> So the question may actually be: > > How do we improve our locking code, so we don't have to spawn millions > of sequences? > > What is the locking method you are using? >
I am not using locking with the million sequence solution. I do not want something that locks because the problems described below I prefer the solution generates a gap (skip a couple of numbers) and not using locks. > > > The lock part is because we solved a similar problem with a counter by > > row locking the counter and increasing it in another part of the > > database. The result is that all the queries using that table are queued > > by pair (group, element) that is not that bad because we are not > > inserting thousands of rows by second. Still is killing cluster > > performance (but performance is still OK from the business point of > > view). The problem using locks is that they are too sensitive to > > developer errors and bugs. Sometimes connected clients aborts and the > > connection is returned to the pool with the lock active until the > > connection is closed or someone unlocks the row. I would prefer to have > > something more resilient to developers/programming errors, if possible. > > > > Now I read this paragraph, I realize I was not clear enough. I am saying we do not want to use locks because of all the problems described.
