I'm not sure. When you say the structure is corrupt, is it corrupt right after the deletes, as in, have you reloaded the objects you are examining for correctness? I haven't had any problems. I'm using a smaller number of items in our case, but I don't see how the number would matter at all (the code either works or it doesn't). Are any exceptions being thrown?

Hi chris, yes, i reloaded the objects after the deletes.
For example before the script, i had a structure like this:

root
  a
    a1
      a11
      a12
    a2
    a3
  b
    b1
    b2
    b3
  etc..

Then after the creation/destroy i had something like:

root
  a
    a1
      a11
      a12
    a2
    a3
      b
        b1
        b2
        b3
    etc..

Having another group of categories under another child (luckily only for lft and rgt, parent_id wasn't touched so i was able to just run a renumber_full_tree)

The only problem i think about is that i got all the categories and then checked them, like
Category.all.each do |category|
category.destroy if category.something = '...' and category.all_children_count.zero? end

Without reloading all the categories, but i think that this isn't the problems as in the before_destroy of the plugin there is a:

   self.reload # in case a concurrent move has altered the indexes

which should be the solution for this cases

There weren't any exceptions or other outputs (just that for the destroy it took a while, as there are a lot of records to update, i think)

Usually it works correctly when you create/destroy a single object, but having a lot of them it collapsed, i don't know if maybe because the db (mysql), the plugin or what else :)

Just an OT question, do you think that there would be problems having a lot (1-2 millions) of children under a single root?
I mean, about updates and stuff like that... if you add/destroy a child at the start and it has to update thousands/millions of rows with the new lft/rgt i think it would take quite a lot (and we know that users hate to wait too long :( )