Oracle – Isn’t Global Temporary Table Global?

It depends on why you are creating the temporary tables in MySQL.

Frequently, people that are creating temporary tables in other databases are doing so in order to work around limitations that don't exist in Oracle where readers don't block writers and writers don't block readers. In other databases, you commonly copy data from a permanent table to a temporary table so that your process doesn't block some other process that needs the same data. Since Oracle provides multi-version read consistency, however, this isn't necessary (or beneficial) in Oracle-- your process can process the data sitting in the real tables without worrying that it is going to block someone else. If that's the situation you're in, the proper response is simply to remove the temporary tables and process the data from the permanent tables.

Assuming that you really need a temporary copy of the data, you can create global temporary tables. You would create these tables outside of your code just like a permanent table and use them just like a permanent table inside your code. The global temporary table ensures that each session can only see the data that session has inserted. The only difference is that you're not dropping and recreating the structure of the table inside your procedure.

Another alternative would be to pull the data into a PL/SQL collection that you work with rather than using a temporary table. PL/SQL collections are stored in the server's PGA (one of Oracle's memory structures) so you generally want to limit the size of the collection particularly if there is a lot of data to process or there could be many sessions processing data simultaneously. You can do something like

DECLARE
  TYPE emp_tbl IS TABLE OF emp%rowtype;
  l_emps emp_tbl;

  CURSOR emp_cur 
      IS SELECT *
           FROM emp;
BEGIN
  OPEN emp_cur;
  LOOP
    -- Fetch 10 rows at a time from the cursor into the collection.
    -- You'd realistically want a larger limit, something between 100 and 1000 generally
    FETCH emp_cur 
     BULK COLLECT INTO l_emps
    LIMIT 10;

    EXIT WHEN l_emps.COUNT = 0;

    -- An example of manipulating the collection in memory
    FOR i IN 1 .. l_emps.COUNT
    LOOP
      l_emps(i).sal := l_emps(i).sal * 2;
    END LOOP;

    -- And an example of using the collection to update a table
    FORALL i IN 1 .. l_emps.COUNT
      UPDATE emp
         SET sal = l_emps(i).sal
       WHERE empno = l_emps(i).empno;
  END LOOP;
END;

Of course, in the example above, it would be far simpler (and more efficient) to simply issue an UPDATE statement that would double every employee's salary.

For me the wording "Distributed transactions are not supported for temporary tables" is clear: In a distributed transaction you should not use statements that involve temporary tables. This means you should not even insert into an already created temporary table in an XA transaction. Of course it can happen that it works but nevertheless it is not supported so you should not use it.

I found Note 1051543.1 "Known issues with Global Temporary Tables and Distributed or XA transactions" on Oracle Metalink stating

• the usage of temporary tables is not supported in distributed transactions
• there are a lot of ORA-00600 errors known when using global temporary tables in XA transactions. Even block corruption ca happen
• there are some types of XA transaction where the usage of temporary tables will work. One of such situations is described in the article:

  A global temporary table can be safely used if there is only single branch transaction at the database using it, but if there are loopback database links or XA transactions involving multiple branches, then problems can occur including block corruption (...)

• this restriction on temporary table is documented in the 10g manual . It is not documented in the 11g manual but this is considered as a bug.

This restriction on temporary table is contained in the 12c manual (you supplied the link).