Hi:

> I believe the problem showed up at about 50 users. Rather than waiting
> quietly for the "cumulative" record to be free, they'd get an error and
> retry immediately. Eventually, the volume of retries overwhelmed the
> server, leading to more retries, etc. I believe it's called "live lock".

The way I did it, the retry is in a loop with a random number used as a delay so all
the others are retrying at random intervals and nothing got locked into a repeated
collision.