Sky' Digests tag:www.heysky.net,2008-07-11:/digest//1 2009-08-27T08:42:03Z Movable Type 4.261 oracle 行列转换 tag:www.heysky.net,2009:/digest//1.338 2009-08-27T08:38:35Z 2009-08-27T08:42:03Z From: http://blog.csdn.net/you_tube/archive/2009/04/09/4059251.aspx1.概述最近论坛很多人提的问题都与行列转换有关系，所以我对行列转换的相关知识做了一个总结，希望对大家有所帮助，同时有何错疏，恳请大家指出，我也是在写作过程中学习，算是一起和大家学习吧！行列转换包括以下六种情况：1)列转行2)行转列3)多列转换成字符串4)多行转换成字符串5)字符串转换成多列6)字符串转换成多行下面分别进行举例介绍。首先声明一点，有些例子需要如下10g及以后才有的知识：A.掌握model子句B.正则表达式C.加强的层次查询讨论的适用范围只包括8i,9i,10g及以后版本。2.列转行CREATE TABLE t_col_row(ID INT,c1 VARCHAR2(10),c2 VARCHAR2(10),c3 VARCHAR2(10));INSERT INTO t_col_row VALUES (1, 'v11', 'v21', 'v31');INSERT INTO t_col_row VALUES (2, 'v12', 'v22', NULL);INSERT INTO t_col_row VALUES (3, 'v13', NULL, 'v33');INSERT INTO t_col_row VALUES (4, NULL, 'v24', 'v34');INSERT INTO t_col_row VALUES (5,... Sky http://www.heysky.net From: http://blog.csdn.net/you_tube/archive/2009/04/09/4059251.aspx

1.
概述
最近论坛很多人提的问题都与行列转换有关系，所以我对行列转换的相关知识做了一个总结，希望对大家有所帮助，同时有何错疏，恳请大家指出，我也是在写作过程中学习，算是一起和大家学习吧！
行列转换包括以下六种情况：
1)
列转行
2)
行转列
3)
多列转换成字符串
4)
多行转换成字符串
5)
字符串转换成多列
6)
字符串转换成多行
下面分别进行举例介绍。
首先声明一点，有些例子需要如下10g及以后才有的知识：
A.
掌握model子句
B.
正则表达式
C.
加强的层次查询
讨论的适用范围只包括8i,9i,10g及以后版本。
2.
列转行
CREATE TABLE t_col_row(
ID INT,
c1 VARCHAR2(10),
c2 VARCHAR2(10),
c3 VARCHAR2(10));
INSERT INTO t_col_row VALUES (1, 'v11', 'v21', 'v31');
INSERT INTO t_col_row VALUES (2, 'v12', 'v22', NULL);
INSERT INTO t_col_row VALUES (3, 'v13', NULL, 'v33');
INSERT INTO t_col_row VALUES (4, NULL, 'v24', 'v34');
INSERT INTO t_col_row VALUES (5, 'v15', NULL, NULL);
INSERT INTO t_col_row VALUES (6, NULL, NULL, 'v35');
INSERT INTO t_col_row VALUES (7, NULL, NULL, NULL);
COMMIT;
SELECT * FROM t_col_row;
2.1
UNION ALL
适用范围：8i,9i,10g及以后版本
SELECT id, 'c1' cn, c1 cv
FROM t_col_row
UNION ALL
SELECT id, 'c2' cn, c2 cv
FROM t_col_row
UNION ALL
SELECT id, 'c3' cn, c3 cv FROM t_col_row;
若空行不需要转换，只需加一个where条件，
WHERE COLUMN IS NOT NULL 即可。
2.2
MODEL
适用范围：10g及以后
SELECT id, cn, cv FROM t_col_row
MODEL
RETURN UPDATED ROWS
PARTITION BY (ID)
DIMENSION BY (0 AS n)
MEASURES ('xx' AS cn,'yyy' AS cv,c1,c2,c3)
RULES UPSERT ALL
(
cn[1] = 'c1',
cn[2] = 'c2',
cn[3] = 'c3',
cv[1] = c1[0],
cv[2] = c2[0],
cv[3] = c3[0]
)
ORDER BY ID,cn;
2.3
COLLECTION
适用范围：8i,9i,10g及以后版本
要创建一个对象和一个集合：
CREATE TYPE cv_pair AS OBJECT(cn VARCHAR2(10),cv VARCHAR2(10));
CREATE TYPE cv_varr AS VARRAY(8) OF cv_pair;
SELECT id, t.cn AS cn, t.cv AS cv
FROM t_col_row,
TABLE(cv_varr(cv_pair('c1', t_col_row.c1),
cv_pair('c2', t_col_row.c2),
cv_pair('c3', t_col_row.c3))) t
ORDER BY 1, 2;

]]> 行转列
CREATE TABLE t_row_col AS
SELECT id, 'c1' cn, c1 cv
FROM t_col_row
UNION ALL
SELECT id, 'c2' cn, c2 cv
FROM t_col_row
UNION ALL
SELECT id, 'c3' cn, c3 cv FROM t_col_row;
SELECT * FROM t_row_col ORDER BY 1,2;
3.1
AGGREGATE FUNCTION
适用范围：8i,9i,10g及以后版本
SELECT id,
MAX(decode(cn, 'c1', cv, NULL)) AS c1,
MAX(decode(cn, 'c2', cv, NULL)) AS c2,
MAX(decode(cn, 'c3', cv, NULL)) AS c3
FROM t_row_col
GROUP BY id
ORDER BY 1;
MAX聚集函数也可以用sum、min、avg等其他聚集函数替代。
被指定的转置列只能有一列，但固定的列可以有多列，请看下面的例子：
SELECT mgr, deptno, empno, ename FROM emp ORDER BY 1, 2;
SELECT mgr,
deptno,
MAX(decode(empno, '7788', ename, NULL)) "7788",
MAX(decode(empno, '7902', ename, NULL)) "7902",
MAX(decode(empno, '7844', ename, NULL)) "7844",
MAX(decode(empno, '7521', ename, NULL)) "7521",
MAX(decode(empno, '7900', ename, NULL)) "7900",
MAX(decode(empno, '7499', ename, NULL)) "7499",
MAX(decode(empno, '7654', ename, NULL)) "7654"
FROM emp
WHERE mgr IN (7566, 7698)
AND deptno IN (20, 30)
GROUP BY mgr, deptno
ORDER BY 1, 2;
这里转置列为empno，固定列为mgr，deptno。
还有一种行转列的方式，就是相同组中的行值变为单个列值，但转置的行值不变为列名：
ID CN_1 CV_1 CN_2 CV_2 CN_3 CV_3
1 c1 v11 c2 v21 c3 v31
2 c1 v12 c2 v22 c3
3 c1 v13 c2 c3 v33
4 c1 c2 v24 c3 v34
5 c1 v15 c2 c3
6 c1 c2 c3 v35
7 c1 c2 c3
这种情况可以用分析函数实现：
SELECT id,
MAX(decode(rn, 1, cn, NULL)) cn_1,
MAX(decode(rn, 1, cv, NULL)) cv_1,
MAX(decode(rn, 2, cn, NULL)) cn_2,
MAX(decode(rn, 2, cv, NULL)) cv_2,
MAX(decode(rn, 3, cn, NULL)) cn_3,
MAX(decode(rn, 3, cv, NULL)) cv_3
FROM (SELECT id,
cn,
cv,
row_number() over(PARTITION BY id ORDER BY cn, cv) rn
FROM t_row_col)
GROUP BY ID;
3.2
PL/SQL
适用范围：8i,9i,10g及以后版本
这种对于行值不固定的情况可以使用。
下面是我写的一个包，包中
p_rows_column_real用于前述的第一种不限定列的转换；
p_rows_column用于前述的第二种不限定列的转换。
CREATE OR REPLACE PACKAGE pkg_dynamic_rows_column AS
TYPE refc IS REF CURSOR;
PROCEDURE p_print_sql(p_txt VARCHAR2);
FUNCTION f_split_str(p_str VARCHAR2, p_division VARCHAR2, p_seq INT)
RETURN VARCHAR2;
PROCEDURE p_rows_column(p_table IN VARCHAR2,
p_keep_cols IN VARCHAR2,
p_pivot_cols IN VARCHAR2,
p_where IN VARCHAR2 DEFAULT NULL,
p_refc IN OUT refc);
PROCEDURE p_rows_column_real(p_table IN VARCHAR2,
p_keep_cols IN VARCHAR2,
p_pivot_col IN VARCHAR2,
p_pivot_val IN VARCHAR2,
p_where IN VARCHAR2 DEFAULT NULL,
p_refc IN OUT refc);
END;
/
CREATE OR REPLACE PACKAGE BODY pkg_dynamic_rows_column AS
PROCEDURE p_print_sql(p_txt VARCHAR2) IS
v_len INT;
BEGIN
v_len := length(p_txt);
FOR i IN 1 .. v_len / 250 + 1 LOOP
dbms_output.put_line(substrb(p_txt, (i - 1) * 250 + 1, 250));
END LOOP;
END;
FUNCTION f_split_str(p_str VARCHAR2, p_division VARCHAR2, p_seq INT)
RETURN VARCHAR2 IS
v_first INT;
v_last INT;
BEGIN
IF p_seq < 1 THEN
RETURN NULL;
END IF;
IF p_seq = 1 THEN
IF instr(p_str, p_division, 1, p_seq) = 0 THEN
RETURN p_str;
ELSE
RETURN substr(p_str, 1, instr(p_str, p_division, 1) - 1);
END IF;
ELSE
v_first := instr(p_str, p_division, 1, p_seq - 1);
v_last := instr(p_str, p_division, 1, p_seq);
IF (v_last = 0) THEN
IF (v_first > 0) THEN
RETURN substr(p_str, v_first + 1);
ELSE
RETURN NULL;
END IF;
ELSE
RETURN substr(p_str, v_first + 1, v_last - v_first - 1);
END IF;
END IF;
END f_split_str;
PROCEDURE p_rows_column(p_table IN VARCHAR2,
p_keep_cols IN VARCHAR2,
p_pivot_cols IN VARCHAR2,
p_where IN VARCHAR2 DEFAULT NULL,
p_refc IN OUT refc) IS
v_sql VARCHAR2(4000);
TYPE v_keep_ind_by IS TABLE OF VARCHAR2(4000) INDEX BY BINARY_INTEGER;
v_keep v_keep_ind_by;
TYPE v_pivot_ind_by IS TABLE OF VARCHAR2(4000) INDEX BY BINARY_INTEGER;
v_pivot v_pivot_ind_by;
v_keep_cnt INT;
v_pivot_cnt INT;
v_max_cols INT;
v_partition VARCHAR2(4000);
v_partition1 VARCHAR2(4000);
v_partition2 VARCHAR2(4000);
BEGIN
v_keep_cnt := length(p_keep_cols) - length(REPLACE(p_keep_cols, ',')) + 1;
v_pivot_cnt := length(p_pivot_cols) -
length(REPLACE(p_pivot_cols, ',')) + 1;
FOR i IN 1 .. v_keep_cnt LOOP
v_keep(i) := f_split_str(p_keep_cols, ',', i);
END LOOP;
FOR j IN 1 .. v_pivot_cnt LOOP
v_pivot(j) := f_split_str(p_pivot_cols, ',', j);
END LOOP;
v_sql := 'select max(count(*)) from ' || p_table || ' group by ';
FOR i IN 1 .. v_keep.LAST LOOP
v_sql := v_sql || v_keep(i) || ',';
END LOOP;
v_sql := rtrim(v_sql, ',');
EXECUTE IMMEDIATE v_sql
INTO v_max_cols;
v_partition := 'select ';
FOR x IN 1 .. v_keep.COUNT LOOP
v_partition1 := v_partition1 || v_keep(x) || ',';
END LOOP;
FOR y IN 1 .. v_pivot.COUNT LOOP
v_partition2 := v_partition2 || v_pivot(y) || ',';
END LOOP;
v_partition1 := rtrim(v_partition1, ',');
v_partition2 := rtrim(v_partition2, ',');
v_partition := v_partition || v_partition1 || ',' || v_partition2 ||
', row_number() over (partition by ' || v_partition1 ||
' order by ' || v_partition2 || ') rn from ' || p_table;
v_partition := rtrim(v_partition, ',');
v_sql := 'select ';
FOR i IN 1 .. v_keep.COUNT LOOP
v_sql := v_sql || v_keep(i) || ',';
END LOOP;
FOR i IN 1 .. v_max_cols LOOP
FOR j IN 1 .. v_pivot.COUNT LOOP
v_sql := v_sql || ' max(decode(rn,' || i || ',' || v_pivot(j) ||
',null))' || v_pivot(j) || '_' || i || ',';
END LOOP;
END LOOP;
IF p_where IS NOT NULL THEN
v_sql := rtrim(v_sql, ',') || ' from (' || v_partition || ' ' ||
p_where || ') group by ';
ELSE
v_sql := rtrim(v_sql, ',') || ' from (' || v_partition ||
') group by ';
END IF;
FOR i IN 1 .. v_keep.COUNT LOOP
v_sql := v_sql || v_keep(i) || ',';
END LOOP;
v_sql := rtrim(v_sql, ',');
p_print_sql(v_sql);
OPEN p_refc FOR v_sql;
EXCEPTION
WHEN OTHERS THEN
OPEN p_refc FOR
SELECT 'x' FROM dual WHERE 0 = 1;
END;
PROCEDURE p_rows_column_real(p_table IN VARCHAR2,
p_keep_cols IN VARCHAR2,
p_pivot_col IN VARCHAR2,
p_pivot_val IN VARCHAR2,
p_where IN VARCHAR2 DEFAULT NULL,
p_refc IN OUT refc) IS
v_sql VARCHAR2(4000);
TYPE v_keep_ind_by IS TABLE OF VARCHAR2(4000) INDEX BY BINARY_INTEGER;
v_keep v_keep_ind_by;
TYPE v_pivot_ind_by IS TABLE OF VARCHAR2(4000) INDEX BY BINARY_INTEGER;
v_pivot v_pivot_ind_by;
v_keep_cnt INT;
v_group_by VARCHAR2(2000);
BEGIN
v_keep_cnt := length(p_keep_cols) - length(REPLACE(p_keep_cols, ',')) + 1;
FOR i IN 1 .. v_keep_cnt LOOP
v_keep(i) := f_split_str(p_keep_cols, ',', i);
END LOOP;
v_sql := 'select ' || 'cast(' || p_pivot_col ||
' as varchar2(200)) as ' || p_pivot_col || ' from ' || p_table ||
' group by ' || p_pivot_col;
EXECUTE IMMEDIATE v_sql BULK COLLECT
INTO v_pivot;
FOR i IN 1 .. v_keep.COUNT LOOP
v_group_by := v_group_by || v_keep(i) || ',';
END LOOP;
v_group_by := rtrim(v_group_by, ',');
v_sql := 'select ' || v_group_by || ',';
FOR x IN 1 .. v_pivot.COUNT LOOP
v_sql := v_sql || ' max(decode(' || p_pivot_col || ',' || chr(39) ||
v_pivot(x) || chr(39) || ',' || p_pivot_val ||
',null)) as "' || v_pivot(x) || '",';
END LOOP;
v_sql := rtrim(v_sql, ',');
IF p_where IS NOT NULL THEN
v_sql := v_sql || ' from ' || p_table || p_where || ' group by ' ||
v_group_by;
ELSE
v_sql := v_sql || ' from ' || p_table || ' group by ' || v_group_by;
END IF;
p_print_sql(v_sql);
OPEN p_refc FOR v_sql;
EXCEPTION
WHEN OTHERS THEN
OPEN p_refc FOR
SELECT 'x' FROM dual WHERE 0 = 1;
END;
END;
/
4.
多列转换成字符串
CREATE TABLE t_col_str AS
SELECT * FROM t_col_row;
这个比较简单，用||或concat函数可以实现：
SELECT concat('a','b') FROM dual;
4.1
|| OR CONCAT
适用范围：8i,9i,10g及以后版本
SELECT * FROM t_col_str;
SELECT ID,c1||','||c2||','||c3 AS c123
FROM t_col_str;
5.
多行转换成字符串
CREATE TABLE t_row_str(
ID INT,
col VARCHAR2(10));
INSERT INTO t_row_str VALUES(1,'a');
INSERT INTO t_row_str VALUES(1,'b');
INSERT INTO t_row_str VALUES(1,'c');
INSERT INTO t_row_str VALUES(2,'a');
INSERT INTO t_row_str VALUES(2,'d');
INSERT INTO t_row_str VALUES(2,'e');
INSERT INTO t_row_str VALUES(3,'c');
COMMIT;
SELECT * FROM t_row_str;
5.1
MAX + DECODE
适用范围：8i,9i,10g及以后版本
SELECT id,
MAX(decode(rn, 1, col, NULL)) ||
MAX(decode(rn, 2, ',' || col, NULL)) ||
MAX(decode(rn, 3, ',' || col, NULL)) str
FROM (SELECT id,
col,
row_number() over(PARTITION BY id ORDER BY col) AS rn
FROM t_row_str) t
GROUP BY id
ORDER BY 1;
5.2
ROW_NUMBER + LEAD
适用范围：8i,9i,10g及以后版本
SELECT id, str
FROM (SELECT id,
row_number() over(PARTITION BY id ORDER BY col) AS rn,
col || lead(',' || col, 1) over(PARTITION BY id ORDER BY col) ||
lead(',' || col, 2) over(PARTITION BY id ORDER BY col) ||
lead(',' || col, 3) over(PARTITION BY id ORDER BY col) AS str
FROM t_row_str)
WHERE rn = 1
ORDER BY 1;
5.3
MODEL
适用范围：10g及以后版本
SELECT id, substr(str, 2) str FROM t_row_str
MODEL
RETURN UPDATED ROWS
PARTITION BY(ID)
DIMENSION BY(row_number() over(PARTITION BY ID ORDER BY col) AS rn)
MEASURES (CAST(col AS VARCHAR2(20)) AS str)
RULES UPSERT
ITERATE(3) UNTIL( presentv(str[iteration_number+2],1,0)=0)
(str[0] = str[0] || ',' || str[iteration_number+1])
ORDER BY 1;
5.4
SYS_CONNECT_BY_PATH
适用范围：8i,9i,10g及以后版本
SELECT t.id id, MAX(substr(sys_connect_by_path(t.col, ','), 2)) str
FROM (SELECT id, col, row_number() over(PARTITION BY id ORDER BY col) rn
FROM t_row_str) t
START WITH rn = 1
CONNECT BY rn = PRIOR rn + 1
AND id = PRIOR id
GROUP BY t.id;
适用范围：10g及以后版本
SELECT t.id id, substr(sys_connect_by_path(t.col, ','), 2) str
FROM (SELECT id, col, row_number() over(PARTITION BY id ORDER BY col) rn
FROM t_row_str) t
WHERE connect_by_isleaf = 1
START WITH rn = 1
CONNECT BY rn = PRIOR rn + 1
AND id = PRIOR id;
5.5
WMSYS.WM_CONCAT
适用范围：10g及以后版本
这个函数预定义按','分隔字符串，若要用其他符号分隔可以用，replace将','替换。
SELECT id, REPLACE(wmsys.wm_concat(col), ',', '/') str
FROM t_row_str
GROUP BY id;
6.
字符串转换成多列
其实际上就是一个字符串拆分的问题。
CREATE TABLE t_str_col AS
SELECT ID,c1||','||c2||','||c3 AS c123
FROM t_col_str;
SELECT * FROM t_str_col;
6.1
SUBSTR + INSTR
适用范围：8i,9i,10g及以后版本
SELECT id,
c123,
substr(c123, 1, instr(c123 || ',', ',', 1, 1) - 1) c1,
substr(c123,
instr(c123 || ',', ',', 1, 1) + 1,
instr(c123 || ',', ',', 1, 2) - instr(c123 || ',', ',', 1, 1) - 1) c2,
substr(c123,
instr(c123 || ',', ',', 1, 2) + 1,
instr(c123 || ',', ',', 1, 3) - instr(c123 || ',', ',', 1, 2) - 1) c3
FROM t_str_col
ORDER BY 1;
6.2
REGEXP_SUBSTR
适用范围：10g及以后版本
SELECT id,
c123,
rtrim(regexp_substr(c123 || ',', '.*?' || ',', 1, 1), ',') AS c1,
rtrim(regexp_substr(c123 || ',', '.*?' || ',', 1, 2), ',') AS c2,
rtrim(regexp_substr(c123 || ',', '.*?' || ',', 1, 3), ',') AS c3
FROM t_str_col
ORDER BY 1;
7.
字符串转换成多行
CREATE TABLE t_str_row AS
SELECT id,
MAX(decode(rn, 1, col, NULL)) ||
MAX(decode(rn, 2, ',' || col, NULL)) ||
MAX(decode(rn, 3, ',' || col, NULL)) str
FROM (SELECT id,
col,
row_number() over(PARTITION BY id ORDER BY col) AS rn
FROM t_row_str) t
GROUP BY id
ORDER BY 1;
SELECT * FROM t_str_row;
7.1
UNION ALL
适用范围：8i,9i,10g及以后版本
SELECT id, 1 AS p, substr(str, 1, instr(str || ',', ',', 1, 1) - 1) AS cv
FROM t_str_row
UNION ALL
SELECT id,
2 AS p,
substr(str,
instr(str || ',', ',', 1, 1) + 1,
instr(str || ',', ',', 1, 2) - instr(str || ',', ',', 1, 1) - 1) AS cv
FROM t_str_row
UNION ALL
SELECT id,
3 AS p,
substr(str,
instr(str || ',', ',', 1, 1) + 1,
instr(str || ',', ',', 1, 2) - instr(str || ',', ',', 1, 1) - 1) AS cv
FROM t_str_row
ORDER BY 1, 2;
适用范围：10g及以后版本
SELECT id, 1 AS p, rtrim(regexp_substr(str||',', '.*?' || ',', 1, 1), ',') AS cv
FROM t_str_row
UNION ALL
SELECT id, 2 AS p, rtrim(regexp_substr(str||',', '.*?' || ',', 1, 2), ',') AS cv
FROM t_str_row
UNION ALL
SELECT id, 3 AS p, rtrim(regexp_substr(str||',', '.*?' || ',',1,3), ',') AS cv
FROM t_str_row
ORDER BY 1, 2;
7.2
VARRAY
适用范围：8i,9i,10g及以后版本
要创建一个可变数组：
CREATE OR REPLACE TYPE ins_seq_type IS VARRAY(8) OF NUMBER;
SELECT * FROM TABLE(ins_seq_type(1, 2, 3, 4, 5));
SELECT t.id,
c.column_value AS p,
substr(t.ca,
instr(t.ca, ',', 1, c.column_value) + 1,
instr(t.ca, ',', 1, c.column_value + 1) -
(instr(t.ca, ',', 1, c.column_value) + 1)) AS cv
FROM (SELECT id,
',' || str || ',' AS ca,
length(str || ',') - nvl(length(REPLACE(str, ',')), 0) AS cnt
FROM t_str_row) t
INNER JOIN TABLE(ins_seq_type(1, 2, 3)) c ON c.column_value <=
t.cnt
ORDER BY 1, 2;
7.3
SEQUENCE SERIES
这类方法主要是要产生一个连续的整数列，产生连续整数列的方法有很多，主要有：
CONNECT BY,ROWNUM+all_objects,CUBE等。
适用范围：8i,9i,10g及以后版本
SELECT t.id,
c.lv AS p,
substr(t.ca,
instr(t.ca, ',', 1, c.lv) + 1,
instr(t.ca, ',', 1, c.lv + 1) -
(instr(t.ca, ',', 1, c.lv) + 1)) AS cv
FROM (SELECT id,
',' || str || ',' AS ca,
length(str || ',') - nvl(length(REPLACE(str, ',')), 0) AS cnt
FROM t_str_row) t,
(SELECT LEVEL lv FROM dual CONNECT BY LEVEL <= 5) c
WHERE c.lv <= t.cnt
ORDER BY 1, 2;
SELECT t.id,
c.rn AS p,
substr(t.ca,
instr(t.ca, ',', 1, c.rn) + 1,
instr(t.ca, ',', 1, c.rn + 1) -
(instr(t.ca, ',', 1, c.rn) + 1)) AS cv
FROM (SELECT id,
',' || str || ',' AS ca,
length(str || ',') - nvl(length(REPLACE(str, ',')), 0) AS cnt
FROM t_str_row) t,
(SELECT rownum rn FROM all_objects WHERE rownum <= 5) c
WHERE c.rn <= t.cnt
ORDER BY 1, 2;
SELECT t.id,
c.cb AS p,
substr(t.ca,
instr(t.ca, ',', 1, c.cb) + 1,
instr(t.ca, ',', 1, c.cb + 1) -
(instr(t.ca, ',', 1, c.cb) + 1)) AS cv
FROM (SELECT id,
',' || str || ',' AS ca,
length(str || ',') - nvl(length(REPLACE(str, ',')), 0) AS cnt
FROM t_str_row) t,
(SELECT rownum cb FROM (SELECT 1 FROM dual GROUP BY CUBE(1, 2))) c
WHERE c.cb <= t.cnt
ORDER BY 1, 2;
适用范围：10g及以后版本
SELECT t.id,
c.lv AS p,
rtrim(regexp_substr(t.str || ',', '.*?' || ',', 1, c.lv), ',') AS cv
FROM (SELECT id,
str,
length(regexp_replace(str || ',', '[^' || ',' || ']', NULL)) AS cnt
FROM t_str_row) t
INNER JOIN (SELECT LEVEL lv FROM dual CONNECT BY LEVEL <= 5) c ON c.lv <= t.cnt
ORDER BY 1, 2;
7.4
HIERARCHICAL + DBMS_RANDOM
适用范围：10g及以后版本
SELECT id,
LEVEL AS p,
rtrim(regexp_substr(str || ',', '.*?' || ',', 1, LEVEL), ',') AS cv
FROM t_str_row
CONNECT BY id = PRIOR id
AND PRIOR dbms_random.VALUE IS NOT NULL
AND LEVEL <=
length(regexp_replace(str || ',', '[^' || ',' || ']', NULL))
ORDER BY 1, 2;
7.5
HIERARCHICAL + CONNECT_BY_ROOT
适用范围：10g及以后版本
SELECT id,
LEVEL AS p,
rtrim(regexp_substr(str || ',', '.*?' || ',', 1, LEVEL), ',') AS cv
FROM t_str_row
CONNECT BY id = connect_by_root id
AND LEVEL <=
length(regexp_replace(str || ',', '[^' || ',' || ']', NULL))
ORDER BY 1, 2;
7.6
MODEL
适用范围：10g及以后版本
SELECT id, p, cv FROM t_str_row
MODEL
RETURN UPDATED ROWS
PARTITION BY(ID)
DIMENSION BY( 0 AS p)
MEASURES( str||',' AS cv)
RULES UPSERT
(cv
[ FOR p
FROM 1 TO length(regexp_replace(cv[0],'[^'||','||']',null))]]> On Tuning by Tracing tag:www.heysky.net,2009:/digest//1.337 2009-07-31T08:21:41Z 2009-07-31T08:29:52Z From: http://www.oracle.com/technology/oramag/oracle/08-jan/o18asktom.htmlOn Tuning by Tracing By Tom Kyte Our technologist does a tune-up with SQL_TRACE. My all-time favorite presentation is called "All About Binds." It takes me about three hours to do the entire talk--which covers performance, memory utilization, scalability,... Sky http://www.heysky.net From: http://www.oracle.com/technology/oramag/oracle/08-jan/o18asktom.html

On Tuning by Tracing
By Tom Kyte Oracle ACE

Our technologist does a tune-up with SQL_TRACE.

