We have a Oracle 19C database (19.0.0.0.ru-2021-04.rur-2021-04.r1) on AWS RDS which is hosted on an 4 CPU 32 GB RAM instance. The size of the database is not big (35 GB) and the PGA Aggregate Limit is 8GB & Target is 4GB. Whenever the scheduled internal Oracle Auto Optimizer Stats Collection Job (ORA$AT_OS_OPT_SY_nnn) runs then it consumes substantially high PGA memory (approx 7GB) and sometimes this makes database unstable and AWS loses communication with the RDS instance so it restarts the database.
We thought this may be linked to existing Oracle bug 30846782 (19C+: Fast/Excessive PGA growth when using DBMS_STATS.GATHER_TABLE_STATS) but Oracle & AWS had fixed it in the current 19C version we are using. There are no application level operations that consume this much PGA and the database restart have always happened when the Auto Optimizer Stats Collection Job was running. There are couple of more databases, which are on same version, where same pattern was observed and the database was restarted by AWS. We have disabled the job now on those databases to avoid further occurrence of this issue however we want to run this job as disabling it may cause old stats being available in the database.
Any pointers on how to tackle this issue?
I found the same issue in my AWS RDS Oracle 18c and 19c instances, even though I am not in the same patch level as you.
In my case, I applied this workaround and it worked.
SQL> alter system set "_fix_control"='20424684:OFF' scope=both;
However, before applying this change, I strongly suggest that you test it on your non production environments, and if you can, try to consult with Oracle Support. Dealing with hidden parameters might lead to unexpected side effects, so apply it at your own risk.
Instead of completely abandoning automatic statistics gathering, try find any specific objects that are causing the problem. If only a small number of tables are responsible for a large amount of statistics gathering, you can manually analyze those tables or change their preferences.
First, use the below SQL to see which objects are causing the most statistics gathering. According to the test case in bug 30846782, the problem seems to be only related to the number of times DBMS_STATS is called.
select *
from dba_optstat_operations
order by start_time desc;
In addition, you may be able to find specific SQL statements or sessions that generate a lot of PGA memory with the below query. (However, if the database restarts, it's possible that AWR won't save the recorded values.)
select username, event, sql_id, pga_allocated/1024/1024/1024 pga_allocated_gb, gv$active_session_history.*
from gv$active_session_history
join dba_users on gv$active_session_history.user_id = dba_users.user_id
where pga_allocated/1024/1024/1024 >= 1
order by sample_time desc;
If the problem is only related to a small number of tables with a large number of partitions, you can manually gather the stats on just that table in a separate session. Once the stats are gathered, the table won't be analyzed again until about 10% of the data is changed.
begin
dbms_stats.gather_table_stats(user, 'PGA_STATS_TEST');
end;
/
It's not uncommon for a database to spend a long time gathering statistics, but it is uncommon for a database to constantly analyze thousands of objects. Running into this bug implies there is something unusual about your database - are you constantly dropping and creating objects, or do you have a large number of objects that have 10% of their data modified every day? You may need to add a manual gather step to a few of your processes.
Turning off the automatic statistics job entirely will eventually cause many performance problems. Even if you can't add manual gathering steps, you may still want to keep the job enabled. For example, if tables are being analyzed too frequently, you may want to increase the table preference for the "STALE_PERCENT" threshold from 10% to 20%:
begin
dbms_stats.set_table_prefs
(
ownname => user,
tabname => 'PGA_STATS_TEST',
pname => 'STALE_PERCENT',
pvalue => '20'
);
end;
/
I’m a longtime MSSQL developer who finds himself back in PL/SQL for the first time since Oracle 7. I’m looking for some tuning advice re a large export stored procedure, which is sporadically and not very reproducably running slow at certain points. This happens around some static working tables which it truncates, fills and uses as part of the export. The code in outline typically looks like this:
create or replace Procedure BigMultiPurposeExport as (
-- about 2000 lines of other code
INSERT WORK_TABLE_5 SELECT WHATEVER1 FROM WHEREVER1;
INSERT WORK_TABLE_5 SELECT WHATEVER2 FROM WHEREVER2;
INSERT WORK_TABLE_5 SELECT WHATEVER3 FROM WHEREVER3;
INSERT WORK_TABLE_5 SELECT WHATEVER4 FROM WHEREVER4;
-- WORK_TABLE_5 now has 0 to ~500k rows whose content can vary drastically from run to run
-- e.g. one hourly run exports 3 whale sightings, next exports all tourist visits to Kenya this decade
-- about 1000 lines of other code
INSERT OUTPUT_TABLE_3
SELECT THIS, THAT, THE_OTHER
