SQL> SELECT KTUXEUSN, KTUXESLT, KTUXESQN, /* Transaction ID */   
  2  KTUXESTA Status,    KTUXECFL Flags ,KTUXESIZ 
  3  FROM x$ktuxe
  4  WHERE ktuxesta!='INACTIVE';


  KTUXEUSN   KTUXESLT   KTUXESQN STATUS           FLAGS                      KTUXESIZ
---------- ---------- ---------- ---------------- ------------------------ ----------
        13          5      47447 ACTIVE           DEAD                        2819919
        39          0        502 ACTIVE           NONE                              1
        43         45        480 ACTIVE           NONE                              0

SQL> /


  KTUXEUSN   KTUXESLT   KTUXESQN STATUS           FLAGS                      KTUXESIZ
---------- ---------- ---------- ---------------- ------------------------ ----------
        13          5      47447 ACTIVE           DEAD                        2819919<----该值没减小。
        39          0        502 ACTIVE           NONE                              1
        43         45        480 ACTIVE           NONE                              0

查询v$px_session和v$fast_start_servers,显示非常多并行进程在rollback，依据以往的project经验：

于是改为

SQL>alter system set fast_start_parallel_rollback=false scope=both;

之后，再次执行

SQL> SELECT KTUXEUSN, KTUXESLT, KTUXESQN,/* Transaction ID */  
 2  KTUXESTA Status,    KTUXECFL Flags ,KTUXESIZ
 3  FROM x$ktuxe
 4  WHERE ktuxesta!='INACTIVE';
 KTUXEUSN   KTUXESLT   KTUXESQN STATUS           FLAGS                      KTUXESIZ
---------- ---------- -------------------------- ------------------------ ----------
       13          5      47447 ACTIVE           DEAD                        2033516
       35         29        502 ACTIVE           NONE                              1
 
SQL> /
 
 KTUXEUSN   KTUXESLT   KTUXESQN STATUS           FLAGS                      KTUXESIZ
---------- ---------- -------------------------- ------------------------ ----------
       13          5      47447 ACTIVE           DEAD                        2033433<---该值不断变小。
       35         29        502 ACTIVE           NONE                              1

使用例如以下脚本查看回滚完成的估计时间（以天为单位）：

SQL> set serveroutput on
SQL> declare
 2   l_start number;
 3   l_end number;
 4   begin
 5   select ktuxesiz into l_startfrom x$ktuxe where KTUXEUSN=13 and KTUXESLT=5;
 6   dbms_lock.sleep(60);
 7   select ktuxesiz into l_endfrom x$ktuxe where KTUXEUSN=13 and KTUXESLT=5;
 8   dbms_output.put_line('time estDay:'|| round(l_end/(l_start -l_end)/60/24,2));
  9   end;
 10   /
time est Day:.21

 24*0.21=5.04小时。即：估计5.04小时后回滚完成。

另外注意：在其它环境使用时，请注意替换KTUXEUSN=13和KTUXESLT=5