My all-time favorite presentation is called "All About Binds." It takes me about three hours to do the entire talk--which covers performance, memory utilization, scalability, security, bind mismatches, bind variable peeking, and cursor sharing--from start to finish.

Today, I got this message on Ask Tom:

I have a query that runs slowly (it takes about two minutes). So what do I do? I turn on tracing (SQL_TRACE=TRUE) before running the query, and consistently, 100 percent of the time, without my changing anything else, when I enable tracing the query comes back instantly!

This message on Ask Tom shows exactly what I demonstrate in the "All About Binds" presentation, during which I say, "I'm going to show you how to tune by setting SQL_TRACE=TRUE. You are probably all thinking that I'm going to run a query, observe it running slowly, trace it, and then tune it. Well, I'm not going to do that--all I'm going to do is set SQL_TRACE=TRUE, and you'll observe that the query performance and resource utilization are dramatically affected--all for the better!"

I know exactly what is happening in the Ask Tom scenario--it is a combination of two things:

  • When you set SQL_TRACE=TRUE, you set up a new "parse environment" (a made-up term). You have changed your session in such a way that it will not share any existing SQL that was not parsed with SQL_TRACE enabled. So it is highly likely that you will either hard-parse a new version of the query or use some existing child cursor that is different from the one you would use with SQL_TRACE disabled.

  • Bind variable peeking happens at hard-parse time and may affect the plan chosen by the optimizer.

]]> Let's start with the setup script to demonstrate this phenomenon. I'll set up a very skewed set of data for emphasis: 

SQL> create table t
  2    as
  3    select case when rownum = 1 
  4    then 1 else 99 end id, a.* 
  5      from all_objects a
  6    /
Table created.

SQL> create index t_idx on t(id);
Index created.

SQL> begin
  2    dbms_stats.gather_table_stats
  3    ( user, 'T',
  4     method_opt=> 
  5    'for all'|| 
  6    'indexed columns'|| 
  7    'size 254' );
  8    end; 
  9    /
PL/SQL procedure successfully completed.

It is quite clear that WHERE ID=1 will return one record and WHERE ID=99 will return all of the rest (about 50,000 records). Also, the optimizer is very aware of this fact, because of the histograms in place, and if we parse by using literals, we can definitely see different plans for different inputs, as shown in Listing 1.

Code Listing 1: Different inputs, different plans 

SQL> set autotrace traceonly explain
SQL> select * from t where id = 1;

Execution Plan
-------------------------------------------------------------------------------------------------------
|  Id  |  Operation                          |  Name  |  Rows  |  Bytes  |  Cost (%CPU)  |  Time      |
-------------------------------------------------------------------------------------------------------
|   0  |  SELECT STATEMENT                   |        |      1 |      96 |      2   (0)  |  00:00:01  |
|   1  |    TABLE ACCESS BY INDEX ROWID      | T      |      1 |      96 |      2   (0)  |  00:00:01  |
|*  2  |      INDEX RANGE SCAN               | T_IDX  |      1 |         |      1   (0)  |  00:00:01  |
-------------------------------------------------------------------------------------------------------
SQL> select * from t where id = 99;

Execution Plan
-------------------------------------------------------------------------------------------------------
|  Id  |  Operation                          |  Name  |  Rows  |  Bytes |  Cost (%CPU)  |  Time       |
-------------------------------------------------------------------------------------------------------
|   0  |  SELECT STATEMENT                   |        |  50244 |  4710K |   235    (4)  |  00:00:02   |
|*  1  |   TABLE ACCESS FULL                 | T      |  50244 |  4710K |   235    (4)  |  00:00:02   |
-------------------------------------------------------------------------------------------------------

Oracle8i Database Release 3 and Earlier

But what happens when we bind the ID input--when we issue SELECT * FROM t WHERE ID = :ID? Well, the answer depends on which release of Oracle Database is being used. With Oracle8i Database Release 3 and earlier, the optimizer will more or less "guess." It will not have any information about the WHERE ID = :ID clause. The information it has is

  • ID has two values in the table: 1 and 99.
  • There are 50,000 records in the table.

And because ID has only two values, this old optimizer will guess that for any query on ID, about half of the table on average will be selected. And the database would come up with one plan, regardless of the inputs. In this case, it would likely do a full scan of the table.

Oracle9i Database and Oracle Database 10g

From Oracle9i Database Release 1 through Oracle Database 10g Release 2, Oracle Database will wait until the cursor is opened to do the actual optimization of the query--it will wait for the bind variable value to be supplied by the application before figuring out the right way to optimize the query. This is called bind variable peeking, when the optimizer first looks at the bind values and then optimizes the query. In this case, however, depending on which inputs are used to first run the query, the database will either choose a full scan or an index range scan plus table access by index rowid. And in Oracle9i Database Release 1 through Oracle Database 10g Release 2, that is the plan that will be used to execute the SELECT * FROM t WHERE ID = :ID query, regardless of the subsequent bind values, until the query is hard-parsed and optimized again.

I'll address what happens in Oracle Database 11g Release 1 in a minute, but first let's take a look at what is happening to the person whose scenario on Ask Tom inspired this discussion:

1. Someone hard-parsed the query in question, and the inputs used resulted in "Plan A."

2. Plan A was the best plan for that person, given the inputs--it was the most efficient plan.

3. Later, someone else executed the same query. Using shared SQL, this person reused the plan generated by the other person (Plan A in No. 1), but the new bind variable inputs were different and the plan generated for No. 1 was not the best plan for these inputs. In fact, the plan was miserable for these new inputs.

4. The person in No. 3 turned on SQL_TRACE and executed the query with inputs from No. 3 again, but because SQL_TRACE was on, the query did not share the SQL. The query was hard-parsed--with the new inputs--and this resulted in a very different plan, "Plan B," and given these inputs, Plan B was much better than Plan A.

Using Oracle Database 10g Release 2, we can observe this easily. Consider the following: 

SQL> variable id number
SQL> set autotrace traceonly statistics
SQL> exec :id := 99
PL/SQL procedure successfully completed.

SQL> select * from t where id = :id;
50254 rows selected.

Statistics
----------------------------------
          
       4031  consistent gets
          
      50254  rows processed

So we started off with ID=99 as the bind, and the optimizer chose a full scan (you can prove that via TKPROF by looking at the row source operation if you wish; I did. . . .) Therefore, regardless of the bind value, the database will execute a full scan from now on. For example, 

SQL> exec :id : = 1
PL/SQL procedure successfully completed.

SQL> select * from t where id = :id;

Statistics
----------------------------------
         
        720  consistent gets
         
          1  rows processed

This result demonstrates that it was unlikely that an index range scan/table access by index rowid was executed. In such a case, we'd expect many fewer logical I/Os (consistent gets)--three or four against an index and one against the table. This result represents our "poorly performing query." Now we turn on SQL_TRACE to find the performance characteristics, and we observe 

SQL> alter session set sql_trace = true;
Session altered.

SQL> select * from t where id = :id;

Statistics
-----------------------------------
          
          4  consistent gets
          
          1  rows processed

Apparently just setting SQL_TRACE=TRUE is our tuning mechanism! It isn't really--bind peeking and hard parsing are what's causing this result, but unless you know that SQL_TRACE sets up a child cursor and that bind peeking happens, this looks very mysterious, especially if you turn SQL_TRACE off: 

SQL> alter session set sql_trace=false;
Session altered.

SQL> select * from t where id = :id;

Statistics
------------------------------------
          
        720  consistent gets
          
          1  rows processed

See it run slowly (720 consistent gets). And now that we've seen slow, turn SQL_TRACE on again: 

SQL> alter session set sql_trace=true;
Session altered.

SQL> select * from t where id = :id;
Statistics
--------------------------------------
          
          4  consistent gets
          
          1  rows processed

And see it go fast. It is not magic; it is bind peeking. To see the real plans, you can query V$SQL_PLAN / V$SQL_PLAN_STATISTICS or use DBMS_XPLAN to dump the plans for each child cursor, as shown in Listing 2.

Code Listing 2: Using DBMS_XPLAN to show the plans for each child cursor 

SQL> select sql_id from v$sql where sql_text = 'select * from t where id = :id';

SQL_ID
---------------------------------------
8s40hfjcbmxzk
8s40hfjcbmxzk

SQL> select * from table( dbms_xplan.display_cursor( '8s40hfjcbmxzk', 0 ) );

PLAN_TABLE_OUTPUT
---------------------------------------
SQL_ID  8s40hfjcbmxzk, child number 0
---------------------------------------
select * from t where id = :id

Plan hash value: 1601196873

--------------------------------------------------------------------------------------
|  Id  |  Operation            |  Name  |  Rows |  Bytes  |  Cost (%CPU)  |  Time    |
--------------------------------------------------------------------------------------
|   0  |  SELECT STATEMENT     |        |       |         |    235 (100)  |          |
|*  1  |    TABLE ACCESS FULL  | T      | 50250 |   4710K |    235   (4)  |  00:00:02|
--------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
--------------------------------------------------------------------------------------
1 - filter("ID"=:ID)

18 rows selected.

SQL> select * from table( dbms_xplan.display_cursor( '8s40hfjcbmxzk', 1 ) );

PLAN_TABLE_OUTPUT
------------------------------------------
SQL_ID  8s40hfjcbmxzk, child number 1
------------------------------------------
select * from t where id = :id

Plan hash value: 470836197

-------------------------------------------------------------------------------------------------
|  Id  |  Operation                     |  Name  |  Rows  |  Bytes  |  Cost (%CPU)  |  Time     |
-------------------------------------------------------------------------------------------------
|   0  |  SELECT STATEMENT              |        |        |         |   2   (100)   |           |
|   1  |   TABLE ACCESS BY INDEX ROWID  |  T     |      1 |     96  |   2      (0)  | 00:00:01  |
|*  2  |    INDEX RANGE SCAN            |  T_IDX |      1 |         |   1      (0)  | 00:00:01  |
-------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
----------------------------------------------------
2 - access("ID"=:ID)

19 rows selected.

You can use v$sql_shared_cursor and the SQL_ID from Listing 2 to see why you have multiple child cursors as well. For example, 

SQL> select child_number, 
  2         stats_row_mismatch
  3      from v$sql_shared_cursor 
  4      where sql_id = '8s40hfjcbmxzk'
  5    /

CHILD_NUMBER   STATS_ROW_MISMATCH 
-------------  ------------------                                                 
                   0 N
                   1 Y

That is what you would look for to verify that you were seeing an "it must be SQL_TRACE creating a new child cursor and a hard parse with different binds" result. (The queries you run to verify this in Oracle9i Database are slightly different from the Oracle Database 10g Release 2 queries above. If you are interested in the Oracle9i Database queries, you can review tkyte.blogspot.com/2007/09/sqltracetrue-part-two.html, where I worked the example through in that release.)

Oracle Database 11g Release 1

The current release of Oracle Database--Oracle Database 11g--changes the above behavior for the better. A new feature--intelligent cursor sharing--enables the optimizer to recognize that performance can be affected when a fixed, one-size-fits-all plan is used and come up with different plans for certain bind ranges. In the following example, I simply used the schema from above--the copy of ALL_OBJECTS, with one row having the value ID=1, the index in place, and statistics gathered. Then I ran a PL/SQL block that would open a cursor by using one of two inputs--either 1 or 99. This time, instead of just one plan being generated, we'll see two. The optimizer recognized that one plan was not sufficient for all inputs and allowed for multiple plans to be generated, as shown in Listing 3.

Code Listing 3: Multiple plans with intelligent cursor sharing (Oracle Database 11g) 

SQL> select sql_id
  2      from v$sql
  3    where sql_text = 'select * from T where id = :x';
no rows selected

SQL> declare
  2    l_cursor sys_refcursor;
  3    type array is table of number;
  4    l_data  array := array(99,1);
  5    type data_array is table of T%rowtype;
  6    l_recs  data_array;
  7    begin
  8    for i in 1 .. l_data.count
  9    loop
 10       open l_cursor for 'select * from T where id = :x' using l_data( i );
 11       loop
 12          fetch l_cursor bulk collect into l_recs limit 500;
 13          exit when l_cursor%notfound;
 14       end loop;
 15       close l_cursor;
 16    end loop;
 17    end;
 18    /
PL/SQL procedure successfully completed.

SQL> select sql_id
  2    from v$sql
  3   where sql_text = 'select * from T where id = :x';
SQL_ID
-----------------------------------------------------------------------
gw1fks2wb4j2g
gw1fks2wb4j2g

SQL> select * from table(dbms_xplan.display_cursor('gw1fks2wb4j2g',0));

SQL_ID  gw1fks2wb4j2g, child number 0
-------------------------------------------------------------------------
select * from T where id = :x
---------------------------------------------------------------------------------------
|  Id   |  Operation            |  Name  |  Rows  |  Bytes |  Cost (%CPU)  |  Time    |
---------------------------------------------------------------------------------------
|   0   |  SELECT STATEMENT     |        |        |        |    292 (100)  |          |
|*  1   |    TABLE ACCESS FULL  |  T     |  68207 |  6927K |   292   (2)   | 00:00:04 |
---------------------------------------------------------------------------------------

SQL> select * from table(dbms_xplan.display_cursor('gw1fks2wb4j2g',1));

SQL_ID  gw1fks2wb4j2g, child number 1
-------------------------------------
select * from T where id = :x
---------------------------------------------------------------------------------------------------
|  Id   |  Operation                     |  Name  |  Rows  |  Bytes  |  Cost (%CPU)  |  Time      |
---------------------------------------------------------------------------------------------------
|   0   |  SELECT STATEMENT              |        |        |         |   2    (100)  |            |
|   1   |    TABLE ACCESS BY INDEX ROWID | T      |     1  |    104  |   2       (0) |  00:00:01  |
|*  2   |      INDEX RANGE SCAN          | T_IDX  |     1  |         |   1       (0) |  00:00:01  |
---------------------------------------------------------------------------------------------------

IF Bind Variable Peeking Is a Problem

Most of the time--almost all of the time, in fact--bind variable peeking works as the designers of the feature intended: in a positive, helpful manner. The query in the application uses bind values that drive it to one plan or the other consistently. It is only when the plan flip-flops between two radically different execution paths, and for some segment of users, that you have a really bad plan. In such cases, Oracle Database 11g might be the right answer for you, because it accommodates multiple plans.

But when bind variable peeking doesn't work "nicely" and Oracle Database 11g is not a viable solution for you, what can you do to get around bind variable peeking? I'll go through the ideas one by one.

Don't bind that query. Yes, you read that correctly--don't use a bind variable (yes, I wrote that--do not use a bind variable!) In my example, ID has two values--1 and 99. Let's say ID wasn't ID but rather was PROCESSED_FLAG and had two values--Y and N. Most of the records are PROCESSED_FLAG=Y; some of them are N. Some people query the processed records (Y=yes). They want a full scan. Others query the unprocessed records in order to process them (N=no). They want an index range scan.

Here my suggestion would be not to bind against PROCESSED_FLAG; instead, use literals. That way we have two queries and two plans.

This is also why I do not like CURSOR_SHARING=FORCE, because when a programmer does this (uses literals) and the DBA turns on CURSOR_SHARING=FORCE, we are back to the original problem! Bind variable peeking comes into play once again--there is one plan for all, and it depends on who runs the query first. (By the way, CURSOR_SHARING=SIMILAR would not have this side effect. It would choose not to bind against PROCESSED_FLAG based on the statistics).

Don't gather statistics that cause the plan to flip-flop. If there is only one plan that the optimizer would ever choose, then bind variable peeking will have no effect. If you were to execute the above example without gathering histograms, the plan would not flip-flop.

Use your domain knowledge of the data. If you know that when a certain query is executed with a date that falls within the last 30 days, the volume of data would be small and indexes would be used and that if the date is way in the past, the volume of data would be large and full scans would be used, then just use if/then/else. 

[[... if/then/else]]: 

if ( p_date > sysdate-30 )
then
  open l_cursor for 
  select * from t less_than_30 where ...
else
   open l_cursor for
   select * from t more_than_30 where ...
end if;
loop
   fetch l_cursor into ...; 
...

This works when you have good knowledge of your data. This approach is similar to what Oracle Database 11g is doing with intelligent cursor sharing.

Use CURSOR_SHARING=SIMILAR. If you use CURSOR_SHARING=SIMILAR, you can let Oracle Database decide what to bind and what not to bind. For more details on that, with a full example, see tkyte.blogspot.com/2007/09/sqltracetrue-part-two.html.

Use stored outlines. Stored outlines are also known as query plan stability. In a test environment, you would exercise the application by using appropriate or representative inputs. Additionally, you would be capturing stored outlines into the outline tables. You would then move these into production and have the application issue ALTER SESSION to use these stored outlines.

Using stored outlines, you--in effect--freeze the plans. (And you can stop gathering statistics as well, because you just froze the plans.)

Disable the bind peeking feature. You can disable bind peeking by using an undocumented init.ora parameter, but--because this really affects you only if you have statistics that cause plans to flip-flop--I would say, instead, that you should stop gathering those statistics to save the time and resources it takes, and you'll have achieved basically the same goal.

]]> Dataguard Configuration: LOG_ARCHIVE_CONFIG and VALID_FOR in Oracle 10G tag:www.heysky.net,2009:/digest//1.336 2009-07-14T05:23:54Z 2009-07-14T05:25:23Z Sky http://www.heysky.net
Subject: Dataguard Configuration: LOG_ARCHIVE_CONFIG and VALID_FOR in Oracle 10G
  Doc ID: 249594.1 Type: BULLETIN
  Modified Date : 12-MAY-2009 Status: PUBLISHED 

    Checked for relevance on 12-May-2009
    PURPOSE
    -------

    This document has been created to explain the new Dataguard Configuration 
    parameters that were added in Oracle 10G. The Dataguard configuration and 
    db_unique_name parameters have been added to remove some complexities, 
    and limitations inherent in previous versions of DataGuard.


 
    SCOPE & APPLICATION
    -------------------

    For use by Customers and Support analysts in understanding and using the 
    new features.


    Dataguard Configuration:  LOG_ARCHIVE_CONFIG and VALID_FOR in Oracle 10G 
    ------------------------------------------------------------------------

    The Data Guard Configuration feature allows the user to optionally identify 
    the current database as well as all the other databases in the configuration
    using the new LOG_ARCHIVE_CONFIG parameter.  When used, the 
    LOG_ARCHIVE_CONFIG parameter defines the list of databases in this 
    configuration.  Log transport will not be allowed to any database not on 
    the list .  

    Each database in the configuration will have a db_unique_name and the 
    configuration can include 1 or more db_unique_names.

    To set this up you will assign each database a unique name, example:


	db_unique_name = 'Chicago_Sales'  # Note this parameter is not dynamic

    You will then add all the databases you want to be in this configuration, 
    example:

	log_archive_config='DG_CONFIG=(Chicago_Sales,Denver_Sales)'


    In this Simple Configuration we will only allow log transport between these 
    2 databases, uniquely identified as: Chicago_Sales and Denver_Sales

    Following this you can now specify the db_unique_name for each log archive 
    destination. This will be used along with the VALID_FOR parameter to 
    determine what logs to transport dependent on the database's current ROLE
    in the configuration.  The advantage of this feature is that we can 
    switchover without having to change parameters to defer/enable destinations.

    Is should be noted that although the DG_CONFIG attribute is an optional 
    attribute of the optional LOG_ARCHIVE_CONFIG initialization parameter, 
    it must be set to enable the dynamic addition of a standby database to a 
    Data Guard configuration that has a Real Application Clusters primary 
    database running in either maximum protection or maximum availability mode. 


    How does the VALID_FOR Parameter Work ?
    ---------------------------------------

    The VALID_FOR Parameter states that a specific log transport destination is 
    VALID_FOR only when this database is in a certain ROLE (STANDBY/PRIMARY).

    In this way you can set up a transport destination that will only be 
    enabled when this db_unique_name is in a specific ROLE.

    The default setting is ..

	VALID_FOR=(ALL_LOGFILES, ALL_ROLES)

    That setting will state that the log_archive_dest_n is valid for online 
    redo log and standby redo log archival. It is also Valid if the database 
    is currently in PRIMARY or STANDBY ROLE.
    
    More simply put: always archive to this destination.


    The full syntax for a local archive destination would be like this..

	log_archive_dest_1='LOCATION=/u01/Sales/oradata/arch VALID_FOR=(ALL_LOGFILES,ALL_ROLES)' db_unique_name='Chicago_Sales'                     	                    db_unique_name=CHicago_Sales' 

    or if using a flash_recovery_area

	log_archive_dest_1='LOCATION=USE_DB_RECOVERY_FILE_DEST VALID_FOR=(ALL_LOGFILES,ALL_ROLES)' db_unique_name='Chicago_Sales'                     	            		    db_unique_name=Chicago_Sales' 

    The real use of this parameter is to dynamically enable/defer destinations 
    dependent on whether the database is mounted in the PRIMARY or STANDBY ROLE.

    In a real situation you may set..
	
	db_unique_name='Chicago_Sales'
	log_archive_config='DG_CONFIG=(Chicago_Sales,Denver_Sales)'
	log_archive_dest_2='SERVICE=Denver_Sales LGWR ASYNC REOPEN=10
             VALID_FOR=(ONLINE_LOGFILES,PRIMARY_ROLE) db_unique_name=Denver_Sales'

    So we would only ship redo to Denver_Sales if Chicago_sales is running in
    the PRIMARY ROLE. Clearly if you were to Switchover and Chicago_Sales would 
    take on the STANDBY_ROLE then this destination would become deferred 
    dynamically.


    What views show the Setup ?
    ---------------------------

    The view V$ARCHIVE_DEST has new columns to handle the new values.

	
	SQL> column DEST_NAME format A19
	SQL> column DB_UNIQUE_NAME format A15
	SQL> column VALID_NOW format A9
	SQL> column VALID_TYPE format A15
	SQL> column VALID_ROLE format A15
	SQL> select DEST_NAME,DB_UNIQUE_NAME,VALID_NOW as CURRENT,VALID_TYPE,VALID_ROLE
             from V$ARCHIVE_DEST WHERE DB_UNIQUE_NAME <> 'NONE';

	DEST_NAME           DB_UNIQUE_NAME  VALID_NOW  VALID_TYPE      VALID_ROLE
	------------------- --------------  ---------  --------------- ---------------
	LOG_ARCHIVE_DEST_2  Denver_Sales    YES        ONLINE_LOGFILE  PRIMARY_ROLE
	LOG_ARCHIVE_DEST_10 Chicago_sales   YES        ALL_LOGFILES    ALL_ROLES


    So this shows us that in fact both the destinations are currently in use 
    as the database Chicago_Sales is the Primary..  
    
    Note LOG_ARCHIVE_DEST_10 is the default log_archive_dest_n parameter used
    when making use of a flash recovery area.



RELATED DOCUMENTS
-----------------

Oracle® Data Guard Concepts and Administration 10g Release 1 (10.1)
Oracle® Data Guard Concepts and Administration 10g Release 2 (10.2)


]]> Comparison Between Features : RAC, Dataguard, Streams, Advanced Replication and Basic Replication tag:www.heysky.net,2009:/digest//1.335 2009-07-14T05:17:32Z 2009-07-14T05:19:02Z Sky http://www.heysky.net
Subject: Comparison Between Features : RAC, Dataguard, Streams, Advanced Replication and Basic Replication
  Doc ID: 370850.1 Type: BULLETIN
  Modified Date : 30-JAN-2009 Status: PUBLISHED

In this Document
  Purpose
  Scope and Application
  Comparison Between Features : RAC, Dataguard, Streams, Advance Replication and Basic Replication

Applies to:

Oracle Server - Enterprise Edition - Version: 9.0.1.4
Information in this document applies to any platform.
Database Features :

Real Application Clusters (RAC)
Dataguard : Physical Standby & Logical Standby
Streams
Advance Replication : Multi-Master Replication & Updatable Materialized Views
Basic Replication : Read-only Materialized Views

Purpose

This article describes similarities and differences between different features available for high availability and distributed systems.

