从数据库BLOB中提取得更慢

时间:2016-12-07 03:42:24

标签: oracle blob procedure utl-file

我们一直在运行一个存储过程,以便在过去的5年中每天从数据库中提取BLOBS。我们通常每晚提取大约25个文件,大多数大小约为500KB,少数几乎接近10,000KB。

这个过程从来都不是最快的,但是在我们移动数据中心之后,这个过程可能需要12个小时。当你只提取~55MB左右时,这本身就令人震惊。我们已聘请所有相关团队来了解Oracle的性能,磁盘I / O等,他们声称一切都很完美。

我一直在阅读有关UTL_FILEDBMS_LOB.read的内容,看到人们谈论在每次循环后重置pos等等。说实话我似乎无法弄清楚是什么这意味着,普遍的共识是,有更好的方法可以达到同样的结果。

不幸的是,我们没有自由重构这一点,所以任何人都可以看到我们程序的任何明显错误吗?我只是在努力解决一些我并不完全理解的问题,那些维护我们基础设施的人会把这一切都归咎于这个代码并洗手。

    CREATE OR REPLACE PROCEDURE PKG_EXTRACT (l_brand IN VARCHAR2) AS

   l_file      UTL_FILE.FILE_TYPE;
   l_buffer    RAW(32767);
   l_amount    BINARY_INTEGER := 32767;
   l_pos       INTEGER;
   l_blob      BLOB;
   l_blob_len  INTEGER;
   x  NUMBER;
   l_file_name VARCHAR2(200);
   l_count       INTEGER := 1;
   v_code  NUMBER;
   v_errm  VARCHAR2(64);
 log_file      UTL_FILE.FILE_TYPE;
rec_num number;

BEGIN
DECLARE
   CURSOR extract_cur IS
      SELECT DATA, BIN_NAME
      FROM   STAGING
      WHERE UPPER(EXTRACTED)='N';

BEGIN
log_file := UTL_FILE.fopen('DATA_DOWNLOAD_DIR','pkg_extract.log','a', 32767);

UTL_FILE.put_line(log_file,'Logging is being done in 24 hours format - V1.5 ',TRUE);
 UTL_FILE.put_line(log_file,'Extract procedure started on Date-Time = '|| TO_TIMESTAMP(LOCALTIMESTAMP, 'DD-MON-RR,HH24.MI.SSXFF'),TRUE);


select count(1) into rec_num from staging;

UTL_FILE.put_line(log_file,'Total Number of records found = ' || rec_num , TRUE);
select count(1) into rec_num from staging where UPPER(EXTRACTED)='N';
UTL_FILE.put_line(log_file,'Total Number of records matching criteria = ' || rec_num , TRUE);

   dbms_output.put_line('Loop through records and write them to file');
   FOR extract_rec IN extract_cur
   LOOP

      l_pos := 1;
      l_blob := extract_rec.DATA;
      l_blob_len := DBMS_LOB.getlength(l_blob);

      -- Save blob length.
      x := l_blob_len;

      l_file_name := extract_rec.BIN_NAME ;

      -- Open the destination file.
      dbms_output.put_line('Open the destination file:- ' || l_file_name);
      l_file := UTL_FILE.fopen('DATA_DOWNLOAD_DIR',l_file_name,'wb', 32767);
      dbms_output.put_line('File opened');

      -- Read chunks of the BLOB and write them to the file until complete.
      dbms_output.put_line('l_pos:- ' || l_pos);
      dbms_output.put_line('l_blob_len:- ' || l_blob_len);
      WHILE l_pos <= l_blob_len
      LOOP
         dbms_output.put_line('DBMS_LOB.read from position: ' || l_pos);
         DBMS_LOB.read(l_blob, l_amount, l_pos, l_buffer);
         dbms_output.put_line('UTL_FILE.put_raw');
         UTL_FILE.put_raw(l_file, l_buffer, TRUE);
         dbms_output.put_line('Written ' || l_amount || ' bytes of data starting at position: ' || l_pos);

         -- Set the start position for the next cut.
         l_pos := l_pos + l_amount;

    --updating the extract field

        dbms_output.put_line(extract_rec.BIN_NAME);

      END LOOP;

      l_count := l_count + 1;
      -- Close the file.
      dbms_output.put_line('Close the file:- ' || l_file_name);
      UTL_FILE.fclose(l_file);

      update staging set extracted='Y', extract_timestamp=sysdate where bin_name=extract_rec.BIN_NAME;
      commit;

   END LOOP;

 UTL_FILE.put_line(log_file,'Extract procedure Completed on Date-Time = '|| TO_TIMESTAMP(LOCALTIMESTAMP, 'DD-MON-RR,HH24.MI.SSXFF'),TRUE);

 IF UTL_FILE.is_open(log_file) THEN
      UTL_FILE.fclose(log_file);
end if;
END;

EXCEPTION
   WHEN OTHERS THEN
   v_code := SQLCODE;
   v_errm := SUBSTR(SQLERRM, 1, 64);
   dbms_output.put_line('Error code ' || v_code || ': ' || v_errm);
UTL_FILE.put_line(log_file,'--------------------------------------' ,TRUE);
 UTL_FILE.put_line(log_file,'Error Occurred while executing '||'Error code ' || v_code || ': ' || v_errm ,TRUE);
 UTL_FILE.put_line(log_file,'Extract procedure Completed with errors - '|| TO_TIMESTAMP(LOCALTIMESTAMP, 'DD-MON-RR,HH24.MI.SSXFF'),TRUE);
UTL_FILE.put_line(log_file,'--------------------------------------' ,TRUE);

   -- Close the file if something goes wrong.
   IF UTL_FILE.is_open(l_file) THEN

      UTL_FILE.fclose(l_file);