FROM BUSINESS_TABLE_1 BT1
INNER JOIN BUSINESS_TABLE_2 ON etc -- typical join on indexed columns
INNER JOIN BUSINESS_TABLE_3 ON etc -- typical join on indexed columns
INNER JOIN BUSINESS_TABLE_4 ON etc -- typical join on indexed columns
LEFT OUTER JOIN WORK_TABLE_1 ON etc -- typical join on indexed columns
LEFT OUTER JOIN WORK_TABLE_2 ON etc -- typical join on indexed columns
LEFT OUTER JOIN WORK_TABLE_3 ON etc -- typical join on indexed columns
LEFT OUTER JOIN WORK_TABLE_4 ON etc -- typical join on indexed columns
LEFT OUTER JOIN WORK_TABLE_5 WT5 ON BT1.ID = WT5.BT1_ID AND WT5.RECORD_TYPE = 21
-- join above is now supported by indexes on BUSINESS_TABLE_1 (ID) and WORK_TABLE_5 (BT1_ID, RECORD_TYPE), originally wasn't
LEFT OUTER JOIN WORK_TABLE_6 ON etc -- typical join on indexed columns
LEFT OUTER JOIN WORK_TABLE_7 ON etc -- typical join on indexed columns
-- about 4000 lines of other code
)
That final insert into OUTPUT_TABLE_3 usually runs in under 10 seconds, but once in a while on certain customer servers it times out at our default 99 minutes. Then we have them take the tiemout off and run it on Friday night, and it finishes but takes 16 hours.
I narrowed the problem down to the join to WORK_TABLE_5, which had no index support, and put an index on the join terms. The next run took 4 seconds. But success has been intermittent, the customer occasionally gets some slow runs when they drastically change their export selection (i.e. drastically change the data in WORK_TABLE_5). And if we update statistics and rebuild indexes after a timed out export, it runs fine at the next attempt.
So, I am wondering about how best to handle truncating/filling static work tables with static indexes, statistics updated overnight, and a stored procedure compiled when the statistics are nothing like runtime.
I have a few general questions about things I'd like to understand better:
Is the nature of the data in the work table going to substantially effect the query plan? Does Oracle form its query plan when you compile the stored procedure? Could we get a highly inappropriate query plan if we compile the stored procedure with the table empty then use a table with 500k rows at runtime?
I expect that if this were an ad-hoc script then updating statistics on the problem table just before selecting from it would eliminate the sporadic slowdowns. But what if I were to update statistics inside the stored procedure, which is compiled with different statistics from runtime?
Anything else you'd like to add...
Thanks for any advice. I hope my MSSQL preconceptions haven't made me too far off base.
This is happening in Oracle 11g, but the code is deployed to assorted customers using Oracle 10 through 12 and I'd like to cater to all of those if possible.
-- Joel
Huge differences in table or index sizes can most definitely cause performance problems. The solution is to add statistics gathering to the procedure instead of relying on the default statistics jobs.
If you've been away from Oracle since version 7, the most important new feature is the Cost Based Optimizer. Oracle now builds query execution plans based on the optimizer statistics of tables, indexes, columns, expressions, system statistics, outlines, directives, dynamic sampling, etc. If you're a full time Oracle developer you should probably spend a day reading about optimizer statistics. Start with Managing Optimizer Statistics and DBMS_STATS in the official documentation.
Eventually the stored procedure should look like this:
--1: Insert into working tables.
insert into work_table...
--2: Gather statistics on working tables.
dbms_stats.gather_table_stats('SCHEMA_NAME', 'WORK_TABLE', ...);
--3: Use working tables.
insert into other_table select * from work_table...
There are so many statistics features it's hard to know exactly what parameters to use in that second step above. Here are some guesses about some features you might find useful:
DEGREE - One reason people avoid gathering statistics inside a process is the time is takes. You can significantly improve the run time by setting the degree. Although this also uses significantly more resources.
NO_INVALIDATE - It can be tricky to know when exactly are the statistics "set" for a query. Gathering statistics usually quickly invalidates execution plans that were based on old statistics. But not always. If you want to be 100% sure that the next query is using the latest statistics you want to set NO_INVALIDATE=>FALSE.
ESTIMATE_PERCENT In 11g and above you definitely want to use the default, which uses a faster algorithm. In 10g and below you may need to set the value to something low to make the gathering fast enough.
Although Oracle 10g and above comes with default statistics gathering jobs you cannot rely on them for a few reasons:
They are scheduled and may not run at the right time. If a process significantly changes the data then new stats are needed right away, not at 10 PM. If there are a lot of tables that need to be analyzed the job may not get to them all in one day.