It contains comparison between multiple features in matrix format

It would help users to select feature useful and feasible in their environment.

Scope and Application

This article is meant for DBA and Support Engineers. It does not require in-depth technical knowledge in any of these features.

The article contains overview of covered topics and not technical details/steps.

Comparison Between Features : RAC, Dataguard, Streams, Advance Replication and Basic Replication

 

        Features

Real  Application Clusters     (RAC)

Dataguard

Streams

Advance Replication

Basic Replication

Topic of Comparison

Physical  Standby

Logical   Standby

Multi-Master Replication

Updatable Materialized Views

Read-only Materialized Views

Description

Allows the Oracle database to run applications across a set of clustered servers.

Provides a physically identical copy of the primary database, with on disk database structures that are identical to the primary database on a block-for-block basis.

Contains the same logical information as the production database, although the physical organization and structure of the data can be different.

Enables information sharing in form of stream of messages. Enables the propagation and management of data, transactions, and events in a data stream.

(Also called peer-to-peer or n-way replication)  Enables multiple sites, acting as equal peers, to manage groups of replicated database objects.

Provides complete or partial copy (replica) of a target table from a single point in time. Enable users to work on a subset of a database while disconnected from the central database server.

Provides complete or partial read-only copy (replica) of a target table from a single point in time. Enable users to view a subset of a database while disconnected from the central database server.

Purpose

- High availability

- Scalability

- Redundancy during failures

- Data protection

- Disaster recovery

- High availability

- Data protection

- Disaster recovery

- Efficient use of redundant hardware

- Data distribution

- Data sharing

- Data sharing

- Data distribution

- Sharing subset of data with update access.

- Data distribution

- Sharing subset of data in read-only mode.

Hardware

All nodes must have hardware that runs same OS.

All sites must have hardware that runs same OS.

All sites must have hardware that runs same OS.

Servers with different hardware can be used.

Servers with different hardware can be used.

Servers with different hardware can be used.

Servers with different hardware can be used.

OS

Same OS on all nodes including Patchset release

Same OS. Patchset release can be different in different sites.

Same OS. Patchset release can be different in different sites.

Can be used with different OS

Can be used with different OS

Can be used with different OS

Can be used with different OS

Oracle Software

Same version on all nodes including Oracle Patchset release

Same version on all nodes including Oracle Patchset release

Same version on all nodes including Oracle Patchset release

Can be used with different Oracle versions

Can be used with different Oracle versions

Can be used with different Oracle versions

Can be used with different Oracle versions

Feature Specific Terms

Servers involved in RAC configuration are generally known as Nodes. High-speed link between nodes is called Interconnect.

Primary and Standby database servers are generally known as Primary Site and Standby Site respectively.

Primary and Standby database servers are generally known as Primary Site and Standby Site respectively.

Database from where changes are captured is called source or capture site. Database from where changes are applied is called destination or apply site.

One database where Replication Administrative activities can be performed is called Master Definition Site. All other replicated databases are called Master sites.

Database with Master table is called Master Site or Target Site. Database where Materialized view is created is called Materialized view site.

Database with Master table is called Master Site or Target Site. Database where Materialized view is created is called Materialized view site.

Database

Multiple instances linked by interconnect to share access to an Oracle database.

One production database and one or more physical standby databases.

One production database and one or more logical standby databases.

Data stream can propagate data either within a database or from one database to another.

Complete copy of replicated table is maintained in multiple databases.

Updatable Materialized view is created in database other than the one containing master table.

Read-only materialized view can be created either within a database or in another database.

Data Storage

Single database on shared storage.

Primary site and Standby sites have their own database.

Primary site and Standby sites have their own database.

Source and Destination can either be same database or they can be different databases.

All replicated master sites are having their own databases.

Master table and Materialized views are part of different databases.

Master table and Materialized views can either be in same database or they can be in different databases.

Logical Database Structure

As there is one database, there is one logical structure of the database.

As physical structure of the database is exactly (block-by-block) same, the logical structure also remains same in primary and standby databases.

When created, logical standby database has same structure as primary. Later, additional schema/objects can be created in logical standby database. However, original objects must remain same as primary.

Streams provide flexibility to have different structure of schema/objects at source and destination databases.  This can be accomplished by using transformation of messages.

Logical structure of replicated objects must be same. Owner of those objects must be same in all master sites.

Both row and column subsetting enable you to create materialized views that contain a partial copy of the data at a master table. However, they are always based on a single master table.

Both row and column subsetting enable you to create materialized views that contain a partial copy of the data at a master table. Read-only materialized views can be created using join between multiple master tables.

Architecture Overview

A cluster comprises multiple interconnected computers or servers that appear as if they are one server to end users and applications. RAC uses Oracle Clusterware for the infrastructure to bind multiple servers so that they operate as a single system. If one clustered server fails, the Oracle database will continue running on the surviving servers.

Standby database is kept synchronized with the primary database, though Redo Apply, which recovers the redo data, received from the primary database and applies the redo to the standby database. If the primary database becomes unavailable, standby database can be switched to the primary role.

Standby database is kept synchronized with the primary database though SQL Apply, which uses logminer, transforms the data from redo logs into SQL statements and then executing the SQL statements on the standby database. If the primary database becomes unavailable, standby database can be switched to the primary role.

Each unit of shared information is called a message. Streams can capture, stage, and manage messages in Queue. Messages can be DML, DDL and user-defined messages. Streams can propagate the messages from one queue to other queue. When messages reach a destination, Streams can consume them based on your specifications.

Internal triggers capture changes applied at one site. The trigger stores those captured transactions locally. The source master site pushes (propagates and applies) those transactions to destination site.

Updatable Materialized view is a view that stores view data in it's own storage. A push process that is same as multi-master replication pushes updated data from MV site. However, changes from master site are pulled by refresh site using refresh process.

Read-only Materialized view is a view that stores view data in it's own storage. Data in materialized view are refreshed by refresh process. Refresh process is initiated at Materialized view site. Refresh process pulls data from master table using SQL query that was used to create Materialized view.

Overview of Installation/Setup

RAC installation is a two-phase installation. In phase one, use Oracle Universal Installer (OUI) to install Oracle Clusterware. In second phase, use OUI to install the RAC software (I.e. Oracle RDBMS with RAC option). You must install Oracle Clusterware and Oracle RDBMS in separate home directories.

Prepare Primary Database by making required changes in parameters, logging, archiving etc. Create a Backup Copy of the Primary Database Datafiles. Create a Control File for the Standby Database. Prepare an Initialization Parameter File for the Standby Database. Copy Files from the Primary System to the Standby System. Set Up the Environment to Support the Standby Database. Start the Physical Standby Database in continuous recovery mode.

Verify prerequisites Logical Standby Database (e.g. Datatypes, Primary Key etc). Create a Physical Standby Database. Stop Redo Apply on the Physical Standby Database. Prepare the Primary Database to Support a Logical Standby Database. Convert to a Logical Standby Database. Adjust Initialization Parameters for the Logical Standby Database. Open the Logical Standby Database and then perform certain Post-Creation Steps.

Create streams administrator user in all databases. In source database, create Capture Process and Propagation schedule for propagation to destination database. Create Apply Process in destination database.  Start Supplemental Logging in source database. . Prepare source database/objects. Create copy of those objects in destination database using export/import (datapump or traditional) or using RMAN. Complete the instantiation of objects. Start Apply, Propagation and Capture processes.

Create replication administrator user at all the sites with required privileges. Create propagation from each site to all other sites. At one site create Master replication group. The group remains in quiesced state when created. This site becomes MDS. Add objects (to be replicated) in the group from MDS. Add all master sites in master group from MDS. Start replication by resuming replication group. It alters replication group from quiesced to normal state.

At master site, create replication administrator. At materialized view (MV) site, create MV administrator. Create propagation from MV site to master site. Create master group at master site and add master objects in master group. Create materialized views at MV site. Create MV group at MV site and add materialized views in the MV group. On the MV site, create refresh group and add materialized views in refresh group.  Start replication by resuming replication group.

If MV is being created in database other than the one containing master table, then create database link in MV database to point to master database for accessing master table. Create materialized view.

Database/Instance status

All or any node of RAC can have instance with database open for DML/DDL access.

Redo apply requires database in recovery mode. When Physical standby database is in recovery mode, it cannot be opened. When not in recovery mode, it can only be opened in read-only mode.

SQL Apply requires database open for running SQL statements. Hence, Logical standby database must be open in normal circumstances.

Apply process requires database open for running SQL statements. Hence, destination database must be open in normal circumstances.

Push job requires master sites to be open when it pushes transaction to other master sites. Hence, in normal circumstances, all the master databases must be open.

Push as well as refresh job requires master and MV sites to be open. Hence, in normal circumstances, the master and MV databases must be open.

Refresh job requires master and MV sites to be open. Hence, in normal circumstances, the master and MV databases must be open.

Restriction on Datatypes

As there is one database, it supports all datatypes.

As physical structure of the database is exactly (block-by-block) same, it supports all datatypes.

There is restriction on datatypes allowed in logical standby setup.

There is restriction on datatypes allowed in streams setup.

There is restriction on datatypes allowed in replicated tables.

There is restriction on datatypes allowed in materialized views.

There is restriction on datatypes allowed in materialized views.

]]> Creating physical standby using RMAN duplicate without shutting down the primary tag:www.heysky.net,2009:/digest//1.334 2009-07-14T05:11:39Z 2009-07-14T05:14:30Z Sky http://www.heysky.net
Subject: Creating physical standby using RMAN duplicate without shutting down the primary
  Doc ID: 789370.1 Type: HOWTO
  Modified Date : 23-JUN-2009 Status: MODERATED

In this Document
  Goal
  Solution
  References

This document is being delivered to you via Oracle Support's Rapid Visibility (RaV) process, and therefore has not been subject to an independent technical review.

Applies to:

Oracle Server - Enterprise Edition - Version: 10.2.0.1 to 10.2.0.4
Information in this document applies to any platform.
Oracle Server Enterprise Edition - Version: 10.2.0.1 to 10.2.0.4

Goal

The following note describes step-by-step procedure to create physical standby by using RMAN duplicate without shutting down the primary (Production) database.

Database Name :- prim
Primary db_unique_name :- prim
standby db_unique_name :- stdby

Primary Hostname :- raca.idc.oracle.com
standby Hostname :- core1.idc.oracle.com

Solution

STEP: 1

Enable Force Logging on primary,

SQL> ALTER DATABASE FORCE LOGGING;

Database altered.

 

NOTE: Create password file if not present, also check if archiving enabled.

STEP: 2

Configure a Standby Redo Log on primary,

SQL> ALTER DATABASE ADD STANDBY LOGFILE GROUP 3 '/u01/app/oracle/databases/prim/redo/log3a.log' size 50m;

Database altered.

SQL> ALTER DATABASE ADD STANDBY LOGFILE GROUP 4 '/u01/app/oracle/databases/prim/redo/log4a.log' size 50m;

Database altered.

SQL> ALTER DATABASE ADD STANDBY LOGFILE GROUP 5 '/u01/app/oracle/databases/prim/redo/log5a.log' size 50m;

Database altered.

SQL> ALTER DATABASE ADD STANDBY LOGFILE GROUP 6 '/u01/app/oracle/databases/prim/redo/log6a.log' size 50m;

Database altered.

 

NOTE: To check the number of SRL,

(maximum number of logfiles for each thread + 1) * maximum number of threads

For example, if the primary database has 3 log files for each thread and 2 threads, then 8 standbys redo log file groups are needed on the standby database.

Verify the standby redo log file groups were created.

SQL> SELECT GROUP#,ThREAD#,SEQUENCE#,ARCHIVED,STATUS FROM V$STANDBY_LOG;


STEP :3

Modify the primary initialization parameter for dataguard on primary,

SQL>alter system set LOG_ARCHIVE_CONFIG='DG_CONFIG=(prim,stdby)';

System altered.


SQL> alter system set LOG_ARCHIVE_DEST_1='LOCATION=/u01/app/oracle/databases/prim/redo/ VALID_FOR=(ALL_LOGFILES,ALL_ROLES) DB_UNIQUE_NAME=prim';

System altered.


SQL>alter system set LOG_ARCHIVE_DEST_2='SERVICE=stdby LGWR ASYNC VALID_FOR=(ONLINE_LOGFILES,PRIMARY_ROLE) DB_UNIQUE_NAME=stdby';

System altered.


SQL> alter system set LOG_ARCHIVE_DEST_STATE_1=ENABLE;

System altered.


SQL>alter system set FAL_SERVER=stdby;

System altered.


SQL>alter system set FAL_CLIENT=prim;

System altered.


SQL>alter system set DB_FILE_NAME_CONVERT='/u01/app/oracle/databases/prim/data/','/u01/app/oracle/databases/stdby/data' scope=spfile;

System altered.


SQL>alter system set LOG_FILE_NAME_CONVERT='/u01/app/oracle/databases/prim/redo/','/u01/app/oracle/databases/stdby/redo' scope=spfile;

System altered.

STEP:4

Run the backup job at the primary by connecting to target and catalog DB(if any)

run
{
allocate channel c1 type disk;
allocate channel c2 type disk;
backup database format '/u01/app/oracle/databases/stage/%U';
backup archivelog all format '/u01/app/oracle/databases/stage/%U';
backup current controlfile for standby format '/u01/app/oracle/databases/stage/%U';
}


using target database control file instead of recovery catalog
allocated channel: c1
channel c1: sid=22 devtype=DISK

allocated channel: c2
channel c2: sid=32 devtype=DISK

Starting backup at 05-MAR-09
channel c1: starting full datafile backupset
channel c1: specifying datafile(s) in backupset
input datafile fno=00001 name=/u01/app/oracle/databases/prim/data/system01.dbf
input datafile fno=00002 name=/u01/app/oracle/databases/prim/data/undotbs01.dbf
channel c1: starting piece 1 at 05-MAR-09
channel c2: starting full datafile backupset
channel c2: specifying datafile(s) in backupset
input datafile fno=00003 name=/u01/app/oracle/databases/prim/data/sysaux01.dbf
channel c2: starting piece 1 at 05-MAR-09
channel c1: finished piece 1 at 05-MAR-09
piece handle=/u01/app/oracle/databases/stage/0tk95ldn_1_1 tag=TAG20090305T143325 comment=NONE
channel c1: backup set complete, elapsed time: 00:01:56
channel c1: starting full datafile backupset
channel c1: specifying datafile(s) in backupset
including current control file in backupset
channel c1: starting piece 1 at 05-MAR-09
channel c2: finished piece 1 at 05-MAR-09
piece handle=/u01/app/oracle/databases/stage/0uk95ldn_1_1 tag=TAG20090305T143325 comment=NONE
channel c2: backup set complete, elapsed time: 00:01:58
channel c2: starting full datafile backupset
channel c2: specifying datafile(s) in backupset
channel c1: finished piece 1 at 05-MAR-09
piece handle=/u01/app/oracle/databases/stage/0vk95lhb_1_1 tag=TAG20090305T143325 comment=NONE
channel c1: backup set complete, elapsed time: 00:00:03
including current SPFILE in backupset
channel c2: starting piece 1 at 05-MAR-09
channel c2: finished piece 1 at 05-MAR-09
piece handle=/u01/app/oracle/databases/stage/10k95lhd_1_1 tag=TAG20090305T143325 comment=NONE
channel c2: backup set complete, elapsed time: 00:00:02
Finished backup at 05-MAR-09

Starting backup at 05-MAR-09
current log archived
channel c1: starting archive log backupset
channel c1: specifying archive log(s) in backup set
input archive log thread=1 sequence=5 recid=1 stamp=679686819
input archive log thread=1 sequence=6 recid=2 stamp=679686958
input archive log thread=1 sequence=7 recid=3 stamp=679687040
input archive log thread=1 sequence=8 recid=4 stamp=679743155
input archive log thread=1 sequence=9 recid=5 stamp=680047441
input archive log thread=1 sequence=10 recid=6 stamp=680621993
channel c1: starting piece 1 at 05-MAR-09
channel c2: starting archive log backupset
channel c2: specifying archive log(s) in backup set
input archive log thread=1 sequence=11 recid=7 stamp=680636808
input archive log thread=1 sequence=12 recid=22 stamp=680703332
input archive log thread=1 sequence=13 recid=32 stamp=680704306
input archive log thread=1 sequence=14 recid=34 stamp=680704331
input archive log thread=1 sequence=15 recid=36 stamp=680704336
input archive log thread=1 sequence=16 recid=38 stamp=680705571
input archive log thread=1 sequence=17 recid=51 stamp=680711731
channel c2: starting piece 1 at 05-MAR-09
channel c2: finished piece 1 at 05-MAR-09
piece handle=/u01/app/oracle/databases/stage/12k95lhl_1_1 tag=TAG20090305T143531 comment=NONE
channel c2: backup set complete, elapsed time: 00:00:26
channel c1: finished piece 1 at 05-MAR-09
piece handle=/u01/app/oracle/databases/stage/11k95lhl_1_1 tag=TAG20090305T143531 comment=NONE
channel c1: backup set complete, elapsed time: 00:01:01
Finished backup at 05-MAR-09

Starting backup at 05-MAR-09
channel c1: starting full datafile backupset
channel c1: specifying datafile(s) in backupset
including standby control file in backupset
channel c1: starting piece 1 at 05-MAR-09
channel c1: finished piece 1 at 05-MAR-09
piece handle=/u01/app/oracle/databases/stage/13k95lji_1_1 tag=TAG20090305T143634 comment=NONE
channel c1: backup set complete, elapsed time: 00:00:02
Finished backup at 05-MAR-09
released channel: c1
released channel: c2

STEP :5

Create parameter file on primary copy it to standby and make the necessary changes,

SQL>create pfile='/u01/app/oracle/databases/prim/stage/initstdby.ora' from spfile;

File created.

STEP :6

Copy the listener.ora,tnsname.ora and sqlnet.ora files into staging directory.

[oracle@raca prim]$ cp /u01/app/oracle/network/admin/*.ora /u01/app/oracle/databases/prim/stage/


Copy the redo logs also to standby to avoid size mismatch,

[oracle@raca prim]$ cp /u01/app/oracle/databases/prim/redo/*.log /u01/app/oracle/databases/prim/stage/

STEP : 7

Move the files to standby server,

a. Create the same directory on standby server and copy the backups.

os standby,

$mkdir -p /u01/app/oracle/databases/stage/

[oracle@raca prim]$ scp /u01/app/oracle/databases/stage/* core1.idc.oracle.com:/u01/app/oracle/databases/stage/


b. Alternatively we can copy the backups to standby different directory. On standby server connect to RMAN target as primary and auxiliary instance or if your primary is having catalog database then connect to target as primary, catalog database and auxiliary instance. Catalog those backup pieces to let the controlfile of primary or catalog database to know the backup information.


For details on how to catalog backup piece refer,

Note 470463.1 - How To Catalog Backups / Archivelogs / Datafile Copies / Controlfile Copies

STEP: 8

Make the necessary changes on the copied initstdby.ora file on standby.

db_name=prim
db_unique_name=stdby
log_archive_config='DG_CONFIG=(prim,stdby)'
log_archive_dest_1='LOCATION=/u01/app/oracle/databases/stdby/redo/ VALID_FOR=(ALL_LOGFILES,ALL_ROLES) DB_UNIQUE_NAME=stdby'
log_archive_dest_2='SERVICE=prim LGWR ASYNC VALID_FOR=(ONLINE_LOGFILES,PRIMARY_ROLE) DB_UNIQUE_NAME=prim'
log_archive_dest_state_1=ENABLE
db_file_name_convert='/u01/app/oracle/databases/prim/data/','/u01/app/oracle/databases/stdby/data/'
log_file_name_convert='/u01/app/oracle/databases/prim/redo/','/u01/app/oracle/databases/stdby/redo/'
REMOTE_LOGIN_PASSWORDFILE=EXCLUSIVE
standby_archive_dest='/u01/app/oracle/databases/stdby/arch/'
standby_file_management=AUTO
fal_client=stdby
fal_server=prim

STEP: 9

Create standby instance,

create the same password as that of primary. Alternatively we can copy the password file from primary and rename it on standby.

For example from primary copry the password file,


scp /u01/app/oracle/dbs/opapwprim core1.idc.oracle.com:/u01/app/oracle/dbs/

on standby,

$mv orapwprim orapwstdby

or create new password same as primary as mentioned below,


a. in UNIX,

$export ORACLE_SID
$orapwd file='$ORACLE_HOME/dbs/orapwstdby' password=sys entries=10;

 in Windows,

d:>set ORACLE_SID=stdby
d:>oradim -new -sid stdby -intpwd sys


b Create necessary directories and start the auxiliary instance on standby 

[oracle@core1 stdby]$ pwd
/u01/app/oracle/databases/stdby

[oracle@core1 stdby]$mkdir data redo oradata stage udump bdump cdump

SQL>create spfile from pfile='/u01/app/oracle/databases/stage/initstdby.ora';

File created.

SQL>start nomount;

ORACLE instance started.

Total System Global Area 612368384 bytes
Fixed Size 1220844 bytes
Variable Size 167776020 bytes
Database Buffers 436207616 bytes
Redo Buffers 7163904 bytes


SQL>exit


STEP :10

Create net services on both primary and standby,

At prim server,

prim =
(DESCRIPTION =
(ADDRESS_LIST =
(ADDRESS = (PROTOCOL = TCP)(HOST =raca.idc.oracle.com )(PORT = 1521))
)
(CONNECT_DATA =
(SERVICE_NAME = prim)
)
)

stdby =
(DESCRIPTION =
(ADDRESS_LIST =
(ADDRESS = (PROTOCOL = TCP)(HOST =core1.idc.oracle.com )(PORT = 1521))
)
(CONNECT_DATA =
(SERVICE_NAME = stdby)
)
)


At standby server,

prim =
(DESCRIPTION =
(ADDRESS_LIST =
(ADDRESS = (PROTOCOL = TCP)(HOST =raca.idc.oracle.com )(PORT = 1521))
)
(CONNECT_DATA =
(SERVICE_NAME = prim)
)
)

stdby =
(DESCRIPTION =
(ADDRESS_LIST =
(ADDRESS = (PROTOCOL = TCP)(HOST =core1.idc.oracle.com )(PORT = 1521))
)
(CONNECT_DATA =
(SERVICE_NAME = stdby)
)
)

STEP : 11  Use RMAN duplicate for standby on standby,

When 'dorecover' is specified in the duplicate for stanby command then do a archivelog switch on the target database and run the duplicate.

