1

Here's one I came up with today (because I didn't like any of the existing answers enough).

This one generates a temp table of random strings, is based off of newid(), but also supports a custom character set (so more than just 0-9 & A-F), custom length (up to 255, limit is hard-coded, but can be changed), and a custom number of random records.

Here's the source code (hopefully the comments help):

/**
 * First, we're going to define the random parameters for this
 * snippet. Changing these variables will alter the entire
 * outcome of this script. Try not to break everything.
 *
 * @var {int}       count    The number of random values to generate.
 * @var {int}       length   The length of each random value.
 * @var {char(62)}  charset  The characters that may appear within a random value.
 */

-- Define the parameters
declare @count int = 10
declare @length int = 60
declare @charset char(62) = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789'

/**
 * We're going to define our random table to be twice the maximum
 * length (255 * 2 = 510). It's twice because we will be using
 * the newid() method, which produces hex guids. More later.
 */

-- Create the random table
declare @random table (
    value nvarchar(510)
)

/**
 * We'll use two characters from newid() to make one character in
 * the random value. Each newid() provides us 32 hex characters,
 * so we'll have to make multiple calls depending on length.
 */

-- Determine how many "newid()" calls we'll need per random value
declare @iterations int = ceiling(@length * 2 / 32.0)

/**
 * Before we start making multiple calls to "newid", we need to
 * start with an initial value. Since we know that we need at
 * least one call, we will go ahead and satisfy the count.
 */

-- Iterate up to the count
declare @i int = 0 while @i < @count begin set @i = @i + 1

    -- Insert a new set of 32 hex characters for each record, limiting to @length * 2
    insert into @random
        select substring(replace(newid(), '-', ''), 1, @length * 2)

end

-- Now fill the remaining the remaining length using a series of update clauses
set @i = 0 while @i < @iterations begin set @i = @i + 1

    -- Append to the original value, limit @length * 2
    update @random
        set value = substring(value + replace(newid(), '-', ''), 1, @length * 2)

end

/**
 * Now that we have our base random values, we can convert them
 * into the final random values. We'll do this by taking two
 * hex characters, and mapping then to one charset value.
 */

-- Convert the base random values to charset random values
set @i = 0 while @i < @length begin set @i = @i + 1

    /**
     * Explaining what's actually going on here is a bit complex. I'll
     * do my best to break it down step by step. Hopefully you'll be
     * able to follow along. If not, then wise up and come back.
     */

    -- Perform the update
    update @random
        set value =

            /**
             * Everything we're doing here is in a loop. The @i variable marks
             * what character of the final result we're assigning. We will
             * start off by taking everything we've already done first.
             */

            -- Take the part of the string up to the current index
            substring(value, 1, @i - 1) +

            /**
             * Now we're going to convert the two hex values after the index,
             * and convert them to a single charset value. We can do this
             * with a bit of math and conversions, so function away!
             */

            -- Replace the current two hex values with one charset value
            substring(@charset, convert(int, convert(varbinary(1), substring(value, @i, 2), 2)) * (len(@charset) - 1) / 255 + 1, 1) +
    --  (1) -------------------------------------------------------^^^^^^^^^^^^^^^^^^^^^^^-----------------------------------------
    --  (2) ---------------------------------^^^^^^^^^^^^^^^^^^^^^^11111111111111111111111^^^^-------------------------------------
    --  (3) --------------------^^^^^^^^^^^^^2222222222222222222222222222222222222222222222222^------------------------------------
    --  (4) --------------------333333333333333333333333333333333333333333333333333333333333333---^^^^^^^^^^^^^^^^^^^^^^^^^--------
    --  (5) --------------------333333333333333333333333333333333333333333333333333333333333333^^^4444444444444444444444444--------
    --  (6) --------------------5555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555^^^^----
    --  (7) ^^^^^^^^^^^^^^^^^^^^66666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666^^^^

            /**
             * (1) - Determine the two hex characters that we'll be converting (ex: 0F, AB, 3C, etc.)
             * (2) - Convert those two hex characters to a a proper hexadecimal (ex: 0x0F, 0xAB, 0x3C, etc.)
             * (3) - Convert the hexadecimals to integers (ex: 15, 171, 60)
             * (4) - Determine the conversion ratio between the length of @charset and the range of hexadecimals (255)
             * (5) - Multiply the integer from (3) with the conversion ratio from (4) to get a value between 0 and (len(@charset) - 1)
             * (6) - Add 1 to the offset from (5) to get a value between 1 and len(@charset), since strings start at 1 in SQL
             * (7) - Use the offset from (6) and grab a single character from @subset
             */

            /**
             * All that is left is to add in everything we have left to do.
             * We will eventually process the entire string, but we will
             * take things one step at a time. Round and round we go!
             */

            -- Append everything we have left to do
            substring(value, 2 + @i, len(value))

end

-- Select the results
select value
from @random

It's not a stored procedure, but it wouldn't be that hard to turn it into one. It's also not horrendously slow (it took me ~0.3 seconds to generate 1,000 results of length 60, which is more than I'll ever personally need), which was one of my initial concerns from all of the string mutation I'm doing.

The main takeaway here is that I'm not trying to create my own random number generator, and my character set isn't limited. I'm simply using the random generator that SQL has (I know there's rand(), but that's not great for table results). Hopefully this approach marries the two kinds of answers here, from overly simple (i.e. just newid()) and overly complex (i.e. custom random number algorithm).

It's also short (minus the comments), and easy to understand (at least for me), which is always a plus in my book.

However, this method cannot be seeded, so it's going to be truly random each time, and you won't be able to replicate the same set of data with any means of reliability. The OP didn't list that as a requirement, but I know that some people look for that sort of thing.

I know I'm late to the party here, but hopefully someone will find this useful.