Many DBAs disable the jobs. This is ridiculous and almost always a mistake. But you'll find many DBAs that disabled the job because they think they can do it better. Instead of working with the auto tasks, and settings preferences, many DBAs like to throw the whole thing out and replace it with a custom procedure that rots over time.
We have one of our system that perform quite a bit of database activity in terms of INSERT/UPDATE/DELETE statements against various tables. Because of this the statistics became stale and this is reflected in overall performance.
We want to create a scheduled job that would periodically invoke DBMS_STATS.GATHER_SCHEMA_STATS. Because we don't want actual stats gathering itself to impact the system processing even more we are thinking to collect statistics quite frequent and use GATHER STALE option:
DBMS_STATS.GATHER_SCHEMA_STATS(OWNNAME => 'MY_SCHEMA', OPTIONS =>'GATHER STALE')
This executes almost instantly but running this statement below before and after stats gathering seems to bring back the same records with the same values:
SELECT * FROM user_tab_modifications WHERE inserts + updates + deletes > 0;
The very short time taking to execute and the fact that user_tab_modifications content stays the same makes me question if OPTIONS =>'GATHER STALE' actually does what we expect it should do. On the other hand if I run this again before and after statistics gathering I can see the tables reported as stale before re no longer reported as stale after:
DECLARE
stale dbms_stats.objecttab;
BEGIN
DBMS_STATS.GATHER_SCHEMA_STATS(ownname => 'MY_SCHEMA', OPTIONS =>'LIST STALE', objlist => stale);
FOR i in 1 .. stale.count
LOOP
dbms_output.put_line( stale(i).objName );
END LOOP;
END;
On another hand if lets say my_table is one of my tables being listed as part of the tables that part of the user_tab_modifications with inserts + updates + deletes > 0 and I run I can see my_table no longer being reported as having changes.
EXECUTE DBMS_STATS.GATHER_TABLE_STATS(ownname => 'MY_SCHEMA', tabname => 'MY_TABLE');
So my questions are:
Is my approach correct. Can I trust I am getting fresh stats just by running options => 'GATHER STALE' or I should manually collect stats for all tables that come back with a reasonable number of inserts, updates, deletes?
When user_tab_modifications would actually get reset; obviously GATHER STALE option does not seem to do it
We are using Oracle Database 11g Enterprise Edition Release 11.2.0.3.0 - 64bit Production
Got the following info from Oracle docs.
You should enable monitoring if you use GATHER_DATABASE_STATS or GATHER_SCHEMA_STATS with the GATHER AUTO or GATHER STALE options.
This view USER_TAB_MODIFICATIONS is populated only for tables with the MONITORING attribute. It is intended for statistics collection over a long period of time. For performance reasons, the Oracle Database does not populate this view immediately when the actual modifications occur. Run the FLUSH_DATABASE_MONITORING_INFO procedure in the DBMS_STATS PL/SQL package to populate this view with the latest information. The ANALYZE_ANY system privilege is required to run this procedure.
Hope this helps you to identify which of your assumptions are incorrect and understand the correct usage of "GATHER STALE".
I'm looking for a best performance solution to detect if data in an Oracle table has changed. This will be used to kick start a calculation that uses lots of data from the same tables. It would be too expensive to poll the data to track changes. The changes happen rarely.
I have analyzed the following solutions to the problem.
ORA_ROWSCN : Too slow, will do a full table scan.
Oracle Audit : Not possible to set up in my environment.
DBMS_ALERT : Writers are not able to signal.
Then I came up with the following simple idea. Add a trigger to the tables that increments a sequence on insert, update or delete. I know this will be materialized if rolled back but I can afford some false positives. My calculations service then polls the sequence current value to detect possible changes (= very cheap query)
CREATE OR REPLACE TRIGGER trg_change_tracker
before insert or update or delete on mytable
declare
dummy number;
begin
select seq_event_seqno.nextval into dummy from dual;
end;
How does this sound? Any pitfalls?
EDIT: Yes there is a major pitfall: When the writer holds back the commit, and the reader sees the new sequence value and query for changes before the writer has committed.
I can suggest you some addition to your trigger soulution:
When your polling see changes of the sequence, you can check opened transactions on interested table, if it exists, then skip current recalc.
UPD:
Also, you can save min SCN of opened transactions, and in case of frequently table changes you will not freeze
UPD2:
It is some heuristic improvement, not full problem solution. If you will skip recal every time when you see opened transaction, then you can freeze much time in case of frequently (and may be long) DML on table.