Example :

on primary,

sql> alter system archive log current;

on standby,

Connect to target(i.e. prim via connect string) and catalog database(if any) and this auxiliary database by,

$RMAN target /@prim catalog RMAN/RMAN@catdb auxiliary /

if no catalog database,

$RMAN target sys/<password>@prim auxiliary /

Recovery Manager: Release 10.2.0.1.0 - Production on Thu Mar 5 12:30:56 2009

Copyright (c) 1982, 2005, Oracle. All rights reserved.

connected to target database: PRIM (DBID=3971986030)
connected to auxiliary database: PRIM (DBID=3971986030, not open)

RMAN> duplicate target database for standby dorecover;

Starting Duplicate Db at 05-MAR-09
using target database control file instead of recovery catalog
allocated channel: ORA_AUX_DISK_1
channel ORA_AUX_DISK_1: sid=36 devtype=DISK

contents of Memory Script:
{
set until scn 503194;
restore clone standby controlfile;
sql clone 'alter database mount standby database';
}
executing Memory Script

executing command: SET until clause

Starting restore at 05-MAR-09
using channel ORA_AUX_DISK_1

channel ORA_AUX_DISK_1: starting datafile backupset restore
channel ORA_AUX_DISK_1: restoring control file
channel ORA_AUX_DISK_1: reading from backup piece /u01/app/oracle/databases/prim/stage/0pk86cs2_1_1
channel ORA_AUX_DISK_1: restored backup piece 1
piece handle=/u01/app/oracle/databases/prim/stage/0pk86cs2_1_1 tag=TAG20090224T175722
channel ORA_AUX_DISK_1: restore complete, elapsed time: 00:00:01
output filename=/u01/app/oracle/databases/stdby/oradata/control01.ctl
Finished restore at 05-MAR-09

sql statement: alter database mount standby database
released channel: ORA_AUX_DISK_1

contents of Memory Script:
{
set until scn 503194;
set newname for datafile 1 to
"/u01/app/oracle/databases/stdby/data/system01.dbf";
set newname for datafile 2 to
"/u01/app/oracle/databases/stdby/data/undotbs01.dbf";
set newname for datafile 3 to
"/u01/app/oracle/databases/stdby/data/sysaux01.dbf";
restore
check readonly
clone database
;
}
executing Memory Script

executing command: SET until clause

executing command: SET NEWNAME

executing command: SET NEWNAME

executing command: SET NEWNAME

Starting restore at 05-MAR-09
allocated channel: ORA_AUX_DISK_1
channel ORA_AUX_DISK_1: sid=36 devtype=DISK

channel ORA_AUX_DISK_1: starting datafile backupset restore
channel ORA_AUX_DISK_1: specifying datafile(s) to restore from backup set
restoring datafile 00003 to /u01/app/oracle/databases/stdby/data/sysaux01.dbf
channel ORA_AUX_DISK_1: reading from backup piece /u01/app/oracle/databases/stage/0uk95ldn_1_1
channel ORA_AUX_DISK_1: restored backup piece 1
piece handle=/u01/app/oracle/databases/stage/0uk95ldn_1_1 tag=TAG20090305T143325
channel ORA_AUX_DISK_1: restore complete, elapsed time: 00:00:07
channel ORA_AUX_DISK_1: starting datafile backupset restore
channel ORA_AUX_DISK_1: specifying datafile(s) to restore from backup set
restoring datafile 00001 to /u01/app/oracle/databases/stdby/data/system01.dbf
restoring datafile 00002 to /u01/app/oracle/databases/stdby/data/undotbs01.dbf
channel ORA_AUX_DISK_1: reading from backup piece /u01/app/oracle/databases/stage/0tk95ldn_1_1
channel ORA_AUX_DISK_1: restored backup piece 1
piece handle=/u01/app/oracle/databases/stage/0tk95ldn_1_1 tag=TAG20090305T143325
channel ORA_AUX_DISK_1: restore complete, elapsed time: 00:00:15
Finished restore at 05-MAR-09

contents of Memory Script:
{
switch clone datafile all;
}
executing Memory Script

datafile 1 switched to datafile copy
input datafile copy recid=4 stamp=680711065 filename=/u01/app/oracle/databases/stdby/data/system01.dbf
datafile 2 switched to datafile copy
input datafile copy recid=5 stamp=680711065 filename=/u01/app/oracle/databases/stdby/data/undotbs01.dbf
datafile 3 switched to datafile copy
input datafile copy recid=6 stamp=680711065 filename=/u01/app/oracle/databases/stdby/data/sysaux01.dbf

contents of Memory Script:
{
set until scn 503194;
recover
standby
clone database
delete archivelog
;
}
executing Memory Script

executing command: SET until clause

Starting recover at 05-MAR-09
using channel ORA_AUX_DISK_1

starting media recovery

channel ORA_AUX_DISK_1: starting archive log restore to default destination
channel ORA_AUX_DISK_1: restoring archive log
archive log thread=1 sequence=11
channel ORA_AUX_DISK_1: restoring archive log
archive log thread=1 sequence=12
channel ORA_AUX_DISK_1: restoring archive log
archive log thread=1 sequence=13
channel ORA_AUX_DISK_1: restoring archive log
archive log thread=1 sequence=14
channel ORA_AUX_DISK_1: restoring archive log
archive log thread=1 sequence=15
channel ORA_AUX_DISK_1: restoring archive log
archive log thread=1 sequence=16
channel ORA_AUX_DISK_1: restoring archive log
archive log thread=1 sequence=17
channel ORA_AUX_DISK_1: reading from backup piece /u01/app/oracle/databases/stage/12k95lhl_1_1
channel ORA_AUX_DISK_1: restored backup piece 1
piece handle=/u01/app/oracle/databases/stage/12k95lhl_1_1 tag=TAG20090305T143531
channel ORA_AUX_DISK_1: restore complete, elapsed time: 00:00:03
channel ORA_AUX_DISK_1: starting archive log restore to default destination
channel ORA_AUX_DISK_1: restoring archive log
archive log thread=1 sequence=8
channel ORA_AUX_DISK_1: restoring archive log
archive log thread=1 sequence=9
channel ORA_AUX_DISK_1: restoring archive log
archive log thread=1 sequence=10
channel ORA_AUX_DISK_1: reading from backup piece /u01/app/oracle/databases/stage/11k95lhl_1_1
channel ORA_AUX_DISK_1: restored backup piece 1
piece handle=/u01/app/oracle/databases/stage/11k95lhl_1_1 tag=TAG20090305T143531
channel ORA_AUX_DISK_1: restore complete, elapsed time: 00:00:15
archive log filename=/u01/app/oracle/databases/stdby/redo/1_8_679684398.dbf thread=1 sequence=8
channel clone_default: deleting archive log(s)
archive log filename=/u01/app/oracle/databases/stdby/redo/1_8_679684398.dbf recid=8 stamp=680711072
archive log filename=/u01/app/oracle/databases/stdby/redo/1_9_679684398.dbf thread=1 sequence=9
channel clone_default: deleting archive log(s)
archive log filename=/u01/app/oracle/databases/stdby/redo/1_9_679684398.dbf recid=10 stamp=680711078
archive log filename=/u01/app/oracle/databases/stdby/redo/1_10_679684398.dbf thread=1 sequence=10
channel clone_default: deleting archive log(s)
archive log filename=/u01/app/oracle/databases/stdby/redo/1_10_679684398.dbf recid=9 stamp=680711078
archive log filename=/u01/app/oracle/databases/stdby/redo/1_11_679684398.dbf thread=1 sequence=11
channel clone_default: deleting archive log(s)
archive log filename=/u01/app/oracle/databases/stdby/redo/1_11_679684398.dbf recid=6 stamp=680711066
archive log filename=/u01/app/oracle/databases/stdby/redo/1_12_679684398.dbf thread=1 sequence=12
channel clone_default: deleting archive log(s)
archive log filename=/u01/app/oracle/databases/stdby/redo/1_12_679684398.dbf recid=7 stamp=680711067
archive log filename=/u01/app/oracle/databases/stdby/redo/1_13_679684398.dbf thread=1 sequence=13
channel clone_default: deleting archive log(s)
archive log filename=/u01/app/oracle/databases/stdby/redo/1_13_679684398.dbf recid=4 stamp=680711066
archive log filename=/u01/app/oracle/databases/stdby/redo/1_14_679684398.dbf thread=1 sequence=14
channel clone_default: deleting archive log(s)
archive log filename=/u01/app/oracle/databases/stdby/redo/1_14_679684398.dbf recid=1 stamp=680711066
archive log filename=/u01/app/oracle/databases/stdby/redo/1_15_679684398.dbf thread=1 sequence=15
channel clone_default: deleting archive log(s)
archive log filename=/u01/app/oracle/databases/stdby/redo/1_15_679684398.dbf recid=3 stamp=680711066
archive log filename=/u01/app/oracle/databases/stdby/redo/1_16_679684398.dbf thread=1 sequence=16
channel clone_default: deleting archive log(s)
archive log filename=/u01/app/oracle/databases/stdby/redo/1_16_679684398.dbf recid=2 stamp=680711066
archive log filename=/u01/app/oracle/databases/stdby/redo/1_17_679684398.dbf thread=1 sequence=17
channel clone_default: deleting archive log(s)
archive log filename=/u01/app/oracle/databases/stdby/redo/1_17_679684398.dbf recid=5 stamp=680711066
media recovery complete, elapsed time: 00:00:01
Finished recover at 05-MAR-09
Finished Duplicate Db at 05-MAR-09

RMAN>exit

NOTE: We can use without dorecover clause also,

Example,

 RMAN>duplicate target database for standby;


STEP : 12 Start the MRP(managed recovery process) on standby,

SQL> select name,db_unique_name,database_role, from v$database;

NAME DB_UNIQUE_NAME DATABASE_ROLE
--------- ---------------- --------------
PRIM STDBY PHYSICAL STANDBY


SQL> alter database recover managed standby database disconnect;

Database altered.

STEP : 13

Enable the log_archive_dest_2 on primary, which is to send the logs to standby server.

SQL>alter system set log_archive_dest_state_2=enable;

System altered.

 

STEP :14

Check the standby whether it is in SYNC with primary,

A. Check the v$archived view on standby,

SQL> SELECT SEQUENCE#, FIRST_TIME, NEXT_TIME FROM V$ARCHIVED_LOG ORDER BY SEQUENCE#;

SEQUENCE# FIRST_TIM NEXT_TIME
---------- --------- ---------
29 06-MAR-09 06-MAR-09
30 06-MAR-09 06-MAR-09
31 06-MAR-09 06-MAR-09
32 06-MAR-09 06-MAR-09
33 06-MAR-09 06-MAR-09
34 06-MAR-09 06-MAR-09
35 06-MAR-09 06-MAR-09
36 06-MAR-09 06-MAR-09
37 06-MAR-09 06-MAR-09
38 06-MAR-09 06-MAR-09
39 06-MAR-09 06-MAR-09


B. Do the log switch on primary,

SQL> ALTER SYSTEM SWITCH LOGFILE;

System altered.

C. On standby,

SQL> SELECT SEQUENCE#, FIRST_TIME, NEXT_TIME FROM V$ARCHIVED_LOG ORDER BY SEQUENCE#;


SEQUENCE# FIRST_TIM NEXT_TIME
---------- --------- ---------
29 06-MAR-09 06-MAR-09
30 06-MAR-09 06-MAR-09
31 06-MAR-09 06-MAR-09
32 06-MAR-09 06-MAR-09
33 06-MAR-09 06-MAR-09
34 06-MAR-09 06-MAR-09
35 06-MAR-09 06-MAR-09
36 06-MAR-09 06-MAR-09
37 06-MAR-09 06-MAR-09
38 06-MAR-09 06-MAR-09
39 06-MAR-09 06-MAR-09

SEQUENCE# FIRST_TIM NEXT_TIME
---------- --------- ---------
40 06-MAR-09 07-MAR-09


SQL> select sequence#,applied from v$archived_log order by sequence#;


SEQUENCE# APP
---------- ---
29 YES
30 YES
31 YES
32 YES
33 YES
34 YES
35 YES
36 YES
37 YES
38 YES
39 YES

SEQUENCE# APP
---------- ---
40 YES

References

Note 183570.1 - Creating a Data Guard Database with RMAN (Recovery Manager) using Duplicate Command
Note 466321.1 - RMAN Duplicate For Standby Fails with Rman-06024: No Backup Or Copy Of The Control File Found To Restore
]]> 10gR2 Dataguard Content tag:www.heysky.net,2009:/digest//1.333 2009-07-14T05:03:51Z 2009-07-14T05:11:02Z Sky http://www.heysky.net
Subject: 10gR2 Dataguard Content
  Doc ID: 739396.1 Type: HOWTO
  Modified Date : 14-MAY-2009 Status: PUBLISHED

In this Document
  Goal
  Solution

Applies to:

Oracle Server - Enterprise Edition - Version: 10.2.0.1 to 10.2.0.4
Information in this document applies to any platform.

Goal

This is a place holder note for most of the dataguard content that can be accessed from one page

This specific note is for Database ver 10.2.x

Solution

Setup and Configuration of Data Guard

Note.343424. Creating a 10gr2 Data Guard Physical Standby database with Real-Time apply

Note.380449.1 Creating a RAC Physical Standby for a RAC Primary

Note.387339.1 Creating a Single Instance Physical Standby for a RAC Primary

Creating a RAC Logical Standby Database for a RAC Primary Database

Client Failover setup in Data Guard Configurations

Data Guard Fast-Start Failover - Best Practices

Setup Multiple Standby Databases - Best Practices

Patching, Upgrade and Migration

Note.278641.1 Applying Patchset with a 10g Physical Standby in Place

Note.278521.1 Upgrading to 10g with a Physical Standby in Place

Note 437276.1 Upgrading Oracle Database with a Logical Standby Database In Place

Note 756671.1 Oracle Recommended Patches -- Oracle Database

Rolling Database Upgrades Using Data Guard SQL Apply

Minimal Downtime Migration to ASM

Oracle 10g Upgrade Companion - for upgrading from Oracle9i to Oracle10g

Managing Data Guard - Physical

Adding a Datafile or Creating a Tablespace
Dropping Tablespaces and Deleting Datafiles

Renaming a Datafile in the Primary Database

Recovering Through the OPEN RESETLOGS Statement

Opening a Physical Standby Database for Read-Only Access

Note.382686.1 CAN DATAGUARD HAVE DIFFERENT SCHEMA ACCOUNT STATUS FOR PRIMARY AND STANDBY DB ?

Note. 394815.1 How to Rename the Datafiles When using OMF in Standby


Switchover and Failover


Note.414043.1 Transitions for Data Guard Configurations Using Mixed Oracle Binaries

Note.738642.1 Step by Step Guide on How To Reinstate Failed Primary Database into Physical Standby

Data Guard Switchover and Failover -  Best Practice

 
Troubleshooting


Note 814417.1 Information to gather and upload for Dataguard related problems

Note.241374.1 Script to Collect Data Guard Primary Site Diagnostic Information

Note.241438.1 Script to Collect Data Guard Physical Standby Diagnostic Information

Note.241512.1 Script to Collect Data Guard Logical Standby Diagnostic Information
Dataguard and RAC


Note.370434.1 How to make CRS aware of the role change in Data Guard environment?

Miscelleanous

Note 386830.1 ALERT: Bug 5380055 Corrupts Index Blocks Especially in Data Guard Configurations

Common Dataguard errors

ORA-16057 DGID from server not in Data Guard configuration
ORA-12154 TNS No Listener

 

 

 



]]> Solaris下调试程序方法：使用truss tag:www.heysky.net,2008:/digest//1.332 2008-09-23T10:07:31Z 2008-09-23T10:14:09Z Sky http://www.heysky.net From: http://blog.chinaunix.net/u1/42574/showart_489928.html

Solaris下调试程序方法：使用truss

truss 介绍

solaris下的truss跟linux下的strace功能一样，用来跟踪系统/库函数调用和系统接收到的信号。并把函数调用跟信号发送到标准出错上。

truss 使用方法

truss  -a  -e  -f  -rall  -wall  -p

truss  -a  -e  -f  -rall  -wall

-a        显示传递给exec函数的参数
-e        显示传递给exec函数的环境变量
-f         显示子进程
-rall     显示所有读取的数据(默认32bytes)
-wall    显示所有写的数据(默认32bytes)
-p        Hook到一个已存在的进程(必须是进程的所有者或者root)
-d        显示跟踪进程运行的时间
<program> 指定要运行的程序

truss 例子

# truss -rall -wall -f -p <PID>
# truss -rall -wall lsnrctl start
# truss -aef lsnrctl dbsnmp_start
0
# truss -d date

Base time stamp:  1066157908.5731  [ Tue Oct 14 14:58:28 EDT 2003 ]

 0.0000 execve("/usr/bin/date", 0xFFBEF29C, 0xFFBEF2A4)  argc = 1

 0.0449 mmap(0x00000000, 8192, PROT_READ|PROT_WRITE|PROT_EXEC,MAP_PRIVATE|MAP_ANON, -1, 0) = 0xFF3A0000

 0.0453 resolvepath("/usr/lib/ld.so.1", "/usr/lib/ld.so.1", 1023) = 16

 0.0457 open("/var/ld/ld.config", O_RDONLY)             Err#2 ENOENT

 0.0460 open("/usr/lib/libc.so.1", O_RDONLY)            = 3

 0.0463 fstat(3, 0xFFBEE9C4)                            = 0

 0.0464 mmap(0x00000000, 8192, PROT_READ|PROT_EXEC, MAP_PRIVATE, 3, 0) = 0xFF390000

 0.0466 mmap(0x00000000, 794624, PROT_READ|PROT_EXEC, MAP_PRIVATE, 3, 0) = 0xFF280000

 0.0470 mmap(0xFF33A000, 24652, PROT_READ|PROT_WRITE|PROT_EXEC,MAP_PRIVATE|MAP_FIXED, 3, 696320) = 0xFF33A000

 0.0474 munmap(0xFF32A000, 65536)                       = 0

 0.0479 memcntl(0xFF280000, 113332, MC_ADVISE, MADV_WILLNEED, 0, 0) = 0

 0.0481 close(3)                                        = 0

 0.0483 open("/usr/lib/libdl.so.1", O_RDONLY)           = 3

 0.0485 fstat(3, 0xFFBEE9C4)                            = 0

 0.0487 mmap(0xFF390000, 8192, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_FIXED,3, 0) = 0xFF390000

 0.0490 close(3)                                        = 0

 0.0493 open("/usr/platform/SUNW,Ultra-5_10/lib/libc_psr.so.1", O_RDONLY) = 3

 0.0496 fstat(3, 0xFFBEE854)                            = 0

 0.0497 mmap(0x00000000, 8192, PROT_READ|PROT_EXEC, MAP_PRIVATE, 3, 0) = 0xFF380000

 0.0500 mmap(0x00000000, 16384, PROT_READ|PROT_EXEC, MAP_PRIVATE, 3, 0) = 0xFF370000

 0.0502 close(3)                                        = 0

 0.0514 munmap(0xFF380000, 8192)                        = 0

 0.0521 brk(0x00022420)                                 = 0

 0.0523 brk(0x00024420)                                 = 0

 0.0526 time()                                          = 1066157908

 0.0531 open("/usr/share/lib/zoneinfo/US/Eastern", O_RDONLY) = 3

 0.0533 read(3, " T Z i f\0\0\0\0\0\0\0\0".., 8192)     = 1250

 0.0536 close(3)                                        = 0

 0.0542 ioctl(1, TCGETA, 0xFFBEEFDC)                    = 0

Tue Oct 14 14:58:28 EDT 2003

 0.0545 write(1, " T u e   O c t   1 4   1".., 29)      = 29

 0.0547 llseek(0, 0, SEEK_CUR)                          = 1829

 0.0549 _exit(0) ]]> Solaris平台上如何知道某个端口被哪个进程和应用程序占用 tag:www.heysky.net,2008:/digest//1.331 2008-09-23T08:48:39Z 2008-09-23T10:06:10Z Sky http://www.heysky.net From: http://blog.csdn.net/Adali/archive/2007/11/23/1899779.aspx

Solaris平台上如何知道某个端口被哪个进程和应用程序占用

我经常在Solaris服务器上启好几个Tomcat和GlassFish, 会遇到"端口"被占用的错误, 很难直接用命令得知这个端口到底被哪个进程或者应用程序占用了， Alex给了我三个解决方案。附带说一句，我只尝试了第一种方案，相当的好用。而Alex愿意花时间找解决方案，却不愿意花几分钟把这些好东西写出来，结果就是被我发表出来。功劳还是要归Alex Peng.

第一种方案：

1。使用下面shell script，先建立一个port.sh文件:

# more /tmp/port.sh

#!/bin/sh

for pid in `ls /proc`

do

       pf=`/usr/bin/pfiles $pid 2>/dev/null`

       if echo $pf | grep $1 > /dev/null 2>&1

       then

           echo $pid

           /usr/bin/pargs $pid

       fi

done

2。运行port.sh, 传入端口号，比如53250 :

#  /tmp/port.sh 53250

3。运行结果如下：

1225

1225:   /usr/lib/thunderbird/thunderbird-bin -UILocale zh-CN

-contentLocale CN

argv[0]: /usr/lib/thunderbird/thunderbird-bin

argv[1]: -UILocale

argv[2]: zh-CN

argv[3]: -contentLocale

argv[4]: CN

4212

4212:   /bin/sh /tmp/port.sh 53250

argv[0]: /bin/sh

argv[1]: /tmp/port.sh

argv[2]: 53250

 
第二种方案：

下载"lsof" package。但可能不适合每种情况


第三种方案：

使用MDB

   from socket info (netstat output), you can know its vnode

   from vnode info, you can know which process owns it

   from process info, you can know its args, so comes the result.

 
Reference

lsof：

http://dmiessler.com/study/lsof/
http://freshmeat.net/projects/lsof/

MDB：

Solaris™ Performance and Tools: DTrace and MDB Techniques for Solaris 10 and OpenSolaris

]]> Introduction to Unix Communication Tools tag:www.heysky.net,2008:/digest//1.330 2008-09-08T05:18:04Z 2008-09-08T05:39:51Z From: http://www.helpdesk.umd.edu/systems/glue/applications/pine/1112/ Introduction to Unix Communication Tools Logging in Remotely Transferring Files with Secure FTP Electronic Mail Real-Time Communication Refusing Real-Time Talks and Messages There are several communications and networking tools that are available on nearly every UNIX system. This... Sky http://www.heysky.net From: http://www.helpdesk.umd.edu/systems/glue/applications/pine/1112/

Introduction to Unix Communication Tools

There are several communications and networking tools that are available on nearly every UNIX system. This document will go over a handful of the more standard ones, including clients that allow you to login remotely, transfer files, send e-mail, and talk (or refuse to talk) in real-time.

]]>

 Logging in Remotely

The ones that are used most often are telnet and rlogin, which let you login to the system remotely. Telnet is used most often. There are telnet clients available for nearly any platform, though usually UNIX systems are the only ones running a telnet server. Telneting into a UNIX machine will allow you to use a command line shell, including full screen display programs if your terminal is set up properly. Rlogin (short for Remote-LOGIN) looks similar to telnet on the surface, but uses a different set of protocols. More information is exchanged between the client and server during an rlogin session than during a telnet session. This gives it more functionality, but also creates some security risks if not used carefully. One of the main features of rlogin is the .rhosts file. This is a file that a user can have in his or her home directory that lists a set of machines and usernames that are to be considered "trusted". This allows the user to login to one of the trusted machines, and then rlogin into the main machine without supplying a password. I'm sure it's obvious at this point why .rhosts files should be used with care. Besides rlogin, there is a whole suite of remote applications including rcp (remote copy), rsh (remote shell) and rexec (remote execute) that rely on the .rhosts file to determine trusted users and machines. Telnet and rlogin are both executed from the command line, with the remote host address as an argument.

File Transfer

Another useful tool is sftp, or Secure File Transfer Protocol. With an sftp client, you can log into any machine that is running an sftp server (no matter what platform) and transfer files back and forth. When you login using sftp, you do not get a shell command line, you stay within the sftp client, but you navigate the directories as if you were in a shell. Using the commands put and get you can transfer files to and from the remote host. One important thing to keep in mind is that executable programs, compressed files, and any other binary files must be transferred in binary mode. Most sftp clients default to ascii (text) mode, so to be safe type binary at the sftp> prompt. Some sftp clients will determine the mode automatically. SFTP sessions can also be automated from a UNIX machine by using macros in your .netrc file.

Electronic Mail

There are several mailers available for UNIX systems. The standard mailer we use is Mail (invoked by mail --note the case; you see, our systems have 'mail' aliased to /usr/ucb/Mail in the default login files). Mail is very simple to use; essentially, you type 'mail' and you read any mail messages waiting for you. Sending mail is like- mail userid1 userid2...useridn. Note: do NOT put the carbon copies here; you put them at the end of the mail session, or inside the mail item (see below). If you put it on the command line, you try to send mail to user "cc:" which usually does not exist.

An alternative to the standard mailer is pine. Pine uses a full screen display, which most users find a lot easier to use than mail. It saves mail in "folders", giving you an inbox, saved-mail, and sent-mail folder by default, and allowing you to define other folders to categorize your mail with.

Another popular mail system available on our UNIX systems is MH. MH differs from Mail in that MH is a collection of small programs that you execute from the shell. MH is more powerful and customizable than Mail, and is on all of the CSC UNIX systems.

Real-Time Communication

On UNIX machines, you have two ways of communicating in real-time. The most preferred way is 'talk.' Talk is a two-way real-time communications program, letting you see what the other person types character-by-character (as opposed to the IBM mainframe, which is line by line). To initiate a talk, type talk userid@address [tty] (the @address part is not needed if you are talking to someone on the same machine as yourself, and the tty part is only necessary if you wish the message to go to a specific tty and not the first one the system finds). The other party will get a talk daemon message: 

Message from Talk_Daemon@lewhoosh.umd.edu at 17:46 ... 
talk: connection requested by matthews@lewhoosh.umd.edu. 
talk: respond with:  talk matthews@lewhoosh.umd.edu

...and he/she can either ignore the message and be bothered for a while, or respond with a talk command (the system will tell you what to type to answer the request -- "respond with:..."). We will ignore the fact that talking to oneself is a sign of impending doom. To end the talk session, just hit ^C (if nobody answers for a few 'rings,' then it's safe to assume they aren't going to). Note that your terminal must be set correctly (including type, number of rows, and number of columns) for a full- screen environment, else ... well, let's just say you won't like the results otherwise and leave it at that. If you get a message: 

[Checking for invitation on caller's machine]

that could mean one of two things. One: The other machine is a Sun workstation running BSD 4.2 talk (while the rest of the world is running BSD 4.3 talk) and the protocol is not compatible. You can try to use the command "otalk" instead, for Old talk. If that doesn't work, then Two applies. And Two is: something is muffed up. It could be the network is down, or it could be the other system is having problems firing up the talk daemon. Either way it's related to the protocol not being met and there's nothing you can do about this situation except wait. If you get the message: 

Target machine is too confused to talk to us

then you've discovered a funny error message most likely written by someone strung out on caffeine at 3 in the morning.

The other method of real-time communications on UNIX is the 'write' command. This will send a one-way message. The format is pretty much the same as the IBM mainframe's TELL command: write  <userid >[tty], where <userid >is the user you want to communicate with and the optional tty specifies which tty to send the message to if/when the user is logged in more than once (otherwise, write will send it to the first one it finds). Note that write will not work across systems, so you have to be logged in to the same machine your recipient is on. When you first issue the command, the recipient will get a message telling him/her you are sending him/her a message. Then you type in your message and end it with a ^d (UNIX EOF), and the recipient will see it (they will get it line by line as well). You can also use i/o redirection (write <userid >< message.file or echo <message>| write <userid>).

How Not to Receive Messages

The commands for enabling/disabling messages to your tty, and thus your screen, are mesg y/n. Some people like to put mesg nin their ~/.login file just to make sure that people won't interrupt them. This is the default setting for the WAM machines, so in order to receive real-time messages on WAM you need to give a "mesg y" command.

The command who will tell you who is on the machine that you are on, or you could use

finger @[machine-name]

to see who is on a remote machine. In order to make sure you are not interrupting someone who is busy, the "w" command will tell you what people on your machine are doing.

One little bit of netiquette: try not to send writes or talk requests to people who do not want to be disturbed! This includes sending random messages to people you don't know just because you're bored, and also sending them to anybody who is using a full screen program, such as a text editor or maybe another talk session.

 

From: http://en.wikipedia.org/wiki/Write_(Unix))

write (Unix)

From Wikipedia, the free encyclopedia

write can refer to several Unix commands. All known variations of write are used to write messages to another user. The most popular variation sends a message directly to another user's TTY

Usage

The correct syntax for the write command is:

write user [tty]
message

The write session is terminated by sending EOF, which can be done by pressing Ctrl+D. The tty argument is only necessary when a user is logged into more than one terminal.

