背景说明:
今天查阅书籍时,偶然间发现“在对某个索引行执行删除操作时,只是为该行增加了一个删除标记,这个索引行并不会释放它的存储空间,Insert产生的新的索引行也不能被插入到该位置。索引列的修改过程其实是将对应的列值删除,然后再插入新的列值(与数据行本身的修改是不一致的,这也正是我们尽量不使用修改频繁的列来创建索引的原因)。所以,无论是插入、修改、删除,都需要消耗存储空间,增大B-Tree索引结构的深度,影响数据的查询速度。尤其是删除和修改,不仅造成了存储空间的浪费,而且增加了扫描索引块的数量”,这就是所谓的索引碎片问题,建议定期对经常使用的表执行检查和重建索引操作。
问题重现:
经测试,收集统计信息等操作,无法释放索引删除块所占用的存储空间。
analyze table tkk29 compute statistics;
select t.index_name, t.distinct_keys, t.num_rows, t.sample_size, t.last_analyzed
, t.blevel, t.leaf_blocks, t.*
from user_indexes t
where t.table_name = upper('tkk29');
delete from tkk29
where mod(trunc((sysdate-createddate) * 24 * 60), 2) = 0;
analyze table tkk29 compute statistics;
analyze index IDX_tkk29_PARTICIPANT validate structure;
select t.name, t.blocks, t.lf_rows, t.del_lf_rows, t.lf_rows - t.del_lf_rows as lf_rows_used
, to_char((t.del_lf_rows/t.lf_rows) * 100, '999.999') as ratio, t.*
from index_stats t
alter index IDX_tkk29_PARTICIPANT rebuild;
alter index IDX_tkk29_ACTUALPARTICIPANT rebuild;
analyze index IDX_tkk29_PARTICIPANT validate structure;
重建索引:
CREATE OR REPLACE TYPE strsplit_type AS TABLE OF VARCHAR2(32676); CREATE OR REPLACE FUNCTION strsplit(p_value VARCHAR2, p_split VARCHAR2 := ',') --usage: select * from table(strsplit('1,2,3,4,5')) RETURN strsplit_type PIPELINED IS v_idx INTEGER; v_str VARCHAR2(500); v_strs_last VARCHAR2(4000) := p_value; BEGIN LOOP v_idx := instr(v_strs_last, p_split); EXIT WHEN v_idx = 0; v_str := substr(v_strs_last, 1, v_idx - 1); v_strs_last := substr(v_strs_last, v_idx + 1); PIPE ROW(v_str); END LOOP; PIPE ROW(v_strs_last); RETURN; END strsplit;
CREATE OR REPLACE PROCEDURE UP_CHECK_TO_REBUILD_INDEX ( tbNames varchar ) IS sqlstr VARCHAR2(100); idx_ratio INT; BEGIN --DECLARE sqlstr VARCHAR2(100); -- idx_ratio INT; BEGIN FOR idx IN (SELECT t.index_name FROM user_indexes t WHERE t.index_type = 'NORMAL' AND t.status = 'VALID' AND t.temporary = 'N' AND t.leaf_blocks > 100 AND t.table_name IN (SELECT UPPER(TRIM(COLUMN_VALUE)) from table(strsplit(tbNames))) --//('TKK29') ORDER BY t.table_name, t.index_name ) LOOP DBMS_OUTPUT.put_LINE(idx.index_name || ' ANALYZE START ' || TO_CHAR(SYSDATE, 'yyyy-MM-dd hh24:mi:ss')); sqlstr := 'ANALYZE INDEX ' || idx.Index_Name || ' VALIDATE STRUCTURE'; EXECUTE IMMEDIATE sqlstr; SELECT TRUNC((t.del_lf_rows/t.lf_rows) * 100) INTO idx_ratio FROM index_stats t WHERE t.name=idx.index_name AND ROWNUM=1; IF (idx_ratio >= 15) THEN DBMS_OUTPUT.put_line(' REINDEX ' || TO_CHAR(SYSDATE, 'yyyy-MM-dd hh24:mi:ss') || ' ratio: ' || idx_ratio); sqlstr := 'ALTER INDEX ' || idx.index_name || ' REBUILD'; EXECUTE IMMEDIATE sqlstr; END IF; END LOOP; END; END UP_CHECK_TO_REBUILD_INDEX; SQL>exec UP_CHECK_TO_REBUILD_INDEX('TKK29, muser'); begin UP_CHECK_TO_REBUILD_INDEX('TKK29, muser'); end;
PK_MUSER ANALYZE START 2016-01-29 17:49:19
IDX_TKK29_ACTIVITYINSTANCEID ANALYZE START 2016-01-29 17:49:19
REBUILD INDEX START 2016-01-29 17:49:20 ratio: 50
IDX_TKK29_ACTUALPARTICIPANT ANALYZE START 2016-01-29 17:49:22
IDX_TKK29_COMPLETEDDATE ANALYZE START 2016-01-29 17:49:22
REBUILD INDEX START 2016-01-29 17:49:22 ratio: 36
IDX_TKK29_PARTICIPANT ANALYZE START 2016-01-29 17:49:23
IDX_TKK29_PROCESSINSTANCEID ANALYZE START 2016-01-29 17:49:23
REBUILD INDEX START 2016-01-29 17:49:24 ratio: 50
IDX_TKK29_STATEDDATE ANALYZE START 2016-01-29 17:49:25
REBUILD INDEX START 2016-01-29 17:49:25 ratio: 33
PK_TKK29 ANALYZE START 2016-01-29 17:49:27
REBUILD INDEX START 2016-01-29 17:49:27 ratio: 50
备注:
真实场景请考虑索引列的修改、数据删除的概率,结合表的数据量大小等多种因素制定合理的维护计划;另外,分区表的不同分区应该有不同的策略。
来源:https://www.cnblogs.com/zhaoguan_wang/p/5169821.html