I mean, when seq is changed and the polling see opened transactions on your table, you can store min(start_scn) from v$transaction, and when the polling see opened transactions at next time you can compare current min(start_scn) with sotred min(start_scn), if current is greater then it is some chance that it is time to recalc.
You can use ( for particular XTABLE ):
SELECT * FROM dba_tab_modifications WHERE TABLE_NAME = 'XTABLE';
it will allow you ( among other data ) to see:
INSERTS UPDATES DELETES
43,708 1,845 0
.. there are issues ( bugs in some Oracle versions ), but if you will test thoroughly in your environment - you may find this useful.
.. as this question is pretty old - it would be nice to know how you implemented the change detection mechanism.
I am developing a DWH on Oracle 11g. We have some big tables (250+ million rows), partitioned by value. Each partition is a assigned to a different feeding source, and every partition is independent from others, so they can be loaded and processed concurrently.
Data distribution is very uneven, we have partition with millions rows, and partitions with not more than a hundred rows, but I didn't choose the partitioning scheme, and by the way I can't change it.
Considered the data volume, we must assure that every partition has always up-to-date statistics, because if the subsequent elaborations don't have an optimal access to the data, they will last forever.
So for each concurrent ETL thread, we
Truncate the partition
Load data from staging area with
SELECT /*+ APPEND */ INTO big_table PARTITION(part1) FROM temp_table WHERE partition_colum = PART1
(this way we have direct path and we don't lock the whole table)
We gather statistics for the modified partition.
In the first stage of the project, we used the APPROX_GLOBAL_AND_PARTITION strategy and worked like a charm
dbms_stats.gather_table_stats(ownname=>myschema,
tabname=>big_table,
partname=>part1,
estimate_percent=>1,
granularity=>'APPROX_GLOBAL_AND_PARTITION',
CASCADE=>dbms_stats.auto_cascade,
degree=>dbms_stats.auto_degree)
But, we had the drawback that, when we loaded a small partition, the APPROX_GLOBAL part was dominant (still a lot faster than GLOBAL) , and for a small partition we had, e.g., 10 seconds of loading, and 20 minutes of statistics.
So we have been suggested to switch to the INCREMENTAL STATS feature of 11g, which means that you don't specify the partition you have modified, you leave all parameters in auto, and Oracle does it's magic, automatically understanding which partition(s) have been touched. And it actually works, we have really speeded up the small partition. After turning on the feature, the call became
dbms_stats.gather_table_stats(ownname=>myschema,
tabname=>big_table,
estimate_percent=>dbms_stats.auto_sample_size,
granularity=>'AUTO',
CASCADE=>dbms_stats.auto_cascade,
degree=>dbms_stats.auto_degree)
notice, that you don't pass the partition anymore, and you don't specify a sample percent.
But, we're having a drawback, maybe even worse that the previous one, and this is correlated with the high level of parallelism we have.
Let's say we have 2 big partition that starts at the same time, they will finish the load phase almost at the same time too.
The first thread ends the insert statement, commits, and launches the stats gathering. The stats procedure notices there are 2 partition modified (this is correct, one is full and the second is truncated, with a transaction in progress), updates correctly the stats for both the partitions.
Eventually the second partition ends, gather the stats, it see all partition already updated, and does nothing (this is NOT correct, because the second thread committed the data in the meanwhile).
The result is:
PARTITION NAME | LAST ANALYZED | NUM ROWS | BLOCKS | SAMPLE SIZE
-----------------------------------------------------------------------
PART1 | 04-MAR-2015 15:40:42 | 805731 | 20314 | 805731
PART2 | 04-MAR-2015 15:41:48 | 0 | 16234 | (null)
and the consequence is that I occasionally incur in not optimal plans (which mean killing the session, refresh manually the stats, manually launch the precess again).
I tried even putting an exclusive lock on the gathering, so no more than one thread can gather stats on the same table at once, but nothing changed.
IMHO this is an odd behaviour, because the stats procedure, the second time it is invoked, should check for the last commit on the second partition, and should see it's newer than the last stats gathering time. But seems it's not happening.
Am I doing something wrong? Is it an Oracle bug? How can I guarantee that all stats are always up-to-date with incremental stats feature turned on, and an high level of concurrency?
I managed to reach a decent compromise with this function.