Example

A conversation initiated between two users on the same machine:

11:19 AM# write root pts/7
test

Will show up to the user on that console as:

Message from root@punch on pts/8 at 11:19 ...
test

History

A version of the write command appeared in the First Edition of the Research Unix operating system. Another variation of write writes a message to a user on a Windows network, using the SMB packet format.

Programmer's Workbench UNIX contained a program wall that wrote a message to all users in the same way.

See also

]]> Automatic SQL Tuning and SQL Profiles Internals in Oracle 10g tag:www.heysky.net,2008:/digest//1.329 2008-08-03T15:00:39Z 2008-08-03T16:14:52Z From: http://www.oracle-base.com/articles/10g/AutomaticSQLTuning10g.phpAutomatic SQL Tuning in Oracle Database 10g This article the discusses the new features which automate the tuning of SQL statements in Oracle 10g: OverviewSQL Tuning AdvisorManaging SQL ProfilesSQL Tuning SetsUseful Views Overview In its normal mode the query... Sky http://www.heysky.net From: http://www.oracle-base.com/articles/10g/AutomaticSQLTuning10g.php

Automatic SQL Tuning in Oracle Database 10g

This article the discusses the new features which automate the tuning of SQL statements in Oracle 10g:

Overview

In its normal mode the query optimizer needs to make decisions about execution plans in a very short time. As a result it may not always be able to obtain enough information to make the best decision. Oracle 10g allows the optimizer to run in tuning mode where it can gather additional information and make recommendations about how specific statements can be tuned further. This process may take several minutes for a single statement so it is intended to be used on high-load resource-intensive statements.

]]>
  • Statistics Analysis - The optimizer recommends the gathering of statistics on objects with missing or stale statistics. Additional statistics for these objects are stored in an SQL profile.

  • SQL Profiling - The optimizer may be able to improve performance by gathering additional statistics and altering session specific parameters such as the OPTIMIZER_MODE. If such improvements are possible the information is stored in an SQL profile. If accepted this information can then used by the optimizer when running in normal mode. Unlike a stored outline which fixes the execution plan, an SQL profile may still be of benefit when the contents of the table alter drastically. Even so, it's sensible to update profiles periodically. The SQL profiling is not performed when the tuining optimizer is run in limited mode.

  • Access Path Analysis - The optimizer investigates the effect of new or modified indexes on the access path. It's index recommendations relate to a specific statement so where necessary it will also suggest the use of the SQL Access Advisor to check the impact of these indexes on a representative SQL workload.

  • SQL Structure Analysis - The optimizer suggests alternatives for SQL statements that contain structures that may impact on performance. The implementation of these suggestions requires human intervention to check their validity.
The automatic SQL tuning features are accessible from Enterprise Manager on the "Advisor Central" page these or from PL/SQL using the DBMS_SQLTUNE package. This article will focus on the PL/SQL API as the Enterprise Manager interface is reasonably intuative.

SQL Tuning Advisor

In order to access the SQL tuning advisor API a user must be granted the ADVISOR privilege:
CONN sys/password AS SYSDBA
GRANT ADVISOR TO scott;
CONN scott/tiger
The first step when using the SQL tuning advisor is to create a new tuning task using the CREATE_TUNING_TASK function. The statements to be analyzed can be retrieved from the Automatic Workload Repository (AWR), the cursor cache, an SQL tuning set or specified manually:
SET SERVEROUTPUT ON

-- Tuning task created for specific a statement from the AWR.
DECLARE
  l_sql_tune_task_id  VARCHAR2(100);
BEGIN
  l_sql_tune_task_id := DBMS_SQLTUNE.create_tuning_task (
                          begin_snap  => 764,
                          end_snap    => 938,
                          sql_id      => '19v5guvsgcd1v',
                          scope       => DBMS_SQLTUNE.scope_comprehensive,
                          time_limit  => 60,
                          task_name   => '19v5guvsgcd1v_AWR_tuning_task',
                          description => 'Tuning task for statement 19v5guvsgcd1v in AWR.');
  DBMS_OUTPUT.put_line('l_sql_tune_task_id: ' || l_sql_tune_task_id);
END;
/

-- Tuning task created for specific a statement from the cursor cache.
DECLARE
  l_sql_tune_task_id  VARCHAR2(100);
BEGIN
  l_sql_tune_task_id := DBMS_SQLTUNE.create_tuning_task (
                          sql_id      => '19v5guvsgcd1v',
                          scope       => DBMS_SQLTUNE.scope_comprehensive,
                          time_limit  => 60,
                          task_name   => '19v5guvsgcd1v_tuning_task',
                          description => 'Tuning task for statement 19v5guvsgcd1v.');
  DBMS_OUTPUT.put_line('l_sql_tune_task_id: ' || l_sql_tune_task_id);
END;
/

-- Tuning task created from an SQL tuning set.
DECLARE
  l_sql_tune_task_id  VARCHAR2(100);
BEGIN
  l_sql_tune_task_id := DBMS_SQLTUNE.create_tuning_task (
                          sqlset_name => 'test_sql_tuning_set',
                          scope       => DBMS_SQLTUNE.scope_comprehensive,
                          time_limit  => 60,
                          task_name   => 'sqlset_tuning_task',
                          description => 'Tuning task for an SQL tuning set.');
  DBMS_OUTPUT.put_line('l_sql_tune_task_id: ' || l_sql_tune_task_id);
END;
/

-- Tuning task created for a manually specified statement.
DECLARE
  l_sql               VARCHAR2(500);
  l_sql_tune_task_id  VARCHAR2(100);
BEGIN
  l_sql := 'SELECT e.*, d.* ' ||
           'FROM   emp e JOIN dept d ON e.deptno = d.deptno ' ||
           'WHERE  NVL(empno, ''0'') = :empno';

  l_sql_tune_task_id := DBMS_SQLTUNE.create_tuning_task (
                          sql_text    => l_sql,
                          bind_list   => sql_binds(anydata.ConvertNumber(100)),
                          user_name   => 'scott',
                          scope       => DBMS_SQLTUNE.scope_comprehensive,
                          time_limit  => 60,
                          task_name   => 'emp_dept_tuning_task',
                          description => 'Tuning task for an EMP to DEPT join query.');
  DBMS_OUTPUT.put_line('l_sql_tune_task_id: ' || l_sql_tune_task_id);
END;
/
If the TASK_NAME parameter is specified it's value is returned as the SQL tune task identifier. If ommitted a system generated name like "TASK_1478" is returned. If the SCOPE parameter is set to scope_limited the SQL profiling analysis is omitted. The TIME_LIMIT parameter simply restricts the time the optimizer can spend compiling the recommendations.

The following examples will reference the last tuning set as it has no external dependancies other than the SCOTT schema. The NVL in the SQL statement was put in to provoke a reaction from the optimizer. In addition we can delete the statistics from one of the tables to provoke it even more:
EXEC DBMS_STATS.delete_table_stats('SCOTT','EMP');
With the tuning task defined the next step is to execute it using the EXECUTE_TUNING_TASK procedure:
EXEC DBMS_SQLTUNE.execute_tuning_task(task_name => 'emp_dept_tuning_task');
During the execution phase you may wish to pause and restart the task, cancel it or reset the task to allow it to be re-executed:
-- Interrupt and resume a tuning task.
EXEC DBMS_SQLTUNE.interrupt_tuning_task (task_name => 'emp_dept_tuning_task');
EXEC DBMS_SQLTUNE.resume_tuning_task (task_name => 'emp_dept_tuning_task');

-- Cancel a tuning task.
EXEC DBMS_SQLTUNE.cancel_tuning_task (task_name => 'emp_dept_tuning_task');

-- Reset a tuning task allowing it to be re-executed.
EXEC DBMS_SQLTUNE.reset_tuning_task (task_name => 'emp_dept_tuning_task');
The status of the tuning task can be monitored using the DBA_ADVISOR_LOG view:
SELECT task_name, status FROM dba_advisor_log WHERE owner = 'SCOTT';

TASK_NAME                      STATUS
------------------------------ -----------
emp_dept_tuning_task           COMPLETED

1 row selected.
Once the tuning task has executed successfully the recommendations can be displayed using the REPORT_TUNING_TASK function:
SET LONG 10000;
SET PAGESIZE 1000
SET LINESIZE 200
SELECT DBMS_SQLTUNE.report_tuning_task('emp_dept_tuning_task') AS recommendations FROM dual;
SET PAGESIZE 24
In this case the output looks like this:
RECOMMENDATIONS
--------------------------------------------------------------------------------
GENERAL INFORMATION SECTION
-------------------------------------------------------------------------------
Tuning Task Name   : emp_dept_tuning_task
Scope              : COMPREHENSIVE
Time Limit(seconds): 60
Completion Status  : COMPLETED
Started at         : 05/06/2004 09:29:13
Completed at       : 05/06/2004 09:29:15

-------------------------------------------------------------------------------
SQL ID  : 0wrmfv2yvswx1
SQL Text: SELECT e.*, d.* FROM   emp e JOIN dept d ON e.deptno = d.deptno
          WHERE  NVL(empno, '0') = :empno

-------------------------------------------------------------------------------
FINDINGS SECTION (2 findings)
-------------------------------------------------------------------------------

1- Statistics Finding
---------------------
  Table "SCOTT"."EMP" and its indices were not analyzed.

  Recommendation
  --------------
    Consider collecting optimizer statistics for this table and its indices.
    execute dbms_stats.gather_table_stats(ownname => 'SCOTT', tabname =>
            'EMP', estimate_percent => DBMS_STATS.AUTO_SAMPLE_SIZE,
            method_opt => 'FOR ALL COLUMNS SIZE AUTO', cascade => TRUE)

  Rationale
  ---------
    The optimizer requires up-to-date statistics for the table and its indices
    in order to select a good execution plan.

2- Restructure SQL finding (see plan 1 in explain plans section)
----------------------------------------------------------------
  The predicate NVL("E"."EMPNO",0)=:B1 used at line ID 2 of the execution plan
  contains an expression on indexed column "EMPNO". This expression prevents
  the optimizer from selecting indices on table "SCOTT"."EMP".

  Recommendation
  --------------
    Rewrite the predicate into an equivalent form to take advantage of
    indices. Alternatively, create a function-based index on the expression.

  Rationale
  ---------
    The optimizer is unable to use an index if the predicate is an inequality
    condition or if there is an expression or an implicit data type conversion
    on the indexed column.

-------------------------------------------------------------------------------
EXPLAIN PLANS SECTION
-------------------------------------------------------------------------------

1- Original
-----------
Plan hash value: 1863486531

----------------------------------------------------------------------------------------
| Id  | Operation                    | Name    | Rows  | Bytes | Cost (%CPU)| Time     |
----------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |         |     1 |   107 |     4   (0)| 00:00:01 |
|   1 |  NESTED LOOPS                |         |     1 |   107 |     4   (0)| 00:00:01 |
|   2 |   TABLE ACCESS FULL          | EMP     |     1 |    87 |     3   (0)| 00:00:01 |
|   3 |   TABLE ACCESS BY INDEX ROWID| DEPT    |     1 |    20 |     1   (0)| 00:00:01 |
|   4 |    INDEX UNIQUE SCAN         | PK_DEPT |     1 |       |     0   (0)| 00:00:01 |
----------------------------------------------------------------------------------------

Note
-----
   - dynamic sampling used for this statement

-------------------------------------------------------------------------------


1 row selected.
Once the tuning session is over the tuning task can be dropped using the DROP_TUNING_TASK procedure:
BEGIN
  DBMS_SQLTUNE.drop_tuning_task (task_name => '19v5guvsgcd1v_AWR_tuning_task');
  DBMS_SQLTUNE.drop_tuning_task (task_name => '19v5guvsgcd1v_tuning_task');
  DBMS_SQLTUNE.drop_tuning_task (task_name => 'sqlset_tuning_task');
  DBMS_SQLTUNE.drop_tuning_task (task_name => 'emp_dept_tuning_task');
END;
/

Managing SQL Profiles

To manage SQL profiles a user needs the following privileges:
CONN sys/password AS SYSDBA
GRANT CREATE ANY SQL PROFILE TO scott;
GRANT DROP ANY SQL PROFILE TO scott;
GRANT ALTER ANY SQL PROFILE TO scott;
CONN scott/tiger
If the recommendations of the SQL tuning advisor include a suggested profile you can choose to accept it using the ACCEPT_SQL_PROFILE procedure:
SET SERVEROUTPUT ON
DECLARE
  l_sql_tune_task_id  VARCHAR2(20);
BEGIN
  l_sql_tune_task_id := DBMS_SQLTUNE.accept_sql_profile (
                          task_name => 'emp_dept_tuning_task',
                          name      => 'emp_dept_profile');
  DBMS_OUTPUT.put_line('l_sql_tune_task_id: ' || l_sql_tune_task_id);
END;
/
The NAME parameter is used to specify a name for the profile. If it is not specified a system generated name will be used.

The STATUS, NAME, DESCRIPTION, and CATEGORY attributes of an SQL profile can be altered using the ALTER_SQL_PROFILE procedure:
BEGIN
  DBMS_SQLTUNE.alter_sql_profile (
    name            => 'emp_dept_profile',
    attribute_name  => 'STATUS',
    value           => 'DISABLED');
END;
/
Existing SQL profiles can be dropped using the DROP_SQL_PROFILE procedure:
BEGIN
  DBMS_SQLTUNE.drop_sql_profile (
    name   => 'emp_dept_profile',
    ignore => TRUE);
END;
/
The IGNORE parameter prevents errors being reported if the specified profile does not exist.

SQL Tuning Sets

An SQL tuning set is a group of statements along with their execution context. These can be created automatically via Enterprise Manager or manually provided you have the necessary privileges:
CONN sys/password AS SYSDBA
GRANT ADMINISTER ANY SQL TUNING SET TO scott;
CONN scott/tiger
An SQL tuning set is created using the CREATE_SQLSET procedure:
BEGIN
  DBMS_SQLTUNE.create_sqlset (
    sqlset_name  => 'test_sql_tuning_set',
    description  => 'A test SQL tuning set.');
END;
/
Statements are added to the set using the LOAD_SQLSET procedure which accepts a REF CURSOR of statements retrieved using one of the following pipelined functions:
  • SELECT_WORKLOAD_REPOSITORY - Retrieves statements from the Automatic Workload Repository (AWR).
  • SELECT_CURSOR_CACHE - Retrieves statements from the cursor cache.
  • SELECT_SQLSET - Retrieves statements from another SQL tuning set.
The following are examples of their usage:
-- Load the SQL set from the Automatic Workload Repository (AWR).
DECLARE
  l_cursor  DBMS_SQLTUNE.sqlset_cursor;
BEGIN
  OPEN l_cursor FOR
    SELECT VALUE(p)
    FROM   TABLE (DBMS_SQLTUNE.select_workload_repository (
                    765,  -- begin_snap
                    766,  -- end_snap
                    NULL, -- basic_filter
                    NULL, -- object_filter
                    NULL, -- ranking_measure1
                    NULL, -- ranking_measure2
                    NULL, -- ranking_measure3
                    NULL, -- result_percentage
                    10)   -- result_limit
                  ) p;

  DBMS_SQLTUNE.load_sqlset (
    sqlset_name     => 'test_sql_tuning_set',
    populate_cursor => l_cursor);
END;
/

-- Load the SQL set from the cursor cache.
DECLARE
  l_cursor  DBMS_SQLTUNE.sqlset_cursor;
BEGIN
  OPEN l_cursor FOR
    SELECT VALUE(p)
    FROM   TABLE (DBMS_SQLTUNE.select_cursor_cache (
                    NULL, -- basic_filter
                    NULL, -- object_filter
                    NULL, -- ranking_measure1
                    NULL, -- ranking_measure2
                    NULL, -- ranking_measure3
                    NULL, -- result_percentage
                    1)    -- result_limit
                  ) p;

  DBMS_SQLTUNE.load_sqlset (
    sqlset_name     => 'test_sql_tuning_set',
    populate_cursor => l_cursor);
END;
/

-- Create a new set and load it from the existing one.
DECLARE
  l_cursor  DBMS_SQLTUNE.sqlset_cursor;
BEGIN
  DBMS_SQLTUNE.create_sqlset(
    sqlset_name  => 'test_sql_tuning_set_2',
    description  => 'Another test SQL tuning set.');

  OPEN l_cursor FOR
    SELECT VALUE(p)
    FROM   TABLE (DBMS_SQLTUNE.select_sqlset (
                    'test_sql_tuning_set', -- sqlset_name
                    NULL,                  -- basic_filter
                    NULL,                  -- object_filter
                    NULL,                  -- ranking_measure1
                    NULL,                  -- ranking_measure2
                    NULL,                  -- ranking_measure3
                    NULL,                  -- result_percentage
                    NULL)                  -- result_limit
                  ) p;

  DBMS_SQLTUNE.load_sqlset (
    sqlset_name     => 'test_sql_tuning_set_2',
    populate_cursor => l_cursor);
END;
/
The contents of an SQL tuning set can be displayed using the SELECT_SQLSET function:
SELECT *
FROM   TABLE(DBMS_SQLTUNE.select_sqlset ('test_sql_tuning_set'));
References can be added to a set to indicate its usage by a client using the ADD_SQLSET_REFERENCE function. The resulting reference ID can be used to remove it using the REMOVE_SQLSET_REFERENCE procedure:
DECLARE
  l_ref_id  NUMBER;
BEGIN
  -- Add a reference to a set.
  l_ref_id := DBMS_SQLTUNE.add_sqlset_reference (
    sqlset_name => 'test_sql_tuning_set',
    reference   => 'Used for manual tuning by SQL*Plus.');

  -- Delete the reference.
  DBMS_SQLTUNE.remove_sqlset_reference (
    sqlset_name  => 'test_sql_tuning_set',
    reference_id => l_ref_id);
END;
/
The UPDATE_SQLSET procedure is used to update specific string (MODULE and ACTION) and number (PRIORITY and PARSING_SCHEMA_ID) attributes of specific statements within a set:
BEGIN
  DBMS_SQLTUNE.update_sqlset (
    sqlset_name     => 'test_sql_tuning_set',
    sql_id          => '19v5guvsgcd1v',
    attribute_name  => 'ACTION',
    attribute_value => 'INSERT');
END;
/
The contents of a set can be trimmed down or deleted completely using the DELETE_SQLSET procedure:
BEGIN
  -- Delete statements with less than 50 executions.
  DBMS_SQLTUNE.delete_sqlset (
    sqlset_name  => 'test_sql_tuning_set',
    basic_filter => 'executions < 50');

  -- Delete all statements.
  DBMS_SQLTUNE.delete_sqlset (
    sqlset_name  => 'test_sql_tuning_set');
END;
/
Tuning sets can be dropped using the DROP_SQLSET procedure:
BEGIN
  DBMS_SQLTUNE.drop_sqlset (sqlset_name => 'test_sql_tuning_set');
  DBMS_SQLTUNE.drop_sqlset (sqlset_name => 'test_sql_tuning_set_2');
END;
/

Useful Views

Useful views related to automatic SQL tuning include:
  • DBA_ADVISOR_TASKS
  • DBA_ADVISOR_FINDINGS
  • DBA_ADVISOR_RECOMMENDATIONS
  • DBA_ADVISOR_RATIONALE
  • DBA_SQLTUNE_STATISTICS
  • DBA_SQLTUNE_BINDS
  • DBA_SQLTUNE_PLANS
  • DBA_SQLSET
  • DBA_SQLSET_BINDS
  • DBA_SQLSET_STATEMENTS
  • DBA_SQLSET_REFERENCES
  • DBA_SQL_PROFILES
  • V$SQL
  • V$SQLAREA
  • V$ACTIVE_SESSION_HISTORY
For further information see:
Hope this helps. Regards Tim...


From http://jonathanlewis.wordpress.com/2007/02/11/profiles/
        http://jonathanlewis.wordpress.com/2007/02/12/sql-profiles-2/
By Jonathan Lewis

SQL Profiles (10g)

When the Tuning Advisor suggests that you accept a SQL Profile, what is it offering you. If you want to find out, the following SQL seems to be the appropriate query to run before you accept the profile:


select
	rat.attr1
from
	wri$_adv_tasks		tsk,
	wri$_adv_rationale	rat
where
	tsk.name = {sql tuning task name}
and	rat.task_id   = tsk.id
;

The {sql tuning task name} will be the task name generated and printed at the top of the Enterprisie Manager screen where you can see the recommendations. The output might list a few lines like the following:


OPT_ESTIMATE(@"SEL$1",INDEX_SCAN, "REQUESTS"@"SEL$1",REQ_UK,SCALE_ROWS=5.786002118 )
OPT_ESTIMATE(@"SEL$1", TABLE, "REQUESTS"@"SEL$1",SCALE_ROWS=0.0004355485551)

These are actually hints - they will be accepted if you include them in the usual fashion in your SQL (although that wouldn't be a good idea at present as they are undocumented, unsupported, and liable to change in their implementation).

Like Stored Outlines, the SQL Profile consists of a stored SQL statement, and a set of hints that will be brought into play when that SQL has to be optimised. Unlike Stored Outlines, the hints for SQL Profiles do not attempt to dictate execution mechanisms directly. Instead they supply arithmetical correction factors to the optimizer as it does its arithemetic as, even with a 100% sample size, it is still possible for the optimizer to misinterpret your statistics and produce an unsuitable execution path.

In principle, if the data distributions do not change, then a stored profile will ensure that the optimizer "understands" your data and does the right thing - even when the data volume changes.


SQL Profiles - 2

Following yesterday's note on SQL Profiles, someone asked how I detect that an opt_estimate hint had been used - with specific reference to the "index_scan" option. The reason for this particular choice is that other opt_estimate hints have a highly visible impact in the 10053 trace files, but this one doesn't.

Here are a few extracts from the Single Table Access section of the trace:

a) Following: opt_estimate(@sel$1, table, t1@sel$1, scale_rows=10) (multiply row count by 10)

   Table: T1  Alias: T1
    Card: Original: 4000    >> Single Tab Card adjusted from: 10.00  to: 100.00
  Rounded: 100  Computed: 100.00  Non Adjusted: 10.00

b) Following: opt_estimate(@sel$1, join, (t2@sel$1, t1@sel$1), scale_rows=225) (multiply join cardinality by 225)

>> Join Card adjusted from 0.00  to: 1.00, prelen=2
Adjusted Join Cards: adjRatio=2666.67 cardHjSmj=1.00 cardHjSmjNPF=1.00
cardNlj=1.00 cardNSQ=1.00 cardNSQ_na=0.00

As you can see, the impact of the opt_estimate() shows up very clearly in these two cases, but when we try: opt_estimate(@sel$1, index_scan, t1@sel$1, t1_i1, scale_rows=0.05) (divide index selectivity by 20), we have to check the before and after trace files:
Before:

   Access Path: index (AllEqRange)
    Index: T1_I1
    resc_io: 168.00  resc_cpu: 1278402
    ix_sel: 0.05  ix_sel_with_filters: 0.05
    Cost: 168.19  Resp: 168.19  Degree: 1
  Best:: AccessPath: IndexRange  Index: T1_I1
         Cost: 168.19  Degree: 1  Resp: 168.19  Card: 10.00  Bytes: 0

After:

   Access Path: index (AllEqRange)
    Index: T1_I1
    resc_io: 12.00  resc_cpu: 89967
    ix_sel: 0.0025  ix_sel_with_filters: 0.0025
    Cost: 12.01  Resp: 12.01  Degree: 1
  Best:: AccessPath: IndexRange  Index: T1_I1
         Cost: 12.01  Degree: 1  Resp: 12.01  Card: 10.00  Bytes: 0

In this case, we don't get any indication of "adjusted values" - we actually have to check the before and after to see what has changed - and infer from the drop in cost that we have used a reduced cardinality. This can be particularly awkward to interpret in cases where the targetted index is not a suitable candidate for a single table access path, but becomes available as an access path later in the join evolution.

DBMS_SQLTUNE.ACCEPT_SQL_PROFILE (
task_name => '',
category => 'MY_CATEGORY');

ALTER SESSION SET SQLTUNE_CATEGORY='MY_CATEGORY' ;

Once satisfied with Plan, change category to default.

Comment by JM -- May 29, 2008 @ 6:27 pm UTC May 29,2008

Yuri,

The view dba_sql_profiles lists the profiles. The text of the query is stored as a CLOB column called sql_text in this view.

If you have lots of profiles and it's hard to find the relevant profile by querying this table, you could use explain plan to generate an execution plan for the query. It should report (in the notes section) something like:

SQL profile "SYS_SQLPROF_01456abbc16a4000″ used for this statement

I think the value reported is stored as the name in the view.

Comment by Jonathan Lewis -- January 28, 2008 @ 10:25 pm UTC Jan 28,2008

Thank for your answer!
But I'm interest in some different thing.
I'm curious in which form does the SQL Profile been stored?
And upon which SQL Profile information bases the cardinality adjustment done
by CBO,
such as "Single Tab Card adjusted from:11.441291 to:28.000000″.
I know some about SQL Profile, for example

SQL> select p.other_xml
2 from dba_sqltune_plans p, dba_sql_profiles s
3 where p.task_id=s.task_id
4 and s.name='profile_lastdoc_1′
5 and p.attribute='Using SQL profile'
6 and p.id=1;

But in this column I can`t find anything about cardinality adjustment factor.

