oracle insert、append、parallel、随后查询的redo与磁盘读写

SQL> set autotrace traceonly statistics;
SQL> insert into big_table_dir_test1 select * from big_table_dir_test;

2853792 rows created.

Statistics
----------------------------------------------------------
    148  recursive calls
     358348  db block gets
     111261  consistent gets
      2  physical reads
  333542568  redo size
    832  bytes sent via SQL*Net to client
    817  bytes received via SQL*Net from client
      3  SQL*Net roundtrips to/from client
      2  sorts (memory)
      0  sorts (disk)
    2853792  rows processed

SQL> commit;

Commit complete.

SQL> select f.owner,f.object_name,f.data_object_id,count(1),max(f.data_object_id),min(f.object_id) from big_table_dir_test1 f
group by f.owner,f.object_name,f.data_object_id;  2  

87653 rows selected.

Statistics
----------------------------------------------------------
      7  recursive calls
      1  db block gets
      41034  consistent gets
      0  physical reads   -- 传统路径insert只写buffer cache, redo保证重做
    176  redo size
    4428645  bytes sent via SQL*Net to client
      64793  bytes received via SQL*Net from client
       5845  SQL*Net roundtrips to/from client
      0  sorts (memory)
      0  sorts (disk)
      87653  rows processed

SQL> truncate table big_table_dir_test1;

Table truncated.

SQL> insert /*+ append nologging */ into big_table_dir_test1 select * from big_table_dir_test;

2853792 rows created.

Statistics
----------------------------------------------------------
    228  recursive calls
      44268  db block gets
      42998  consistent gets
      2  physical reads
     376672  redo size
    827  bytes sent via SQL*Net to client
    841  bytes received via SQL*Net from client
      3  SQL*Net roundtrips to/from client
      1  sorts (memory)
      0  sorts (disk)
    2853792  rows processed

SQL> SQL> commit;

Commit complete.

SQL> select f.owner,f.object_name,f.data_object_id,count(1),max(f.data_object_id),min(f.object_id) from big_table_dir_test1 f
group by f.owner,f.object_name,f.data_object_id;
  2  

87653 rows selected.

Statistics
----------------------------------------------------------
      5  recursive calls
      1  db block gets
      40831  consistent gets
      40752  physical reads --直接路径插入后，不经过buffer cache
    168  redo size
    4413020  bytes sent via SQL*Net to client
      64793  bytes received via SQL*Net from client
       5845  SQL*Net roundtrips to/from client
      0  sorts (memory)
      0  sorts (disk)
      87653  rows processed

SQL> SQL> /

87653 rows selected.

Statistics
----------------------------------------------------------
      0  recursive calls
      0  db block gets
      40766  consistent gets
      0  physical reads
      0  redo size
    4310178  bytes sent via SQL*Net to client
      64793  bytes received via SQL*Net from client
       5845  SQL*Net roundtrips to/from client
      0  sorts (memory)
      0  sorts (disk)
      87653  rows processed

SQL> truncate table big_table_dir_test1;

Table truncated.

Elapsed: 00:00:00.62
SQL>
SQL> alter session enable parallel dml;

Session altered.

Elapsed: 00:00:00.00
SQL> SQL> insert /*+ parallel(c,4) */ into big_table_dir_test1 c select * from big_table_dir_test;

2853792 rows created.

Elapsed: 00:00:03.69

Statistics
----------------------------------------------------------
     13  recursive calls
       2574  db block gets
      43108  consistent gets
      0  physical reads
     119108  redo size     -- insert中的parallel导致走了直接路径加载
    830  bytes sent via SQL*Net to client
    840  bytes received via SQL*Net from client
      3  SQL*Net roundtrips to/from client
      1  sorts (memory)
      0  sorts (disk)
    2853792  rows processed

SQL> commit;

Commit complete.
SQL> select f.owner,f.object_name,f.data_object_id,count(1),max(f.data_object_id),min(f.object_id) from big_table_dir_test1 f
group by f.owner,f.object_name,f.data_object_id;  2  

87653 rows selected.

Elapsed: 00:00:03.33

Statistics
----------------------------------------------------------
      5  recursive calls
      1  db block gets
      40896  consistent gets
      40752  physical reads   -- 没有写buffer cache
    168  redo size
    4470876  bytes sent via SQL*Net to client
      64793  bytes received via SQL*Net from client
       5845  SQL*Net roundtrips to/from client
      0  sorts (memory)
      0  sorts (disk)
      87653  rows processed

SQL> truncate table big_table_dir_test1;

insert into big_table_dir_test1 select /*+ parallel(b 4) */ * from big_table_dir_test b;
Table truncated.

Elapsed: 00:00:00.05
SQL> SQL> 

2853792 rows created.

Elapsed: 00:00:04.66

Statistics
----------------------------------------------------------
    139  recursive calls
     358365  db block gets
     110606  consistent gets
      2  physical reads
  333527468  redo size
    846  bytes sent via SQL*Net to client
    840  bytes received via SQL*Net from client
      3  SQL*Net roundtrips to/from client
      1  sorts (memory)
      0  sorts (disk)
    2853792  rows processed

SQL> select /*+ parallel(4) */f.owner,f.object_name,f.data_object_id,count(1),max(f.data_object_id),min(f.object_id) from big_table_dir_test1 f
group by f.owner,f.object_name,f.data_object_id;  2  

87653 rows selected.

Elapsed: 00:00:02.07

Statistics
----------------------------------------------------------
     38  recursive calls
      1  db block gets
      41750  consistent gets
      0  physical reads    -- parallel走了buffer cache
    176  redo size
    4557551  bytes sent via SQL*Net to client
      64793  bytes received via SQL*Net from client
       5845  SQL*Net roundtrips to/from client
      0  sorts (memory)
      0  sorts (disk)
      87653  rows processed

目前暂时无法做到直接路径加载同时满足不生成redo，同时又写一份到buffer cache，这只能依赖于操作系统缓存，但是过多的并发append会导Disk file operations I/O致等待事件。

This event is used to wait for disk file operations (for example, open, close, seek, and resize). It is also used for miscellaneous I/O operations such as block dumps and password file accesses.

Wait Time: The wait time is the actual time it takes to do the I/O

Parameter	Description
FileOperation	Type of file operation
fileno	File identification number
filetype	Type of file (for example, log file, data file, and so on)

我们知道操作系统在操作文件的时候，需要打开文件、关闭文件、定位文件位置等，当这些操作在进行的时候，Oracle就处于等待状态。

操作系统的这些文件操作可以划分如下：

1.file creation
2 file open
3 file resize
4 file deletion
5 file close
6 wait for all aio requests to finish
7 write verification
8 wait for miscellaneous io (ftp, block dump, passwd file)
9 read from snapshot files