PROCEDURE gather_tb_partiz(
p_tblname IN VARCHAR2,
p_partname IN VARCHAR2)
IS
v_stale all_tab_statistics.stale_stats%TYPE;
BEGIN
BEGIN
SELECT stale_stats
INTO v_stale
FROM user_tab_statistics
WHERE table_name = p_tblname
AND object_type = 'TABLE';
EXCEPTION
WHEN NO_DATA_FOUND THEN
v_stale := 'YES';
END;
IF v_stale = 'YES' THEN
dbms_stats.gather_table_stats(ownname=>myschema,
tabname=> p_tblname,
partname=>p_partname,
degree=>dbms_stats.auto_degree,
granularity=>'APPROX_GLOBAL AND PARTITION') ;
ELSE
dbms_stats.gather_table_stats(ownname=>myschema,
tabname=>p_tblname,
partname=>p_partname,
degree=>dbms_stats.auto_degree,
granularity=>'PARTITION') ;
END IF;
END gather_tb_partiz;
At the end of each ETL, if the number of added/deleted/modified rows is low enough not to mark the table as stale (10% by default, can be tuned with STALE_PERCENT parameter), I collect only partition statistics; otherwise i collect global and partition statistics.
This keeps ETL of small partition fast, because no global partition must be regathered, and big partition safe, because any subsequent query will have fresh statistics and will likely use an optimal plan.
Incremental stats is anyway enabled, so whenever the global has to be recalculated, it is pretty fast because aggregates partition level statistics and does not perform a full scan.
I am not sure if, with incremental enabled, "APPROX_GLOBAL AND PARTITION" and "GLOBAL AND PARTITION" do differ in something, because both incremental and approx do basically the same thing: aggregate stats and histograms without doing a full scan.
Have you tried to have incremental statistics on, but still explicitly name a partition to analyze?
dbms_stats.gather_table_stats(ownname=>myschema,
tabname=>big_table,
partname=>part,
degree=>dbms_stats.auto_degree);
For your table, stale (yesterday's) global stats are not as harmful as completely invalid partition stats (0 rows). I can propose 2 a bit alternative approaches that we use:
Have a separate GLOBAL stats gathering executed by your ETL tool right after all partitions are loaded. If it's taking too long, play with estimate_percent as dbms_stats.auto_degree will likely to be more than 1%
Gather the global (as well as all other stale) stats in a separate database job run later during the day, after all data is loaded into DW.
The key point is that stale statistics which differ only slightly from fresh are almost just as good. If statistics show you 0 rows, they'll kill any query.
Considering what you are trying to achieve, you need to run stats on specific intervals of time for all Partitions and not at the end of the process that loads each partition. It could be challenging if this is a live table and has constant data loads happening round the clock, but since these are LARGE DW tables I really doubt that's the case. So the best bet would be to collect stats at the end of loading all partitions, this will ensure that the statistics is collected for partitions where data has change or statistics are missing and update the global statistics based on the partition level statistics and synopsis.
However to do so, you need to turn on incremental feature for the table (11gR1).
EXEC DBMS_STATS.SET_TABLE_PREFS('<Owner>','BIG_TABLE','INCREMENTAL','TRUE');
At the end of every load, gather table statistics using GATHER_TABLE_STATS command. You don't need to specify the partition name. Also, do not specify the granularity parameter.
EXEC DBMS_STATS.GATHER_TABLE_STATS('<Owner>','BIG_TABLE');
Kindly check if you have used DBMS_STATS to set table preference to gather incremental statistics.This oracle blog explains that statistics will be gathered after each row affected.
Incremental statistics maintenance needs to gather statistics on any partition that will change the global or table level statistics. For instance, the min or max value for a column could change after just one row is inserted or updated in the table
BEGIN
DBMS_STATS.SET_TABLE_PREFS(myschema,'BIG_TABLE','INCREMENTAL','TRUE');
END;
I'm a bit rusty about it, so first of all a question:
did you try serializing partition loading? If so, how long and how well does statistics run? Notice that since loading time is so much smaller than statistics gathering, i guess this could also act as a temporary workaround.
Append hint does affects redo size, meaning the transaction just traces something, thus statistics may not reckon new data:
http://oracle-base.com/articles/misc/append-hint.php
Thinking out loud: since the direct path insert does append rows at the end of the partition and eventually updates metadata at the end, the already running thread gathering statistics could have read non-updated (stale) data. Thus it may not be a bug, and locking threads would accomplish nothing.
You may test this behaviour temporarily switching your table/partition to LOGGING, for instance, and see how it works (slower, of course, but it's a test). Can you do it?
EDIT: incremental stats should work anyway, even disabling a parallel statistics gathering, since it reiles on the incremental values no matter how they were collected:
https://blogs.oracle.com/optimizer/entry/incremental_statistics_maintenance_what_statistics