Comment by Yuri A.P. -- January 29, 2008 @ 8:39 pm UTC Jan 29,2008

The content of the SQL profile is not stored in a "public" view, but it can be found in table sql$profattr - for example:

select
       sp.sp_name, sa.attr#, sa.attr_val
from
       sqlprof$      sp,
       sqlprof$attr  sa
where
       sp.signature = sa.signature
and     sp.category  = sp.category
order by
       sp.sp_name,
       sa.attr#
;

SP_NAME                             ATTR#
------------------------------ ----------
ATTR_VAL
-------------------------------------------------------------------------
SYS_SQLPROF_0145f1e5167c4000            1
OPT_ESTIMATE(@"SEL$1", JOIN, ("T2"@"SEL$1", "T1"@"SEL$1"), SCALE_ROWS=15)

SYS_SQLPROF_0145f1e5167c4000            2
OPT_ESTIMATE(@"SEL$1", TABLE, "T2"@"SEL$1", SCALE_ROWS=200)

SYS_SQLPROF_0145f1e5167c4000            3
OPTIMIZER_FEATURES_ENABLE(default)

In this example, we have two adjustments - the single table cardinality of table t2 in query block sel$1 should be multiplied by 200; and the join cardinality of (t1,t2) when they are joined in query block sel$1 should be multiplied by 15.

Comment by Jonathan Lewis -- January 29, 2008 @ 10:22 pm UTC Jan 29,2008

Thank you very much!
By the way, I have discovered that on 11g database such information can be retrieved from non-documented view DBMSHSXP_SQL_PROFILE_ATTR.

Comment by Yuri A.P. -- January 30, 2008 @ 8:34 pm UTC Jan 30,2008


Subject: Automatic SQL Tuning - SQL Profiles
  Doc ID: Note:271196.1 Type: BULLETIN
  Last Revision Date: 29-JUL-2008 Status: PUBLISHED 

Checked for relevance on 12-May-2008.

Clarification/Explanation
==========================

The query optimizer can sometimes produce inaccurate estimates about 
an attribute of a statement due to lack of information,leading to poor
execution plans.  Traditionally, users have corrected  this problem by
manually adding hints to the application code to guide the optimizer 
into making correct decisions. For packaged applications, changing 
application code is not an option and the only alternative available 
is to log a bug with the application vendor and wait for a fix.

Automatic SQL Tuning deals with this problem with its SQL Profiling 
capability. The Automatic Tuning Optimizer creates a profile of the 
SQL statement called a SQL Profile, consisting of auxiliary statistics 
specific to that statement. The query optimizer under normal mode makes 
estimates about cardinality, selectivity, and cost that can sometimes be 
off by a significant amount resulting in poor execution plans. SQL Profile 
addresses this problem by collecting additional information using
sampling and partial execution techniques to verify and, if necessary, 
adjust these estimates.

During SQL Profiling, the Automatic Tuning Optimizer also uses execution
history information of the SQL statement to appropriately set optimizer 
parameter settings, such as changing the OPTIMIZER_MODE initialization 
parameter setting from ALL_ROWS to FIRST_ROWS for that SQL statement.

The output of this type of analysis is a recommendation to accept the 
SQL Profile. A SQL Profile, once accepted, is stored persistently in 
the data dictionary. Note that the SQL Profile is specific to a 
particular query. If accepted, the optimizer under normal mode uses the 
information in the SQL Profile in conjunction with regular
database statistics when generating an execution plan. 
The availability of the additional information makes it possible
to produce well-tuned plans for corresponding SQL statement without
requiring any change to the application code.

The scope of a SQL Profile can be controlled by the CATEGORY profile attribute. 
This attribute determines which user sessions can apply the profile. 
You can view the CATEGORY attribute for a SQL Profile in CATEGORY column 
of the DBA_SQL_PROFILES view. By default, all profiles are created in the 
DEFAULT category. This means that all user sessions where the SQLTUNE_CATEGORY 
initialization parameter is set to DEFAULT can use the profile.

By altering the category of a SQL profile, you can determine which sessions are 
affected by the creation of a profile. For example, by setting the category 
of a SQL Profile to DEV, only those users sessions where the SQLTUNE_CATEGORY 
initialization parameter is set to DEV can use the profile. All other sessions 
do not have access to the SQL Profile and execution plans for SQL statements 
are not impacted by the SQL profile. This technique enables you to test 
a SQL Profile in a restricted environment before making it available to other 
user sessions.

It is important to note that the SQL Profile does not freeze the execution plan
of a SQL statement, as done by stored outlines. As tables grow or indexes 
are created or dropped, the execution plan can change with the same SQL Profile. 
The information stored in it continues to be relevant even as the data 
distribution or access path of the corresponding statement change. 
However, over a long period of time, its content can become outdated and 
would have to be regenerated. This can be done by running Automatic 
SQL Tuning again on the same statement to regenerate the SQL Profile.

SQL Profiles apply to the following statement types:

     SELECT statements 
     UPDATE statements 
     INSERT statements (only with a SELECT clause) 
     DELETE statements 
     CREATE TABLE statements (only with the AS SELECT clause) 
     MERGE statements (the update or insert operations)

Managing SQL Profiles
=====================

While SQL Profiles are usually handled by Oracle Enterprise Manager as part 
of the Automatic SQL Tuning process, SQL Profiles can be managed through the
DBMS_SQLTUNE package. To use the SQL Profiles APIs, you need the 
CREATE ANY SQL_PROFILE, DROP ANY SQL_PROFILE, and ALTER ANY SQL_PROFILE
system privileges.

Accepting a SQL Profile
=======================

You can use the DBMS_SQLTUNE.ACCEPT_SQL_PROFILE procedure to accept a 
SQL Profile recommended by the SQL Tuning Advisor. This creates and stores a
SQL Profile in the database. For example:

DECLARE
 my_sqlprofile_name VARCHAR2(30);
BEGIN
 my_sqlprofile_name := DBMS_SQLTUNE.ACCEPT_SQL_PROFILE ( 
    task_name => 'my_sql_tuning_task',
    name      => 'my_sql_profile');
END;


Where my_sql_tuning_task is the name of the SQL tuning task.
You can view information about a SQL Profile in the DBA_SQL_PROFILES view.

Altering a SQL Profile
======================

You can alter the STATUS, NAME, DESCRIPTION, and CATEGORY attributes of 
an existing  SQL Profile with the ALTER_SQL_PROFILE procedure. For example:

BEGIN
  DBMS_SQLTUNE.ALTER_SQL_PROFILE(
     name            => 'my_sql_profile', 
     attribute_name  => 'STATUS', 
     value           => 'DISABLED');
END;
/


In this example, my_sql_profile is the name of the SQL Profile that you want to alter. 
The status attribute is changed to disabled which means the SQL Profile is
not used during SQL compilation.

Dropping a SQL Profile
======================
You can drop a SQL Profile with the DROP_SQL_PROFILE procedure. For example:

BEGIN
  DBMS_SQLTUNE.DROP_SQL_PROFILE(name => 'my_sql_profile');
END;
/


In this example, my_sql_profile is the name of the SQL Profile you want to drop. 
You can also specify whether to ignore errors raised if the name does not exist.
For this example, the default value of FALSE is accepted.

Example
=======

SESSION 1 -- SCOTT
=========

SQL> create table test (n number );

Table created.

declare
begin
for i in 1 .. 10000
loop
insert into test values(i);
commit;
end loop;
end;

PL/SQL procedure successfully completed.


create index test_idx on test(n);

Index created.


analyze table test estimate statistics (OR use dbms_stats)

Table analyzed.


select /*+ no_index(test test_idx) */ * from test where n=1

Execution Plan
----------------------------------------------------------
   0      SELECT STATEMENT Optimizer=ALL_ROWS (Cost=6 Card=1 Bytes=13)
   1    0   TABLE ACCESS (FULL) OF 'TEST' (TABLE) (Cost=6 Card=1 Bytes
          =13)



SESSION 2  -- SYS
=========

  1  DECLARE
  2   my_task_name VARCHAR2(30);
  3   my_sqltext   CLOB;
  4  BEGIN
  5   my_sqltext := 'select /*+ no_index(test test_idx) */ * from test where n=
1';
  6   my_task_name := DBMS_SQLTUNE.CREATE_TUNING_TASK(
  7           sql_text    => my_sqltext,
  8           user_name   => 'SCOTT',
  9           scope       => 'COMPREHENSIVE',
 10           time_limit  => 60,
 11           task_name   => 'my_sql_tuning_task_2',
 12           description => 'Task to tune a query on a specified table');
 13* END;
 14  /

PL/SQL procedure successfully completed.


  1  BEGIN
  2    DBMS_SQLTUNE.EXECUTE_TUNING_TASK( task_name => 'my_sql_tuning_task_2');
  3* end;
SQL> /

PL/SQL procedure successfully completed.


  1  SET LONG 1000
  2  SET LONGCHUNKSIZE 1000
  3  SET LINESIZE 100
  4* SELECT DBMS_SQLTUNE.REPORT_TUNING_TASK( 'my_sql_tuning_task_2') from DUAL;



DBMS_SQLTUNE.REPORT_TUNING_TASK('MY_SQL_TUNING_TASK_2')
--------------------------------------------------------------------------------
--------------------
GENERAL INFORMATION SECTION
-------------------------------------------------------------------------------
Tuning Task Name   : my_sql_tuning_task_2
Scope              : COMPREHENSIVE
Time Limit(seconds): 60
Completion Status  : COMPLETED
Started at         : 05/04/2004 17:36:05
Completed at       : 05/04/2004 17:36:05

-------------------------------------------------------------------------------
SQL ID  : d4wgpc5g0s0vu

DBMS_SQLTUNE.REPORT_TUNING_TASK('MY_SQL_TUNING_TASK_2')
--------------------------------------------------------------------------------
--------------------
SQL Text: select /*+ no_index(test test_idx) */ * from test where n=1

-------------------------------------------------------------------------------
FINDINGS SECTION (1 finding)
-------------------------------------------------------------------------------

1- SQL Profile Finding (see explain plans section below)
--------------------------------------------------------
  A potentially better execution plan was found for this statement.

  Recommendation (estimated benefit: 83.84%)

DBMS_SQLTUNE.REPORT_TUNING_TASK('MY_SQL_TUNING_TASK_2')
--------------------------------------------------------------------------------
--------------------
  ------------------------------------------
    Consider accepting the recommended


  1  DECLARE
  2   my_sqlprofile_name VARCHAR2(30);
  3  BEGIN
  4   my_sqlprofile_name := DBMS_SQLTUNE.ACCEPT_SQL_PROFILE (
  5      task_name => 'my_sql_tuning_task_2',
  6      name      => 'my_sql_profile');
  7* END;
  8  /

PL/SQL procedure successfully completed.


SQL> select to_char(sql_text) from           dba_sql_profiles;

TO_CHAR(SQL_TEXT)
------------------------------------------------------------------------
select /*+ no_index(test test_idx) */ * from test where n=1


SESSION 1 --- SCOTT


SQL> select /*+ no_index(test test_idx) */ * from test where n=1;

Execution Plan
----------------------------------------------------------
   0      SELECT STATEMENT Optimizer=ALL_ROWS (Cost=1 Card=1 Bytes=13)
   1    0   INDEX (RANGE SCAN) OF 'TEST_IDX' (INDEX) (Cost=1 Card=1 By
          tes=13


From http://asktom.oracle.com/pls/asktom/f?p=100:11:0::::P11_QUESTION_ID:61313086268493

Rasin -- Thanks for the question regarding "SQL Profiles", version 10.2

You Asked

Tom,
can you explain what are SQL Profiles, what are they used for how the CBO uses them and
how can we utilize this feature.

Thanks.
 
and we said...

Ok, this is my conceptual explanation. 

In the past, the DBA could analyze single 'things'

the DBA could analyze A TABLE - finding the number of blocks, rows, things like that. 
about A TABLE.

the DBA could analyze A COLUMN - get histograms, high/low values, number of distinct
values, things like that - about A COLUMN

the DBA could analyze AN INDEX - get keys/get, leaf blocks, height, clustering factor.

the DBA could analyze A SYSTEM - find the single block IO time, the multi-block IO time,
cpu speed and so on.


But - most of our queries (many of them) involve more than A TABLE, A COLUMN, AN INDEX,
and so on... They are complex.

For example (blatant theft of example from Jonathan Lewis here - his CBO book is awesome,
I remember him first using this example in 2003 at the NoCoug meeting...)

o How many of you are Pisces?  1/12th of the room will raise hands.

o How many of you are born in December?  1/12 of the room will raise hands.


Now, how many of you Pisces were born in December?  (silence, none of them are - I know,
I am Pisces :)

Now, ask the optimizer!


ops$tkyte@ORA10GR2> create table t
  2  as
  3  select decode( mod(rownum,12), 0, 'Capricorn',
  4                                 1, 'Gemini',
  5                                                             2, 'Libra',
  6                                                             3, 'Aquarius',
  7                                                             4, 'Cancer',
  8                                                             5, 'Scorpio',
  9                                                             6, 'Pisces',
 10                                                             7, 'Ariea',
 11                                                             8, 'Leo',
 12                                                             9, 'Sagittarius',
 13                                                             10, 'Taurus',
 14                                                             11, 'Virgo' ) zodiac,
 15         add_months( to_date( '01-jan-2006' ), mod(rownum,12) ) dob
 16    from all_objects
 17  /

Table created.

ops$tkyte@ORA10GR2>
ops$tkyte@ORA10GR2> exec dbms_stats.gather_table_stats( user, 'T', method_opt=> 'for all
columns size 254' );

PL/SQL procedure successfully completed.

ops$tkyte@ORA10GR2>
ops$tkyte@ORA10GR2> set autotrace traceonly explain
ops$tkyte@ORA10GR2> select * from t;

Execution Plan
----------------------------------------------------------
Plan hash value: 1601196873

--------------------------------------------------------------------------
| Id  | Operation         | Name | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |      | 51292 |   751K|    52   (6)| 00:00:01 |
|   1 |  TABLE ACCESS FULL| T    | 51292 |   751K|    52   (6)| 00:00:01 |
--------------------------------------------------------------------------

ops$tkyte@ORA10GR2> select * from t where zodiac = 'Pisces';

Execution Plan
----------------------------------------------------------
Plan hash value: 1601196873

--------------------------------------------------------------------------
| Id  | Operation         | Name | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |      |  4107 | 61605 |    53   (8)| 00:00:01 |
|*  1 |  TABLE ACCESS FULL| T    |  4107 | 61605 |    53   (8)| 00:00:01 |
--------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   1 - filter("ZODIAC"='Pisces')

ops$tkyte@ORA10GR2> select * from t where dob = to_date( '01-dec-2006' );

Execution Plan
----------------------------------------------------------
Plan hash value: 1601196873

--------------------------------------------------------------------------
| Id  | Operation         | Name | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |      |  4494 | 67410 |    53   (8)| 00:00:01 |
|*  1 |  TABLE ACCESS FULL| T    |  4494 | 67410 |    53   (8)| 00:00:01 |
--------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   1 - filter("DOB"=TO_DATE('2006-12-01 00:00:00', 'yyyy-mm-dd
              hh24:mi:ss'))

ops$tkyte@ORA10GR2> select * from t where zodiac = 'Pisces' and dob = to_date(
'01-dec-2006' );

Execution Plan
----------------------------------------------------------
Plan hash value: 1601196873

--------------------------------------------------------------------------
| Id  | Operation         | Name | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |      |   360 |  5400 |    53   (8)| 00:00:01 |
|*  1 |  TABLE ACCESS FULL| T    |   360 |  5400 |    53   (8)| 00:00:01 |
--------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   1 - filter("ZODIAC"='Pisces' AND "DOB"=TO_DATE('2006-12-01
              00:00:00', 'yyyy-mm-dd hh24:mi:ss'))


We can sort of see this with dynamic sampling - which is *similar* to a sql profile,
but not stored like a profile would be:


ops$tkyte@ORA10GR2> select /*+ dynamic_sampling( t 4 ) */ * from t where zodiac =
'Pisces' and dob = to_date( '01-dec-2006' );

Execution Plan
----------------------------------------------------------
Plan hash value: 1601196873

--------------------------------------------------------------------------
| Id  | Operation         | Name | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |      |     1 |    15 |    53   (8)| 00:00:01 |
|*  1 |  TABLE ACCESS FULL| T    |     1 |    15 |    53   (8)| 00:00:01 |
--------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   1 - filter("ZODIAC"='Pisces' AND "DOB"=TO_DATE('2006-12-01
              00:00:00', 'yyyy-mm-dd hh24:mi:ss'))

Note
-----
   - dynamic sampling used for this statement

ops$tkyte@ORA10GR2> set autotrace off

the optimizer ran this query for us to parse:

SELECT /* OPT_DYN_SAMP */ /*+ ALL_ROWS opt_param('parallel_execution_enabled',
   'false') NO_PARALLEL(SAMPLESUB) NO_PARALLEL_INDEX(SAMPLESUB) NO_SQL_TUNE
  */ NVL(SUM(C1),:"SYS_B_0"), NVL(SUM(C2),:"SYS_B_1")
FROM
 (SELECT /*+ NO_PARALLEL("T") FULL("T") NO_PARALLEL_INDEX("T") */ :"SYS_B_2"
  AS C1, :"SYS_B_3" AS C2  FROM "T" "T" WHERE "T"."ZODIAC"=:"SYS_B_4" AND
  "T"."DOB"=TO_DATE(:"SYS_B_5", :"SYS_B_6")) SAMPLESUB


and used the results of that as the estimated cardinality.  SQL profiles would do the
same - but would look at the entire query - cross table, column, whatever and do partial
executes on it (try out the data in your database) and remember that in the dictionary as
extended statistics.

It can use that next time to parse and optimize the query.

It'll remember other things too - like "this query is used to retrieve just the first
rows, we should not optimize for all rows, but first rows - let us remember that in the
future..."

So, a SQL profile is sort of like gathering statistics on A QUERY - which involves many
tables, columns and the like....

In fact - it is just like gathering statistics for a query, it stores additional
information in the dictionary which the optimizer uses at optimization time to determine
the correct plan.  The SQL Profile is not "locking a plan in place", but rather giving
the optimizer yet more bits of information it can use to get the right plan.

SQL PROFILES  
Reviewer: Mark

Tom,

Do SQL Profiles get stale?  If so, is there some way to refresh stale profiles like we do for
stats?

Thanks for your time..

Followup   April 2, 2008 - 1pm US/Eastern:
you re-profile the sql statement.

Yes, they can get "stale" - basically, sql profiles are better estimated cardinality values - better in that they are based on sampling your where clause - not based on basic 'statistical projection'.

and if the underlying data changes - well, then they (the better card=values) would become wrong.

Eg: if someone changes the zodiac lookup table so that pisces are born in december, the sample we ran that says "zero are" would be wrong

sql profile vs Sql plan management of 11g 
Reviewer: Aman Sharma from India

Hi  sir,
Its an excellent explanation.Can you please explain the difference in the sql profiles and sql plan
management /sql baselines of 11g ?
regards
Aman....

Followup   April 3, 2008 - 8pm US/Eastern:
the sql baselines is more akin to storing query plans and ensuring those plans are used. they are the baseline plans..

sql profiles are what they are - cardinality information, subject to influencing a plan, but not dictating it.


From http://asktom.oracle.com/pls/asktom/f?p=100:11:0::::P11_QUESTION_ID:729346500346644123

Joe -- Thanks for the question regarding "10g SQL Profiles", version 10.2

You Asked
Can you explain how SQL profiles work when it comes to recommendations from the SQL Tuning Advisor in 10g?

Thanks in advance.

and we said...

simply put:

sql profiles work by taking the query itself and doing partial executions of various predicates - and remembering the ACTUAL card= values (cardinality values) and hinting the query so that it *KNOWS* "this step will return about 5 records, not the 50,000 you guessed at, but only 5"

and the next time the query is hard parsed, we use those card= values to come up with the best plan.

the key is to get the CORRECT card=value numbers, when we get the right number - you get the right plan. When we get the wrong number - that is when you see suboptimal plans.

So, sql profiles simply take the query - execute bits of it, figure out how many rows are actually returned (or look at the execution HISTORY to see how many rows are actually flowing from the row sources as well) and STORE that information as extended statistics - and use that information the next time the query is hard parsed

Profiles
Reviewer: Dana from Phoenix, AZ USA

Are SQL Profiles an extension of sql outlines?
Many of the features sound similar or same.

Followup   March 26, 2008 - 8am US/Eastern:

similar, but not the same.

stored outlines are a set of hints that say "use this index, do this table first, do that next, use this access path, perform this filter then that filter"....

sql profiles are more like extended statistics - they are the result of "analyzing a query", the information provided to the optimizer is not HOW to perform the query - but rather better information about how many rows will flow out of a step in the plan, how selective something is.

say you have a query, you generate a stored outline for it. You now add an index to the underlying table. This index would be GREAT for the query. A stored outline won't use it, the stored outline says "use this index - query that table - then do this". Since the GREAT index did not exist when the outline was generated - it won't be used.

say you have a query, you generate a profile for it. You now add an index to the underlying table. This index would be GREAT for the query. A profile can use it - since the profile is just more information for the CBO - better cardinality estimates.


SQLProfiles_20060622.pdf


]]> Hashing in Oracle tag:www.heysky.net,2008:/digest//1.328 2008-07-29T23:26:28Z 2008-07-29T23:30:43Z Sky http://www.heysky.net From http://www.dbasupport.com/oracle/ora10g/hash.shtml

 

Hashing in Oracle
Steve Callan, stevencallan@hotmail.com

The internal workings of the Oracle RDBMS are a marvel to behold. Think about all of the things you do know with respect to what it takes to issue a select statement, perform an update, or create a table. Latches, locks, serialization, caching, pinning, flushing, checkpointing, sorting, space management, undo, redo, and the list goes on and on. Aside from visible functions, there are a slew of transparent operations. Many of these transparent operations, as the adjective implies, not only work behind the scenes, but also stay there. There are also operations whose work is largely hidden, but whose results are exposed to us. One such operation found in several key areas of how Oracle works is hashing.

]]> What is Hashing?
Hash isn't a formal term, but more of a description for what takes place: chop and mix. A hash function returns a value for an input, and the output is generally shorter or more compact than the input. For a relatively simple hash function, there is no guarantee that inputs map one to one for outputs, that is, two different inputs can have the same hashed output value. Add more sophistication to the hash function, or add yet another hashing, and the probability of two distinct inputs having the same hashed output can be greatly decreased.

Before getting more into where hashing takes place in Oracle, let's look at some examples of how hashing is used. A good general overview of hashing and various types of hashing can be found at a Mathematica site. The table of names takes the sum of the ASCII values of each letter, sums them, and then lists the mod 13 remainder. The table is reproduced below with an additional column to help illustrate what is taking place. I also provided some code to help reproduce the results.

create table hash_example
(lastname varchar2(12),
  ascii_sum number,
  hashed_value number);
insert into hash_example (lastname) values ('Brecker');
insert into hash_example (lastname) values ('Corea');
insert into hash_example (lastname) values ('Davis');
insert into hash_example (lastname) values ('Hancock');
insert into hash_example (lastname) values ('Harris');
insert into hash_example (lastname) values ('Marsalis');
insert into hash_example (lastname) values ('Parker');
commit;

declare
  cursor c is
  select lower(lastname) lastname
  from hash_example
  for update;
  l_name hash_example.lastname%type;
  l_sum hash_example.ascii_sum%type := 0;
  l_tmp number := 0;
begin
  for r in c loop
    for i in 1..length(r.lastname) loop
      l_tmp := ascii(substr(r.lastname,i,1));
      l_sum := l_sum + l_tmp;
    end loop;
    update hash_example
    set ascii_sum = l_sum,
        hashed_value = mod(l_sum,13)
    where current of c;
    l_tmp := 0;
    l_sum := 0;
  end loop;
end;
/

PL/SQL procedure successfully completed.

SQL> select * from hash_example;

LASTNAME      ASCII_SUM HASHED_VALUE
------------ ---------- ------------
Brecker             734            6
Corea               522            2
Davis               535            2
Hancock             727           12
Harris              649           12
Marsalis            860            2
Parker              645            8


Overall, this was a pretty simple example. Given that you can pass in a value and get an output based on the hashing algorithm, and that the output is like a fingerprint of the original data, how do you think this can applied to a real table? Suppose you have a table that contains employee identification numbers along with salaries. Aside from using an already built in hashing function (e.g., SHA-1), you could devise a function to hash the combined employee ID and salary values. This hashed value could be stored elsewhere, and then auditing of the table would compare the current values and the stored values. Any (unknown) change in the fingerprint would be cause for concern. This is just one safeguard, and no doubt, you'd want to have some auditing (before the fact, not after!) on a sensitive table like this.

Hashing in a Database
So far, we've seen two uses of hashing that can be applied to a database. In the first example, a hashed value of some key attribute can be used for indexing (the hashed value points back to several values), and in the second example, hashing can be used as a comparison tool. These two concepts lend themselves to extensive use within Oracle.

How does Oracle know if a SQL statement is new or already parsed? A parsed statement has a hash value that is by and large unique enough (collisions, or same output for different inputs can happen, but are rare). Shown below is an extract from a trace file, and the lines show each statement's identifying information.

1 PARSING IN CURSOR #1 len=46 dep=0 uid=5 hv=810835083 ad='1d9cd494'

  select ename from scott.emp where empno = 7369
 
2 PARSING IN CURSOR #2 len=46 dep=0 uid=5 hv=1290101743 ad='1d9cca00'

  select ename from scott.emp where empno = 7499
 
3 PARSING IN CURSOR #1 len=46 dep=0 uid=5 hv=1589217292 ad='1d9cc7c4'

  SELECT ENAME FROM SCOTT.EMP WHERE 7499 = EMPNO
 
4 PARSING IN CURSOR #2 len=46 dep=0 uid=5 hv=810835083 ad='1d9cd494'

  select ename from scott.emp where empno = 7369
 
Statements 1, 2 and 3 are different, so each has its own hashed value (the hv value). Statements 1 and 4 are the same, and note both have the same hv value. One of the more desirable characteristics of a hashing function is that it should be quite fast and inexpensive to perform. If your statement has already been parsed (has a hash ID value), then a comparison is all that is needed. As pointed out in numerous other sources, hard parsing is undesirable. Re-use of statements can be accomplished two ways. One is via the use of bind variables, and the other is to embed commonly used statements in functions or procedures.

