In this particular example, it's totally OK for module2.py to import main.py, there are some gotcha's though.
The most obvious is that main.py is probably being run from the command line, like python main.py, which has the effect of making that file think it's called __main__.  You could import that module in module2, but that's sort of unusual; it's called "main.py" and you want to import main.  Doing so will cause the module to be imported a second time.
For that to be OK, you have to arrange for importing the file to have no side effects unless it's imported as __main__.  A very common idiom in python is to test that condition at the end of a module.  
import module2
global x
hello="Hello"
def main():
    x=module2.message()
    x.say()
if __name__ == '__main__':
    main()
And now it's just fine for module2.py to actually import main.  On the other hand, importing variables from one module into another gets hard to predict when the imports can be recursive, you may not have that variable yet because the module is already trying to import you.  On the other hand, it's always safe to refer to a variable in a module using dotted syntax.  So your module2.py should be:
import main
class message:
    def say():
        print main.hello
which also makes it more obvious just where hello came from.