Can gather_stats_job delete my statistics that collected via dbms_stats...

Question

Hello all,
We are using automatic_statistic_gathering via oracle bundle (gather_stats_job) in our 10.2 database.
We have a large partitioned table. We gathered statistics for this table and its indexes by using dbms_stats package.
Yesterday we found some of table's index partition's statistics not all as deleted state !
How can we fix this issue?
I wonder that what is a relation between statistics collected via dbms_stats package and "gather_stats_job" job's one?
Thank you?

Answer 1

Sukan wrote:
select * from v$version;
Oracle Database 10g Enterprise Edition Release 10.2.0.3.0 - 64bi
PL/SQL Release 10.2.0.3.0 - Production
"CORE     10.2.0.3.0     Production"
TNS for IBM/AIX RISC System/6000: Version 10.2.0.3.0 - Productio
NLSRTL Version 10.2.0.3.0 - Production
Beside;
we found index statistics as deleted via belov sql statement;
select * from dba_ind_partitions where index_owner = 'HIST'
When we check the distinct_keys, num_rows and clustering_factor, we found them as different from the actual value.
Table actual number of rows are about 6 million but we saw it as 65 thousand.
I think gather_stats_job might choose the sample method for collect statistics then it deleted my statistics.If I understand you correctly, then the statistics are not "deleted" (so that no statistics are there) but replaced with statistics obviously gathered by the default "automatic statistics" collection job.
You can use the view DBA_TAB_STATS_HISTORY to check for modifications to the statistics.
Note that depending on how you collect your statistics manually, these might be considered as "STALE" although they're actually not stale. E.g. in case of index statistics, if you gather the index statistics separately, e.g. using DBMS_STATS.GATHER_INDEX_STATS, and afterwards collect the corresponding table statistics using "DBMS_STATS.GATHER_TABLE_STATS(...cascade=>false...)", then index statistics will show up as "STALE" in the dictionary, because the dictionary view contains the simple check if the table statistics are more recent than the index statistics.
If you want to make sure that any individual statistics collected are left untouched, you can "lock" the statistics on particular tables or schemas using DBMS_STATS.LOCK_TABLE/SCHEMA_STATS. You need then to use the "FORCE=>true" parameter when gathering statistics yourself to overwrite the lock. Note that there is a peculiarity when having the statistics locked: ALTER INDEX REBUILD resp. CREATE INDEX won't update the statistics any longer (which is consistent, since the statistics are locked) but there is no "FORCE" option to the commands, so you need to take care of this by either collecting statistics separately or unlocking the stats temporarily.
Alternatively, if you don't want the default statistics job to work on any of your objects, you can re-configure it using "DBMS_STATS.SET_PARAM('AUTOSTATS_TARGET', 'ORACLE')" in which case the job only analyzes objects owned and maintained by Oracle (data dictionary, workload repository, etc.).
Regards,
Randolf
Oracle related stuff blog:
http://oracle-randolf.blogspot.com/
SQLTools++ for Oracle (Open source Oracle GUI for Windows):
http://www.sqltools-plusplus.org:7676/
http://sourceforge.net/projects/sqlt-pp/