Where else is hashing used? Hash partitioning is useful for data that doesn't cleanly fall into categories such as range or list. Composite partitioning such as range-hash can be used to further categorized and distribute data. In these examples, the indexing feature or benefit is being used. Since this is such a highly performant arrangement of data, the use of hash clusters would be a natural extension. "Overview of Hash Clusters" in the Concepts guide provides a good description of how useful clusters can be. Clustered data is easier to find, and for sorted data, the benefits are even better.

Maybe not as obvious for usage as in the past, given how much better the cost based optimizer is, but hash joins are a frequently used means of accessing data. Incorporating a hint in a statement can result in seeing the hashing step in an execution plan. Without looking at execution plans, and by letting the optimizer do its thing, Oracle isn't going to "tell" you that a hash join was used.

Hidden Hashing
Now it's time for a behind the scene use of hashing. In fact, this particular use of hashing is so under the radar that you would probably never even know it existed in Oracle unless you encountered one of several bugs. From note 4960265.9 we have, "In some cases when using a join filter in a RAC configuration (parallel hash-join) wrong results can be produced." The stated fix for this is to set the undocumented parameter _BLOOM_FILTER_ENABLED to false. What is a Bloom filter?

There are lots of hashing techniques, and "General Purpose Hash Function Algorithms," by Arash Partow, gives a decent summary without delving too deep into some esoteric computer science. The definition of a Bloom filter given in the paper states the following:

A Bloom filter allows for the state of existence of a very large set of possible type values to be repsented with a much smaller piece of memory. This is achieved through the use of multiple distinct hash functions and also by allowing the result of a query for the existence of a particular type to have a certain amount of error. This error can be controlled by varying the size of the table used for the Bloom filter and also by varying the number of hash functions.

So now, we know another technique Oracle uses to store data. In a sense, the data can be "over stored," that is, we want to make really, really sure it exists in one place or another. We may be wrong about the place (and then go find an accurate location), but the main point is that it exists. With this type of hashing, it will be difficult to say we don't have it anywhere, when in fact we do.

In Closing
Hashing is but one of many cleverly implemented functions embedded in Oracle. What would be interesting is being able to document all of the hash function implementation used in Oracle. For example, how does Oracle come up with 810835083 from "select ename from scott.emp where empno = 7369?" And why do hashed partitions work better (better = data spread more evenly) when the number of buckets is a power of two?

]]> Oracle 数据库 10g 版本 1 中的 PL/SQL 纯编译 (NCOMP) tag:www.heysky.net,2008:/digest//1.327 2008-07-29T20:01:09Z 2008-07-29T20:07:20Z Sky http://www.heysky.net From http://www.oracle.com/technology/global/cn/tech/pl_sql/htdocs/ncomp_faq.html

 

Oracle 数据库 10g 版本 1 中的 PL/SQL 纯编译 (NCOMP)

2004 年 9 月 4 日更新 

--------------------------------------------------------------------------------

什么是纯编译？
如何实现 PL/SQL 纯编译？
Oracle 数据库 10g 有什么新特性？
什么对象可以进行 PL/SQL 纯编译？
共享库 (DLL) 的命名惯例是什么？
共享的动态链接库是否可以移植？
在部署端是否需要 C 编译环境？
哪些 Oracle 参数与纯编译相关？
plsql_native_library_dir 参数
plsql_native_library_subdir_count 参数
plsql_code_type 参数
$ORACLE_HOME/plsql/spnc_commands 文件的格式
纯编译生成的共享库存储在哪个字典表中？
当删除一个"纯编译"单元时，文件系统上的共享库将发生什么？
如何更改 plsql_native_library_dir？
是否还有其他场合需要手动删除共享库？
如果误删除了一个共享库该怎么办？
甲骨文公司是否建议混合使用纯编译单元和解释单元？
在操作系统升级时，是否需要重新生成纯编译的共享库？
在应用程序部署期间是否能够使用纯编译模式来节省时间（通过提供预先生成的 NCOMP 共享库）？
纯编译测试：简单测试
将数据库中的所有 PL/SQL 单元进行纯编译

]]> 您可以通过将 PL/SQL 模块（程序包、触发器、过程、函数和类型）编译成驻留在共享库中的纯代码来加速它们的运行。过程被转换成 C 代码，然后用一个 C 编译器进行编译，并与 Oracle 进程动态链接。

您可以将该技术用于随付的 Oracle 程序包以及您自己编写的过程。用这种方式编译的过程可以用于所有服务器环境，如共享服务器配置（以前称为多线程服务器）和 Oracle 真正应用集群。

使用了大量有代表性测试程序的性能实验（发布在 OTN 上）表明，在 Oracle 数据库 10g 中，纯 PL/SQL 程序的应用程序在纯编译时的速度与解释编译时的速度相比（其他保持不变），预计可以获得大约 1.05 到 2.4 倍的性能提高。提高的程度取决于程序的种类。

注意：许多 Oracle 提供的程序包（如 UTL_FILE、UTL_RAW、DBMS_LOB）已经用 C 实施，而仅使用 PL/SQL 来提供 API。这些程序包不大可能从纯编译中获得显著的性能提高。同样，仅调用 SQL 语句的 PL/SQL 程序单元也可能几乎得不到加速。不过，纯编译的 PL/SQL 往往至少和相应的解释代码一样快。编译的代码将与解释的代码执行相同的库调用，因此它们的行为是完全相同的。

 

--------------------------------------------------------------------------------

如何实现 PL/SQL 纯编译？

如果未使用纯编译，则将每个 PL/SQL 程序单元编译成一种中间形式的机器可读码 (MCode)。MCode 存储在数据库字典中，并在运行时进行解释。

使用 PL/SQL 纯编译，将 PL/SQL 程序编译成无需运行时解释的纯机器码，从而加快运行速度。

PL/SQL 使用命令文件 $ORACLE_HOME/plsql/spnc_commands 和支持的操作系统 C 编译器和链接器，将生成的 C 代码编译并链接到共享库中。共享库存储在数据字典中，这样便可将它们自动备份并防止它们被删除。这些共享库文件被复制到文件系统中，并在调用 PL/SQL 子程序时被加载和运行。如果在关闭数据库时从文件系统中删除了这些文件，或者如果您更改了这些库所在的目录，则自动重新提取这些文件。

 

--------------------------------------------------------------------------------

Oracle 数据库 10g 有什么新特性？

这个特性现在需要更少的设置和维护。

不需要用相同的代码类型设置来编译程序包主体及其规范。例如，可以对程序包主体进行纯编译，而对程序包规范进行解释编译。
经过纯编译的子程序存储在数据库中，并在需要的时候自动提取相应的共享库。对于备份共享库、清理原有共享库等事项，您无需操心。
而且简化了纯编译的初始化参数和命令的设置。唯一需要的参数是 plsql_native_library_dir。不再使用与编译器、链接器和 make 实用程序相关的参数。$ORACLE_HOME/plsql/spnc_commands 命令文件用于进行纯编译。它包含了用于编译和链接的命令和选项，从而不再依赖于 make 实用程序。不再需要或支持 plsql_native_make_utility 和 plsql_native_make_file_name 参数，但仍须像在 9.0.1 版和 9.2.0 版中一样指定 plsql_native_library_dir。这将指定用于生成共享库（在共享库被复制到数据库中之前）的目录的完整路径。
纯编译期间出现的所有错误都存储在 USER_ERRORS 字典视图中，可通过 SQL*Plus 命令 SHOW ERRORS 来查询。
开启和关闭纯编译是由一个单独的参数 plsql_code_type 来控制的，而不是 plsql_compiler_flags 参数的几个选项之一，现在不建议使用后者。
RAC 支持：现在在没有集群（共享）文件系统的 RAC 环境中也支持 PL/SQL 纯编译。
共享库清理：当删除（或在解释模式下重新编译）纯编译的 PL/SQL 模块时，现在将删除（在 plsql_native_library_dir 指定的位置下的）共享库的文件系统副本。在 RAC 配置中，将从所有实例中删除共享库。


--------------------------------------------------------------------------------

什么对象可以进行 PL/SQL 纯编译？

任何存储的 PL/SQL 模块都可以进行纯编译。匿名的 PL/SQL 程序块不能进行纯编译。

注意：Oracle 支持以下类型的存储 PL/SQL 模块：

FUNCTION
PACKAGE
PACKAGE BODY
PROCEDURE
TRIGGER
TYPE
TYPE BODY

 

--------------------------------------------------------------------------------

共享库 (DLL) 的命名惯例是什么？

纯编译为每个单元创建一个共享库。例如，将为程序包规范和程序包主体分别创建一个共享库。以下是生成的一些名称：

PKG__SCOTT__S__54682.so
PKG__SCOTT__B__54683.so

生成的名称一般包含单元名称、模式名称、对象类型和对象编号。不过，这是一个可能会变更的内部惯例，用户不应依赖这种文件名格式规范。

文件名的扩展名在不同平台上可能不同。例如，在 Solaris 上共享库的后缀为 .so，而在 HP-UX 上共享库的后缀为 .sl。

返回页首
共享的动态可链接库是否可以移植？

不，绝对不行！共享库包含特定于平台的目标代码。

注意：在解释方式下创建的 MCode 也是特定于平台的。

 

--------------------------------------------------------------------------------

在部署端是否需要 C 编译环境？

是的。没有商量的余地。PL/SQL 单元可能需要在部署端进行编译，PL/SQL 纯编译依靠一个 C 编译器来生成目标（机器）代码。

例如，各种操作（如向表中增加一列）可能导致与该表相关的 PL/SQL 单元的自动失效。对这种单元的后续访问将触发重新编译。在这种情况下，在纯编译模式下再次重新编译该单元需要 C 编译环境。

另一种情况是需要在部署数据库上应用补丁。可能需要在部署环境中重新加载和编译个别程序包。

 

--------------------------------------------------------------------------------

哪些 Oracle 参数与纯编译相关？

plsql_native_library_dir
plsql_native_library_subdir_count
plsql_code_type
返回页首
plsql_native_library_dir 参数

这个参数指定保存共享库 (DLL) 的 OS 副本的目录位置。

当对一个模块进行纯编译时，将在这个位置创建共享库，然后将其复制到数据库字典表 (ncomp_dll$) 中。虽然共享库的主副本驻留在数据库中，但还将共享对象在文件系统中进行了物化，以便能够动态地将它们加载到 Oracle 的地址空间中。

注释 1：用户 (DBA) 决不能在系统启动并运行时从 plsql_native_library_dir 中手动删除共享库，因为这些 DLL 可能被映射到了 Oracle 进程上。只有在系统关闭的时候才能安全地删除共享库的 OS 副本。

注释 2：在 RAC 配置中，必须在每个实例中设置这个参数。不要求各实例拥有共享文件系统。在每个实例上，必须将 plsql_native_library_dir 设置为指向一个实例本地目录。或者，如果 RAC 配置支持共享（集群）文件系统，您可以将（共享文件系统上的）一个公共的目录用于所有实例。

注释 3：您必须创建这个目录。Oracle 不会自动为您创建这个目录。

注释 4：甲骨文公司在单实例或 RAC 配置中都不为 plsql_native_library_dir 提供 NFS 挂载目录支持。这是因为 NFS 在写或删除文件时会导致一些不可预测的时序问题。

 

--------------------------------------------------------------------------------

plsql_native_library_subdir_count 参数

一些文件系统可能无法在单个目录中处理大量的文件；您可能需要在 plsql_native_library_dir 指定的目录下创建子目录来存储纯编译生成的共享库。

如果您使用的是生产品质的文件系统（例如 Veritas），那么您可能无需创建子目录。然而，如果您使用的是 vanilla 文件系统，那么它可能无法在单个目录中轻松地处理大量的文件。对于拥有大约 70,000 个 PL/SQL 对象的 Oracle 应用程序，在使用 vanilla 文件系统时，我们使用了大约 500 个子目录。

这个参数的默认值是 0。为了指示 Oracle 将共享库分配到不同的子目录中，而不是将所有共享库存储在 plsql_native_library_dir 指定的单个目录中，请将 plsql_native_library_subdir_count 参数设为您要创建的子目录数。

例如，要使用 500 个子目录，则将以下行添加到您的初始化参数文件中：

plsql_native_library_subdir_count=500

然后您必须创建 500 个子目录（在 plsql_native_library_dir 参数指定的目录下），目录名称即为 d0、d1、d2、..、d498、d499。

 

--------------------------------------------------------------------------------

plsql_code_type 参数

plsql_code_type 参数决定对 PL/SQL 代码是进行纯编译还是解释。默认设置是 INTERPRETED。要启用 PL/SQL 纯编译，请将 plsql_code_type 的值设为 NATIVE。如果您想以 NATIVE 方式编译整个数据库，那么甲骨文公司建议您在系统级或在初始化参数文件中设置 plsql_code_type。

使用以下语法设置该参数：

对于纯编译模式：
alter session set plsql_code_type='NATIVE'
或者，
alter system set plsql_code_type='NATIVE'
对于解释模式：
alter session set plsql_code_type='INTERPRETED'
或者，
alter system set plsql_code_type='INTERPRETED'


--------------------------------------------------------------------------------

spnc_commands 文件的格式

$ORACLE_HOME/plsql 目录中的 spnc_commands 文件包含了编译和链接各个程序的命令模板。一些特殊的名称（如 %(src)）被预先定义，并用相应的文件名进行了替换。变量 $(ORACLE_HOME) 也自动被 Oracle 主目录的位置替换。您可以包含命令行（以 # 字符开始）。该文件包含解释所有特殊符号的注释。

spnc_commands 文件根据特定的操作系统为 C 编译器包含了一个预先定义的路径。（每种操作系统支持一种特定的编译器。）在大多数情况下，您不需要更改这个路径，但如果系统管理员将编译器安装在了另一个位置，您可能就需要更改它了。

这种方法意味着（与 9.0.1 版和 9.2.0 版不同）这里与 make 实用程序不存在依赖关系。

 

--------------------------------------------------------------------------------

纯编译生成的共享库存储在哪个字典表中？

纯编译生成的共享库被作为 BLOB 存储在数据库中的 ncomp_dll$ 字典表中。

SQL> connect / as sysdba;
Connected.
SQL> describe ncomp_dll$;
Name         Null?Type
 ------------   --------------    --------
OBJ#         NOT NULL          NUMBER
VERSION                    NUMBER
DLL             BLOB
DLLNAME                    RAW(1024)

 

--------------------------------------------------------------------------------

当删除一个"纯编译"单元时，文件系统上的共享库将发生什么？

直到 9.2.0，未发生任何事情。不会对共享库进行垃圾收集。

在 10.1.0 版中，当删除或在解释模式下重新编译一个纯编译的单元时，将从 ncomp_dll$ 字典表以及从 plsql_native_library_dir 参数指定的文件系统位置中删除该共享库。在 RAC 配置的情况下，将从每个实例的 plsql_native_library_dir 位置中删除该共享库。

 

--------------------------------------------------------------------------------

如何更改 plsql_native_library_dir？

如果您需要更改该参数的值，建议采取以下步骤：

关闭数据库实例
将初始化参数文件中的 plsql_native_library_dir 设置更改为指向新的位置。
重新启动数据库实例。Oracle 将根据需要自动将共享库提取到这个新的位置中。
注释 1：虽然可以通过 alter system..命令来更改这个参数，但 Oracle 不建议这么做。

注释 2：一旦执行了上述步骤，就可以安全地删除与 plsql_native_library_dir 先前的设置对应的原有目录了。Oracle 不会 自动从原有位置删除共享库。

 

--------------------------------------------------------------------------------

是否还有其他场合需要手动删除共享库？

没有。

唯一的例外是当您想修改 plsql_native_library_dir 的设置的时候。请按照先前所述步骤更改 plsql_native_library_dir。然后就可以从原有位置删除共享库了。

 

--------------------------------------------------------------------------------

如果误删除了一个共享库该怎么办？

这可能导致非预期的内部错误和问题。建议的补救办法是关闭然后重新启动实例。Oracle 将根据需要自动从 ncomp_dll$ 字典表中提取共享库。

 

--------------------------------------------------------------------------------

甲骨文公司是否建议混合使用纯编译单元和解释单元？

虽然支持混合使用，并且决不会导致错误的行为，但当纯编译单元调用解释单元时，需要付出一定代价。如果纯编译单元频繁调用解释单元，那么您可能牺牲一些潜在的性能改善。

如果将所有模块纯编译成共享库的时间是个问题，那么一种选择是考虑将程序包和类型规范在解释模式下进行编译，而将所有其他类型的 PL/SQL 单元（程序包主体、类型主体、过程、函数和触发器）在纯编译模式下进行编译。因为规范一般没有太多可执行代码，因此就运行时性能而言您不会有太大损失。

 

--------------------------------------------------------------------------------

在操作系统升级时，是否需要重新生成纯编译的共享库？

假设您已经确定甲骨文支持这种类型的 OS 升级，而无需重新生成普通的 Oracle 可执行文件，那么 PL/SQL 纯编译产生的共享库也将可用，而无需重新生成。

 

--------------------------------------------------------------------------------

在应用程序部署期间是否能够使用纯编译模式来节省时间（通过提供预先生成的 NCOMP 共享库）？

不能！即使共享库是在与最终部署环境相同的平台上生成的，您也无法这么做。您也无法在解释模式下利用 MCode 这么做。

其原因与编译模式无关。基于 PL/SQL 的应用程序在数据库中的安装需要在数据库中加载和编译 PL/SQL 模块，以便可以正确解析所有的外部引用。

 

--------------------------------------------------------------------------------

纯编译测试：简单测试

创建一个目录，用于存储纯编译生成的共享库，并在初始化参数文件中将 plsql_native_library_dir 参数设置为指向该目录路径。例如，您需要在初始化参数文件中添加如下内容：
plsql_native_library_dir=/home/kmuthukk/oracle10g/dbs/ncomp_libraries


还可以在初始化参数文件中设置 plsql_native_library_subdir_count 并创建子目录。
确保 spnc_commands 文件中使用的 C 编译器 -- 以及链接器（如果适用） -- 的 OS 路径是正确的。
系统参数完整性检查： 必须指定 plsql_native_library_dir。plsql_native_library_subdir_count 是可选的，可以保留为 0（如果您不打算使用基于子目录的模式）。最好在初始化参数文件中指定这些参数。
SQL> show parameters plsql_native
NAME         TYPE  VALUE
------------------------------------ ----------- ------------------------------
plsql_native_library_dir      string  /home/kmuthukk/oracle10g/dbs/ncomp_libraries
plsql_native_library_subdir_count    integer  0

对 PL/SQL 模块进行纯编译

要在纯编译模式下加载和编译新的模块，必须将 plsql_code_type 参数设为 NATIVE。
例如，尝试在 SQL*Plus 中执行以下操作：

connect scott/tiger
alter session set plsql_code_type='NATIVE';

create or replace procedure my_test is
begin
dbms_output.put_line('hello world');
end;
/
show errors;

set serveroutput on;
execute my_test;

验证步骤：

看一下 plsql_native_library_dir 指定的位置。应找到一个与 MY_TEST 过程相对应的新的共享库。在 Solaris 上，DLL 可能有一个类似 MY_TEST__SCOTT__P__54765.so 的名称。
[要阅读有关共享库命名惯例的更多信息，请点击此处。]


此外，通过执行以下查询，确保该模块的 plsql_code_type 设置的确为 NATIVE：
select name, type, plsql_code_type
from user_plsql_object_settings
where name = 'MY_TEST';

如果您想手动地对特定的已经加载的模块进行纯编译，那么您可以使用 alter ... compile .. 语句。
假定您有一个程序包 MYPKG，其规范和主体是用解释方式编译的。现在，如果您只想对其主体进行纯编译，那么可以执行以下命令：

alter package MYPKG compile body plsql_code_type=native reuse settings
  
上面的 reuse settings 子句指示编译器应当保留该单元所有以前的设置 -- 被显式覆盖的设置（如在上述示例中设定的 plsql_code_type 的设置）除外。

如果从上面的语句中删除了 body 子句，如：

alter package MYPKG compile plsql_code_type=native reuse settings
... 那么规范和主体都将在纯编译模式下重新编译。不过请注意，重新编译规范将造成依赖于该规范的其他模块失效。

 

--------------------------------------------------------------------------------

将数据库中的所有 PL/SQL 单元进行纯编译

在试图将数据库中的所有 PL/SQL 单元进行纯编译之前，在一个简单的测试案例上试验一下（如上一部分所述）。这将帮助确保您的纯编译设置是正确的。

操作系统文件描述符限制
$ORACLE_HOME/rdbms/admin 目录下的两个 SQL*Plus 脚本（dbmsupgnv.sql 和 utlrp.sql）用来在纯编译模式下重新编译所有的 PL/SQL 模块。它们使用并行作业（取决于可用的 CPU 的数量）。在这种情况下，许多作业同时生成 C 文件并试图编译它们。因此不将文件描述符的最大数的 OS 限制设得过低非常重要。

例如，在我的 Solaris 计算机上：

% ulimit -a
time(seconds)        unlimited
file(blocks)         unlimited
data(kbytes)         2097148
stack(kbytes)        8192
coredump(blocks)     3
nofiles(descriptors) 64

vmemory(kbytes)      unlimited

注意描述符的最大数（值为 64）。这在生产环境中太少了，因为有许多作业要进行并行纯编译。您可能想查看您计算机上的限制，并将其提高。例如：

在 Solaris 上，需要编辑 /etc/system 文件，使之包含：

set rlim_fd_max=4096
set rlim_fd_cur=1024

在 HP-UX 上，相应的参数位于 /stand/system 文件中。

maxfiles        2048
maxfiles_lim    2048


系统表空间
因为 DLLs 的主副本存储在字典表中，因此您可能需要比以前更多的 SYSTEM 表空间。如果您遇到了与 SYSTEM 表空间的空间限制相关的错误，那么您将需要执行如下操作：

alter tablespace system
add datafile '/home/kmuthukk/oracle10g/dbs/mydata10.dbf'
size 500 m;

.. 或在数据文件上设置 autoextend。


以纯编译模式重新编译所有的模块（dbmsupgnv.sql 和 utlrp.sql）

关闭所有的应用程序服务（包括 Forms Processe、Web Server、Reports Server 和 Concurrent Manager Server）。在关闭所有的应用程序服务之后，确保中断了所有与数据库的连接。

关闭数据库的 TNS 监听器，以确保不会创建新的连接。

关闭数据库（以正常模式或即时模式）。

确定您已在初始化参数文件中设置了 plsql_native_library_dir。如果没有，则请查看上一部分所述的步骤。

在初始化参数文件中将 plsql_code_type 设为 NATIVE。
plsql_code_type=NATIVE


在 UPGRADE 模式下启动数据库（使用 STARTUP UPGRADE 或 ALTER DATABASE OPEN UPGRADE）。

运行以下查询，获取无效对象的数量。这个值将在稍后用于作比较，以确保重新编译没有由于纯编译故障而产生额外的无效对象。
Select count(*)
From all_objects
Where status='INVALID';

以下查询将为您展示，纯编译模式和解释模式分别编译了多少对象。
SELECT plsql_code_type, count(*)
FROM dba_plsql_object_settings
GROUP BY plsql_code_type;

您可能看到一些对象的 plsql_code_type 为 NULL。这些是特殊的内部对象，可以忽略。


cd（更改目录）至 $ORACLE_HOME/rdbms/admin。

运行 $ORACLE_HOME/rdbms/admin/dbmsupgnv.sql（作为 SYS 用户）。
注释 1： dbmsupgnv.sql 使所有的 PL/SQL 模块失效并更新字典表，将它们的 plsql_code_type 设置修改为 NATIVE。这一阶段必须在数据库上没有其他活动进行时完成；因此，我们要求在运行 dbmsupgnv.sql 期间必须以 UPGRADE 模式启动数据库。

注释 2：如果您要从数据库的一个低版本向数据库高版本升级（例如，升级到 Oracle 数据库 10g），那么您必须首先运行升级脚本，然后运行 dbmsupgnv.sql。例如，如果您正从 9.2.0.3 升级到 10.1.0，那么请首先运行升级脚本 (u0902000.sql)，然后关闭数据库并运行 dbmsupgnv.sql 脚本。


在运行 dbmsupgnv.sql 之后提交更改
SQL> commit;


关闭数据库。

以正常模式重新启动数据库。

运行 $ORACLE_HOME/rdbms/admin/utlrp.sql 脚本（作为 SYS 用户）。这一步骤将重新编译所有的 PL/SQL 模块。[在这个过程期间，您可能想定期查看 plsql_native_library_dir 指定的目录，以确保编译过程的确创建了新的共享库。]
注意：这一步骤是可重新启动的。如果由于任何原因这个步骤异常终止，那么您只需重新启动这个步骤（即重新运行 utlrp.sql 脚本）。它将重新编译余下的任何无效的 PL/SQL 模块。


在编译成功完成之后，您应当验证无效对象数少于或等于重新编译之前的无效对象数。您可以使用以下查询获取无效对象数。
Select count(*)
From all_objects
Where status='INVALID';

再运行以下查询，以确保没有遗漏解释的模块：

SELECT plsql_code_type, count(*)
FROM dba_plsql_object_settings
GROUP BY plsql_code_type;