 IF UTL_FILE.is_open(log_file) THEN
      UTL_FILE.fclose(log_file);
end if;
   END IF;
   RAISE;
END;
/

修改

CURSOR extract_cur的执行计划。

Plan hash value: 3428151562
-----------------------------------------------------------------------------
| Id  | Operation         | Name    | Rows  | Bytes | Cost (%CPU)| Time     |
-----------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |         |     6 |   678 |     9   (0)| 00:00:01 |
|*  1 |  TABLE ACCESS FULL| STAGING |     6 |   678 |     9   (0)| 00:00:01 |
-----------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   1 - filter(UPPER("S"."EXTRACTED")='N')

1 个答案:

答案 0 :(得分:0)

好吧所以我改变了你的代码,试着避免这些事情select count(1) into rec_num from staging,因为oracle在这里做了什么,它将pl / sql转换为sql和sql转换为pl / sql所以更好地使用游标,这里是更改的代码。并尝试添加索引create index extracted_idx on staging(extracted asc);,所以这里是修改后的代码:我不知道它是否编译,因为我无法检查它,但它应该工作。

CREATE OR REPLACE PROCEDURE PKG_EXTRACT(l_brand IN VARCHAR2) AS

  l_file      UTL_FILE.FILE_TYPE;
  l_buffer    RAW(32767);
  l_amount    BINARY_INTEGER := 32767;
  l_pos       INTEGER;
  l_blob      BLOB;
  l_blob_len  INTEGER;
  x           NUMBER;
  l_file_name VARCHAR2(200);
  l_count     INTEGER := 1;
  v_code      NUMBER;
  v_errm      VARCHAR2(64);
  log_file    UTL_FILE.FILE_TYPE;
  rec_num     number;

  CURSOR extract_cur IS
    SELECT DATA, 
               BIN_NAME 
            FROM STAGING 
         WHERE EXTRACTED = 'N';

  cursor c_count is
    select count(1) cnt 
          from staging;

  cursor c_rec_num is
    select count(1)  cnt
         from staging 
        where EXTRACTED = 'N';

BEGIN
  log_file := UTL_FILE.fopen('DATA_DOWNLOAD_DIR',
                             'pkg_extract.log',
                             'a',
                             32767);

  UTL_FILE.put_line(log_file,
                    'Logging is being done in 24 hours format - V1.5 ',
                    TRUE);
  UTL_FILE.put_line(log_file,
                    'Extract procedure started on Date-Time = ' ||
                    TO_TIMESTAMP(LOCALTIMESTAMP, 'DD-MON-RR,HH24.MI.SSXFF'),
                    TRUE);

  open c_cnt;
  fetch c_cnt into rec_num;
  close c_cnt;

  open c_rec_num;
  fetch c_rec_num into rec_num;
  close c_rec_num;

  UTL_FILE.put_line(log_file,
                    'Total Number of records found = ' || rec_num,
                    TRUE);
  --dont know why you doing this again
  open c_rec_num;
  fetch c_rec_num into rec_num;
  close c_rec_num;

  UTL_FILE.put_line(log_file,
                    'Total Number of records matching criteria = ' ||
                    rec_num,
                    TRUE);

  FOR extract_rec IN extract_cur LOOP

    l_pos      := 1;
    l_blob     := extract_rec.DATA;
    l_blob_len := DBMS_LOB.getlength(l_blob);

    -- Save blob length.
    x := l_blob_len;

    l_file_name := extract_rec.BIN_NAME;

    -- Open the destination file.

    l_file := UTL_FILE.fopen('DATA_DOWNLOAD_DIR', l_file_name, 'wb', 32767);

    -- Read chunks of the BLOB and write them to the file until complete.

    WHILE l_pos <= l_blob_len LOOP

      DBMS_LOB.read(l_blob, l_amount, l_pos, l_buffer);

      UTL_FILE.put_raw(l_file, l_buffer, TRUE);

      -- Set the start position for the next cut.
      l_pos := l_pos + l_amount;

    --updating the extract field

    END LOOP;

    l_count := l_count + 1;
    -- Close the file.
    UTL_FILE.fclose(l_file);

    update staging
       set extracted = 'Y', 
                 extract_timestamp = sysdate
     where bin_name = extract_rec.BIN_NAME;
    commit;

  END LOOP;

  UTL_FILE.put_line(log_file,
                    'Extract procedure Completed on Date-Time = ' ||
                    TO_TIMESTAMP(LOCALTIMESTAMP, 'DD-MON-RR,HH24.MI.SSXFF'),
                    TRUE);

  IF UTL_FILE.is_open(log_file) THEN
    UTL_FILE.fclose(log_file);
  end if;
END;

EXCEPTION
WHEN OTHERS THEN 
        v_code := SQLCODE; v_errm := SUBSTR(SQLERRM, 1, 64); 
        UTL_FILE.put_line(log_file, '--------------------------------------', TRUE); 
        UTL_FILE.put_line(log_file, 'Error Occurred while executing ' || 'Error code ' || v_code || ': ' || v_errm, TRUE);
        UTL_FILE.put_line(log_file, 'Extract procedure Completed with errors - ' || TO_TIMESTAMP(LOCALTIMESTAMP, 'DD-MON-RR,HH24.MI.SSXFF'), TRUE); 
        UTL_FILE.put_line(log_file, '--------------------------------------', TRUE);

  -- Close the file if something goes wrong.
    IF UTL_FILE.is_open(l_file) THEN

    UTL_FILE.fclose(l_file);

        IF UTL_FILE.is_open(log_file) THEN 
            UTL_FILE.fclose(log_file);
        end if;
    END IF; 
END;
/