Randomizing Strings In A Microservices World

[Peter Pearson]

On Mon, 9 Dec 2019 21:38:43 -0600, Tim Daneliuk <info at tundraware.com> wrote:
> On 12/9/19 8:54 PM, Dennis Lee Bieber wrote:
>> On Mon, 9 Dec 2019 18:52:11 -0600, Tim Daneliuk <info at tundraware.com>
>> declaimed the following:
>> 
>>> - Each of these services needs to produce a string of ten digits
>>> guaranteed to be unique on a per service instance basis AND to not
>>> collide for - oh, let's say - forever :)s
>>>
>>> Can anyone suggest a randomization method that might achieve this
>>> efficiently?
>>>
>>> My first thought was to something like nanonseconds since the epoch
>>> plus something unique about the service instance - like it's IP?
>>> (This is in a K8s cluster) - to see the randomization and
>>> essentially eliminate the string being repeated.
>>>
>>> Ideas welcome ..
>> 
>> 	Well, 10 digits is rather short, but studying
>> https://en.wikipedia.org/wiki/Universally_unique_identifier might provide
>> some ideas.
>
> For a variety of reasons the length of this string cannot exceed 10
> digits.  It is believed that - in this application - the consumption
> of values will be sparse over time.  All I really need is a high
> entropy way to select from among the billion possible values to
> minimize the possibility of collisions (which require retry and time
> we don't want to burn).

Just to be sure: you *are* aware that the "Birthday Paradox" says
that if you pick your 10-digit strings truly randomly, you'll probably
get a collision by the time of your 10**5th string . . . right?

-- 
To email me, substitute nowhere->runbox, invalid->com.