]]> Oracle 11g Linux单机STANDBY配置 tag:www.heysky.net,2008:/digest//1.326 2008-07-28T19:28:45Z 2008-07-28T19:32:57Z Sky http://www.heysky.net From: http://www.itpub.net/835722.html By blue_prince

 

Oracle 11g Linux单机STANDBY配置

环境：RHEL 4 U5+Oracle 11.1.0.6
主库SID：ora11g 备用库SID:standby
主库数据文件存放目录：/home/oracle/opt/oradata/ora11g/
备用库数据文件存放目录：/home/oracle/opt/oradata/standby/

1、        验证主库是否归档，如果是未归档模式的话必须更改为归档模式：

Sys@ORA11G> archive log list
Database log mode              Archive Mode
Automatic archival             Enabled
Archive destination            /home/oracle/opt/oradata/ora11g/archive
Oldest online log sequence     61
Next log sequence to archive   63
Current log sequence           63

2、        将主库置于FORCE LOGGING模式：

Sys@ORA11G> alter database force logging;

3、        对主库做一个全库的RMAN备份，用于STANDBY配置： 
PHP code:



[oracle@test51 bin]$ ./rman target /





Recovery Manager: Release 11.1.0.6.0 - Production on Thu Aug 16 15:51:22 2007





Copyright (c) 1982, 2007, Oracle.  All rights reserved.





connected to target database: ORA11G (DBID=4026454982)





RMAN> backup database format '/home/oracle/liuyun/%U.bak';





Starting backup at 16-AUG-07


using target database control file instead of recovery catalog


allocated channel: ORA_DISK_1


channel ORA_DISK_1: SID=121 device type=DISK


channel ORA_DISK_1: starting full datafile backup set


channel ORA_DISK_1: specifying datafile(s) in backup set


input datafile file number=00004 name=/home/oracle/opt/oradata/ora11g/users01.dbf


input datafile file number=00002 name=/home/oracle/opt/oradata/ora11g/sysaux01.dbf


input datafile file number=00001 name=/home/oracle/opt/oradata/ora11g/system01.dbf


input datafile file number=00005 name=/home/oracle/opt/oradata/ora11g/example01.dbf


input datafile file number=00003 name=/home/oracle/opt/oradata/ora11g/undotbs01.dbf


channel ORA_DISK_1: starting piece 1 at 16-AUG-07


channel ORA_DISK_1: finished piece 1 at 16-AUG-07


piece handle=/home/oracle/liuyun/03iphp8l_1_1.bak tag=TAG20070816T155148 comment=NONE


channel ORA_DISK_1: backup set complete, elapsed time: 00:01:45


channel ORA_DISK_1: starting full datafile backup set


channel ORA_DISK_1: specifying datafile(s) in backup set


including current control file in backup set


including current SPFILE in backup set


channel ORA_DISK_1: starting piece 1 at 16-AUG-07


channel ORA_DISK_1: finished piece 1 at 16-AUG-07


piece handle=/home/oracle/liuyun/04iphpc9_1_1.bak tag=TAG20070816T155148 comment=NONE


channel ORA_DISK_1: backup set complete, elapsed time: 00:00:01


Finished backup at 16-AUG-07


..
4、        准备STANDBY数据文件路径和其他路径：

[oracle@test51 oradata]$ pwd  
/home/oracle/opt/oradata
[oracle@test51 oradata]$ mkdir standby
[oracle@test51 oradata]$ ls
ora11g  orcl  standby
[oracle@test51 oradata]$ cd $ORACLE_BASE/admin
[oracle@test51 admin]$ mkdir standby
[oracle@test51 admin]$ ls
ora11g  orcl  standby
[oracle@test51 standby]$mkdir adump
[oracle@test51 standby]$mkdir dpdump
[oracle@test51 standby]$mkdir pfile
[oracle@test51 standby]$ ls
adump  dpdump  pfile

5、更改tnsnames配置，添加主库和备用库的连接字：

primary =
  (DESCRIPTION =
    (ADDRESS = (PROTOCOL = TCP)(HOST = test51)(PORT = 1522))
    (CONNECT_DATA =
      (SERVER = DEDICATED)
      (SERVICE_NAME = ora11g)
    )
  )

standby =
  (DESCRIPTION =
    (ADDRESS = (PROTOCOL = TCP)(HOST = test51)(PORT = 1522))
    (CONNECT_DATA =
      (SERVER = DEDICATED)
      (SERVICE_NAME = standby)
    )
  )

6、生成STANDBY控制文件：

Sys@ORA11G> alter database create standby controlfile as '/home/oracle/opt/oradata/standby/control01.ctl';

Database altered.

[oracle@test51 standby]$ cp control01.ctl control02.ctl
[oracle@test51 standby]$ cp control01.ctl control03.ctl
[oracle@test51 standby]$ ls
archive  control01.ctl  control02.ctl  control03.ctl

6、生成standby 初始化参数文件：

Sys@ORA11G> create pfile='$ORACLE_BASE/admin/standby/pfile/init.ora' from spfile;

File created.

添加以下几个参数，需要注意的是从11G开始原来备用库归档参数standby_archive_dest这个参数已经废除了，STANDBY的归档路径改为常规的归档路径log_archive_dest_n。还有原来单机配置STANDBY需要的参数lock_name_space也废除了：

*.log_archive_config='DG_CONFIG=(ora11g,standby)'
*.fal_client='standby'
*.fal_server='primary'
*.db_file_name_convert='ora11g','standby'
*.log_file_name_convert='ora11g','standby'
*.standby_file_management='auto'
*.log_archive_dest_1='location=/home/oracle/opt/oradata/standby/archive VALID_FOR=(STANDBY_LOGFILE,STANDBY_ROLE)'

更改后的参数文件如下，注意db_unique_name要和主库区分，否则无法MOUNT STANDBY：

standby.__db_cache_size=234881024
standby.__java_pool_size=12582912
standby.__large_pool_size=4194304
standby.__oracle_base='/home/oracle/opt'#ORACLE_BASE set from environment
standby.__pga_aggregate_target=209715200
standby.__sga_target=629145600
standby.__shared_io_pool_size=0
standby.__shared_pool_size=369098752
standby.__streams_pool_size=0
*.audit_file_dest='/home/oracle/opt/admin/standby/adump'
*.audit_trail='db'
*.compatible='11.1.0.0.0'
*.control_files='/home/oracle/opt/oradata/standby/control01.ctl','/home/oracle/opt/oradata/standby/control02.ctl','/home/oracle/opt/oradata/standby/control03.ctl'
*.db_block_size=8192
*.db_domain=''
*.db_name='ora11g'
*.db_recovery_file_dest='/home/oracle/opt/flash_recovery_area'
*.db_recovery_file_dest_size=2147483648
*.db_unique_name='standby'
*.ddl_lock_timeout=10
*.diagnostic_dest='/home/oracle/opt'
*.dispatchers=''
*.job_queue_processes=0
*.local_listener='LISTENER_ora11g'
*.memory_target=838860800
*.open_cursors=300
*.processes=150
*.remote_login_passwordfile='EXCLUSIVE'
*.undo_tablespace='UNDOTBS1'

*.log_archive_config='DG_CONFIG=(ora11g,standby)'
*.fal_client='standby'
*.fal_server='primary'
*.db_file_name_convert='ora11g','standby'
*.log_file_name_convert='ora11g','standby'
*.standby_file_management='auto'
*.log_archive_dest_1='location=/home/oracle/opt/oradata/standby/archive VALID_FOR=(STANDBY_LOGFILE,STANDBY_ROLE)'

7、启动STANDBY数据库，进行RESTORE： 
PHP code:



[oracle@test51 standby]$ export ORACLE_SID=standby


[oracle@test51 standby]$ sql





SQL*Plus: Release 10.2.0.1.0 - Production on Thu Aug 16 16:27:08 2007





Copyright (c) 1982, 2005, Oracle.  All rights reserved.





Connected to an idle instance.





idle> startup nomount pfile='$ORACLE_BASE/admin/standby/pfile/init.ora'


ORACLE instance started.





Total System Global Area  836976640 bytes


Fixed Size                  1303132 bytes


Variable Size             595594660 bytes


Database Buffers          234881024 bytes


Redo Buffers                5197824 bytes


idle> alter database mount standby database;





Database altered.





[oracle@test51 bin]$ ./rman target /





Recovery Manager: Release 11.1.0.6.0 - Production on Thu Aug 16 16:30:11 2007





Copyright (c) 1982, 2007, Oracle.  All rights reserved.





connected to target database: ORA11G (DBID=4026454982, not open)





RMAN> restore database;





Starting restore at 16-AUG-07


Starting implicit crosscheck backup at 16-AUG-07


using target database control file instead of recovery catalog


allocated channel: ORA_DISK_1


channel ORA_DISK_1: SID=147 device type=DISK


Crosschecked 4 objects


Finished implicit crosscheck backup at 16-AUG-07





Starting implicit crosscheck copy at 16-AUG-07


using channel ORA_DISK_1


Crosschecked 2 objects


Finished implicit crosscheck copy at 16-AUG-07





searching for all files in the recovery area


cataloging files...


no files cataloged





using channel ORA_DISK_1





channel ORA_DISK_1: starting datafile backup set restore


channel ORA_DISK_1: specifying datafile(s) to restore from backup set


channel ORA_DISK_1: restoring datafile 00001 to /home/oracle/opt/oradata/standby/system01.dbf


channel ORA_DISK_1: restoring datafile 00002 to /home/oracle/opt/oradata/standby/sysaux01.dbf


channel ORA_DISK_1: restoring datafile 00003 to /home/oracle/opt/oradata/standby/undotbs01.dbf


channel ORA_DISK_1: restoring datafile 00004 to /home/oracle/opt/oradata/standby/users01.dbf


channel ORA_DISK_1: restoring datafile 00005 to /home/oracle/opt/oradata/standby/example01.dbf


channel ORA_DISK_1: reading from backup piece /home/oracle/liuyun/03iphp8l_1_1.bak


channel ORA_DISK_1: piece handle=/home/oracle/liuyun/03iphp8l_1_1.bak tag=TAG20070816T155148


channel ORA_DISK_1: restored backup piece 1


channel ORA_DISK_1: restore complete, elapsed time: 00:02:05


Finished restore at 16-AUG-07


..
8、添加STANDBY LOGFILE，启动STANDBY至恢复管理模式：
[php]

idle> alter database add standby logfile '/home/oracle/opt/oradata/standby/redo04.log' size 50M;

Database altered.

Elapsed: 00:00:01.06
idle> alter database add standby logfile '/home/oracle/opt/oradata/standby/redo05.log' size 50M;

Database altered.

Elapsed: 00:00:01.18
idle> alter database add standby logfile '/home/oracle/opt/oradata/standby/redo06.log' size 50M;

Database altered.

Elapsed: 00:00:00.85

idle> recover managed standby database disconnect from session;
Media recovery complete.
..
[php]
9、主库配置到STANDBY的归档：

Sys@ORA11G> alter system set log_archive_dest_state_2=enable;

System altered.

Sys@ORA11G> alter system set log_archive_dest_2='service=standby ASYNC VALID_FOR=(ONLINE_LOGFILES,PRIMARY_ROLE) DB_UNIQUE_NAME=standby';

System altered.

VALID_FOR=(ONLINE_LOGFILES,PRIMARY_ROLE)这个在我配置的过程中是必须要的，我自作主张把里面的PRIMARY_ROLE改为STANDBY _ROLE，归档无法从主库传至STANDBY了。

10、可以看到STANDBY端已经收到主库传递过来的归档日志了：

[oracle@test51 archive]$ ls -l
total 7236
-rw-r-----  1 oracle oracle 7392768 Aug 16 16:46 1_63_630254857.dbf

Redo Shipping Client Connected as PUBLIC
-- Connected User is Valid
RFS[2]: Assigned to RFS process 5315
RFS[2]: Identified database type as 'physical standby'
Primary database is in MAXIMUM PERFORMANCE mode
Primary database is in MAXIMUM PERFORMANCE mode
RFS[2]: Successfully opened standby log 4: '/home/oracle/opt/oradata/standby/redo04.log'
Thu Aug 16 16:46:21 2007
Media Recovery Log /home/oracle/opt/oradata/standby/archive/1_63_630254857.dbf
Media Recovery Waiting for thread 1 sequence 64 (in transit)

这样STANDBY就配置完成了。
]]> 在linux上单机模拟Oracle 10g RAC集群（OCFS+ASM) tag:www.heysky.net,2008:/digest//1.325 2008-07-28T19:13:50Z 2008-07-28T19:26:54Z Sky http://www.heysky.net From: http://www.linuxsir.org/bbs/thread198024.html By daniey


Word 文档下载：  linux_srac.doc

 

在linux上单机模拟Oracle 10g RAC集群（OCFS+ASM)

       作者：daniey MSN:danieyren@hotmail.com

申明：请珍惜作者劳动，如需要转载，请注明出处

    本文是在参考oracle 10g 双机集群的基础上通过实践并且精简而得，撰写本文的目的是为了方便大家探讨技术，相信本文仍有许多不足之处

参考文档：http://www.oracle.com/technology/global/cn/pub/articles/hunter_rac10g.html

注：对于与多机多节点集群相同的地方，文中将直接引用参考文档相关部分

一、简介

    熟悉 Oracle 真正应用集群 (RAC) 10g 技术的最有效方法之一是访问一个实际的 Oracle RAC 10g 集群。没有什么方法比直接体验它们能够更好地理解其好处的了 -- 包括容错、安全性、负载均衡和可伸缩性。

    Oracle RAC 的核心是共享磁盘子系统。集群中的所有节点必须能够访问集群中所有节点的所有数据、重做日志文件、控制文件和参数文件。数据磁盘必须在全局范围内可用，以便允许所有节点访问数据库。每个节点拥有自己的重做日志和控制文件，但是其他节点必须能够访问这些文件，以便在系统故障时恢复该节点。

  希望本文能对一些手头上只有一台PC而没有真正的双机环境的朋友提供一个参考。

二、本文实践环境

1、PC主要配置：

Celeron(R) CPU 1.80GHz

Maxtor 6E040L0, ATA DISK drive 40G

RelTek 8139C 网卡 一块

内存DDR333 512MB * 2

显示卡：ATI [Radeon 9200 SE]

操作系统：White Box Enterprise Linux 3 （本文同样适用于RedHat企业版,不同之处是WBEL3目前可以免费运用于商业)

远程终端一台，装有windowsXP和远程X-server软件

2、服务器分区方案

Oracle 数据库文件

RAC节点名

实例名

数据库名

$ORACLE_BASE

文件系统

dbrac

orcl1

orcl

/home/oracle

ASM

Oracle CRS 共享文件

文件类型

文件名

分区

挂载点

文件系统

oracle集群注册表

/u01/orcl/orcfile

/dev/hda8

/u01

OCFS

CRS表决磁盘

/u01/orcl/cssfile

/dev/hda8

/u01

OCFS

3、所涉及软件

1)    oracle 10g 数据库软件

ship.db.lnx32.cpio.gz

2)    oracle 10g 集群服务软件

ship.crs.lnx32.cpio.gz

3)    OCFS文件系统支持

ocfs-2.4.21-EL-1.0.14-1.i686.rpm

ocfs-support-1.0.10-1.i386.rpm

ocfs-tools-1.0.10-1.i386.rpm

4)    ASMlib驱动程序

oracleasm-2.4.21-EL-1.0.3-1.i686.rpm

oracleasm-support-1.0.3-1.i386.rpm

oracleasmlib-1.0.0-1.i386.rpm

以上软件包均可以从oracle官方网站下载

WBEL linux下载地址：http://www.whiteboxlinux.org/download.html    

三、基础操作

1、安装linux

安装过程中一些需要注意的地方：

1)    磁盘分区:swap分区大小建议是内存的2倍，这里是2048MB,划出一些系统必要的分区根分区/,var分区/var,usr分区/usr,home分区/home,临时文件分区/tmp。注意：这里不要把所有的硬盘空间划分进操作系统，留下一半给后面安装oarcle集群磁盘使用，本文示例

2)    文件系统     容量  挂载点

/dev/hda1          1012M  /

/dev/hda2          7.7G   /home

/dev/hda7          1012M   /tmp

/dev/hda3          5.8G   /usr

/dev/hda5          2.0G   /var

3)    组件选取：一定要选上delvelopment tools和X-windows两项，为了节省空间其他可以不要

4)    防火墙：最好不要

5)    网络设置：eth0

取消选中 [Configure using DHCP] 复选项

选中 [Activate on boot]

IP 地址：192.168.22.44

网络掩码： 255.255.255.0

6)    主机名:dbrac

2、.安装完成后检查必需的 RPM

3、必须安装以下程序包（或更高版本）：

make-3.79.1

gcc-3.2.3-34

glibc-2.3.2-95.20

glibc-devel-2.3.2-95.20

glibc-headers-2.3.2-95.20

glibc-kernheaders-2.4-8.34

cpp-3.2.3-34

compat-db-4.0.14-5

compat-gcc-7.3-2.96.128

compat-gcc-c++-7.3-2.96.128

compat-libstdc++-7.3-2.96.128

compat-libstdc++-devel-7.3-2.96.128

openmotif-2.2.2-16

setarch-1.3-1

四、设置

1、更改/etc/hosts

vi /etc/hosts

127.0.0.1                localhost.localdomain localhost

192.168.22.44   dbrac   int-dbrac

192.168.22.244       vip-dbrac

确保RAC节点名没有出现在回送地址中。

此处设置相当重要，不能跳过，一定按照此设置，IP和主机别名可以自己定

oracle 10g RAC中 使用了虚拟IP(VIP)技术，这是一个令人心动的高可用性、多机无缝切换的解决方案,但在单机模拟环境中仅仅是个形式而已，为了以后的顺利安装，不得不配置它

2、调整内核网络设置参数

编辑/etc/sysctl.conf，增加下面的设置：

vi /etc/sysctl.conf

# Default setting in bytes of the socket receive buffer

net.core.rmem_default=262144

# Default setting in bytes of the socket send buffer

net.core.wmem_default=262144

# Maximum socket receive buffer size which may be set by using

# the SO_RCVBUF socket option

net.core.rmem_max=262144

# Maximum socket send buffer size which may be set by using

# the SO_SNDBUF socket option

net.core.wmem_max=262144

3、添加模块选项：

将下列行添加到 /etc/modules.conf 中：

options sbp2 sbp2_exclusive_login=0

4、创建"oracle"用户和目录

$su -

#groupadd dba

#useradd -g dba -m oracle

#passwd oracle

5、编辑.bash_profile文件，增加oracle环境变量

$vi .bash_profile

export PATH

unset USERNAME

export LANG=zh_CN.EUC

ORACLE_BASE=/home/oracle;export ORACLE_BASE

export ORACLE_HOME=$ORACLE_BASE/product/10.1.0/db_1

export ORA_CRS_HOME=$ORACLE_BASE/product/10.1.0/crs_1

export ORACLE_SID=rac1

export NLS_LANG='SIMPLIFIED CHINESE_CHINA.ZHS16GBK'

PATH=$ORACLE_HOME/bin:/bin:/sbin:/usr/bin:/usr/ccs/bin:/usr/local/bin:/usr/ucb;export PATH

LD_LIBRARY_PATH=$ORACLE_HOME/lib:$ORACLE_HOME/network/lib:$ORACLE_HOME/oracm/lib:/usr/local/lib:/usr/lib;export LD_LIBRARY_PATH

export ORACLE_TERM=xterm

export CLASSPATH=$ORACLE_HOME/JRE:$ORACLE_HOME/jlib:$ORACLE_HOME/rdbms/jlib:$ORACLE_HOME/rdbms/jlib:$ORACLE_HOME/network/jlib

export THREADS_FLAG=native

export TEMP=/tmp

export TMPDIR=/tmp

export LD_ASSUME_KERNEL=2.4.1

6、创建CRS分区和数据文件分区

1)    先建立CRS分区挂载点

mkdir /u01

chown oracle:dba /u01

2)    接着创建CRS分区和共享数据文件分区

fdisk /dev/hda

CRS分区只要500M就够了，剩下的全部划分给数据文件分区，这里为数据文件只分了1个区/dev/hda9

CRS分区为/dev/hda8

[root@dbrac root]# fdisk /dev/hda

The number of cylinders for this disk is set to 4997.

[root@dbrac root]# fdisk /dev/hda

The number of cylinders for this disk is set to 4997.

There is nothing wrong with that, but this is larger than 1024,

and could in certain setups cause problems with:

1) software that runs at boot time (e.g., old versions of LILO)

2) booting and partitioning software from other OSs

   (e.g., DOS FDISK, OS/2 FDISK)

Command (m for help): p

Disk /dev/hda: 41.1 GB, 41109061120 bytes

255 heads, 63 sectors/track, 4997 cylinders

Units = cylinders of 16065 * 512 = 8225280 bytes

   Device Boot    Start       End    Blocks   Id  System

/dev/hda1   *         1       131   1052226   83  Linux

/dev/hda2           132      1151   8193150   83  Linux

/dev/hda3          1152      1916   6144862+  83  Linux

/dev/hda4          1917      4998  24756165    f  Win95 Ext'd (LBA)

/dev/hda5          1917      2177   2096451   83  Linux

/dev/hda6          2178      2438   2096451   82  Linux swap

/dev/hda7          2439      2569   1052226   83  Linux

Command (m for help): n

First cylinder (2570-4998, default 2570):

Using default value 2570

Last cylinder or +size or +sizeM or +sizeK (2570-4998, default 4998): +500M

Command (m for help): n

First cylinder (2632-4998, default 2632):

Using default value 2632

Last cylinder or +size or +sizeM or +sizeK (2632-4998, default 4998): +15000M

Command (m for help): p

Disk /dev/hda: 41.1 GB, 41109061120 bytes

255 heads, 63 sectors/track, 4997 cylinders

Units = cylinders of 16065 * 512 = 8225280 bytes

   Device Boot    Start       End    Blocks   Id  System

/dev/hda1   *         1       131   1052226   83  Linux

/dev/hda2           132      1151   8193150   83  Linux

/dev/hda3          1152      1916   6144862+  83  Linux

/dev/hda4          1917      4998  24756165    f  Win95 Ext'd (LBA)

/dev/hda5          1917      2177   2096451   83  Linux

/dev/hda6          2178      2438   2096451   82  Linux swap

/dev/hda7          2439      2569   1052226   83  Linux

/dev/hda8          2570      2631    497983+  83  Linux

/dev/hda9          2632      4456  14659281   83  Linux

Command (m for help): w

The partition table has been altered!

Calling ioctl() to re-read partition table.

WARNING: Re-reading the partition table failed with error 16: 设备或资源忙.

The kernel still uses the old table.

The new table will be used at the next reboot.

Syncing disks.

7、编辑/etc/sysctl.conf，增加下面2行以设置系统共享内存大小和文件句柄

kernel.shmmax=2147483648

kernel.sem=250 32000 100 128

本文涉及到的linux操作系统的内核参数默认设置大多符合oracle安装所需，不需要做更大的改动

8、配置 hangcheck-timer 内核模块

在/etc/modules.conf中增加下面一行

options hangcheck-timer hangcheck_tick=30 hangcheck_margin=180

为了确保系统每次重新启动的时候都能自动加载hangcheck-timer 模块，需要在/etc/rc.local文件中增加下面一行

echo "modprobe hangcheck-timer" >> /etc/rc.local

重启系统并检查hangcheck-timer模块是否已经加载

[root@dbrac root]# lsmod | grep hangcheck-timer

hangcheck-timer         2616   0  (unused)

9、配置 RAC 节点以进行远程访问

    在 RAC 节点上运行 Oracle Universal Installer 时，它将使用 rsh 、rcp或scp命令将 Oracle 软件复制到 RAC 集群中的所有其他节点。虽然是单机模拟，但仍然要配置，无法跳过，从oracle 10g开始已经支持ssh协议，本文将首先尝试使用它

使用oracle用户创建ssh公共密匙：

[oracle@dbrac oracle]$ ssh-keygen -t dsa

Generating public/private dsa key pair.

Enter file in which to save the key (/home/oracle/.ssh/id_dsa):

Enter passphrase (empty for no passphrase):

Enter same passphrase again:

Your identification has been saved in /home/oracle/.ssh/id_dsa.

Your public key has been saved in /home/oracle/.ssh/id_dsa.pub.

The key fingerprint is:

2d:09:9a:c0:40:c7:99:46:ea:43:0d:22:4b:d0:a0:26 oracle@dbrac

复制公匙到其他节点（这里是单机）

[oracle@dbrac oracle]$ cp -v .ssh/id_dsa.pub .ssh/authorized_keys

测试密匙是否生效

[oracle@dbrac oracle]$ ssh dbrac

The authenticity of host 'dbrac (192.168.22.44)' can't be established.

RSA key fingerprint is e7:ff:ce:5e:92:ac:c4:96:a8:ca:3e:20:2e:5c:75:ae.

Are you sure you want to continue connecting (yes/no)? yes

Warning: Permanently added 'dbrac,192.168.22.44' (RSA) to the list of known hosts.

[oracle@dbrac oracle]$

无需密码即可登陆表示密匙已经生效

10、安装和配置 OCFS

OCFS是Oracle 集群文件系统 (OCFS)，由 Oracle 开发，用于消除数据库管理员和系统管理员管理原始设备这一负担，它提供了与通常的文件系统相同的功能和用法。尽量不要在OCFS文件系统上使用linux二进制文件系统操作命令

在目前的版本１种支持以下文件类型：

·       Oracle 数据库文件

·       联机重做日志文件

·       归档重做日志文件

·       控制文件

·       服务器参数文件 (SPFILE)

·       Oracle 集群注册表 (OCR) 文件

·       CRS 表决磁盘。

 

安装和配置