Oracle DBA – A lifelong learning experience

Archive for the ‘Oracle’ Category

A change of career after 29 years working with Oracle databases

Posted by John Hallas on July 1, 2019

I have worked with Oracle since the early 90’s. I did work on V6 but I do not remember that much about it. V7.3 was were I had a big learning curve. This book was the mainstay of my life at that time.

In recent years I have diversified into managing teams across a number of areas including SQLServer, Middleware (ODI, SOA, JMS, Weblogic), System Management and others. I have been quite involved in both Oracle and Microsoft licensing which leads me neatly onto my new role.

I have now moved to being the Software Asset Manager for my company and I am also taking the opportunity to move to a 3 day week. For someone who started work at 16 and has never had any time out to suddenly have 2 days a week free is quite momentous.

I am really enthused by the new opportunity and will definitely involve myself as deeply as possible in trying to understand the nuances of managing software across a large organisation.

One of the nicer aspects is that I will no longer be on-call or expected to be contactable in the event of production issues. I have probably been on call for at least 20 of the 29 years – possibly more

I have not updated this blog that much in the last 6 months or so and I think new posts will be infrequent going forward but I will still be working with the DBA team so if I hear of any interesting activities I will post them up.

 

Advertisements

Posted in Oracle, SAM | 3 Comments »

Oracle car parking

Posted by John Hallas on June 17, 2019

I have been involved over the years with both Microsoft SQL Server licensing and Oracle licensing management and the following image made me chuckle.

Produced by House of Brick who have a lot of interesting content about license management

 

 

 

 

Posted in Licensing, Oracle | Leave a Comment »

Oracle Cloud Total Cost of ownership Calculator – rather confusing

Posted by John Hallas on August 1, 2018

I was interested enough to try out the Oracle  TCO calculator that I was sent an email about. I can honestly say it has rather confused me in what it is comparing and how the results are calculated.

I will share a few screen shots and I will take you through my thoughts

I put a dummy company name in and my country – bizarrely it does not auto select the currency.

I then hit enter and waited and waited as I had to with every screenshot I show. I must admit that  trying this a few times performance was sometimes acceptable but when it was slow it was very slow. I then get the fantastic news that moving to the Cloud could save me £403K over 3 years. I had my credit card out and was ready to sign-up.

The basics of my comparison were shown assuming that I had was using 12Tb of storage and 24 cores. I then looked at the detailed savings

Firstly we have gone from using 12Tb storage to only 4Tb, although interestingly the 4Tb in the Cloud costs not far short of the on-premise 12Tb. We have also reduced from 24 cores to only 8 cores. Perhaps Oracle could just come on-site and sprinkle the magic dust that shrinks data and reduces processing requirements and we wouldn’t need to move to the Cloud.

Each of the lines on the diagram can be viewed in more detail and adjustments made. The network component shows that there is zero cost for running in the Cloud but I assume that data will have to be loaded into the database via some form of network load.

Overall as a calculator it is well laid out but I found the like for like comparisons to be rather not like for like.

 

The link for anyone to try it is https://valuenavigator.oracle.com/benefitcalculator

 

Posted in Oracle | Leave a Comment »

12C Grid – GENO process causes high CPU usage

Posted by John Hallas on July 11, 2018

This is an interesting bug which has affected a number of our databases  – perhaps because we are not patched to the latest levels in every case.
The MoS note is this one – Patch 20373598: GEN0 TIMEOUT ACTION ALWAYS TRY TO INITIALIZE OCR CONTENT    p20373598_121022_Linux-x86-64
OR
fixed in Bug 22291127  12.1.0.2.160419 (Apr 2016) Database Patch Set Update (DB PSU)
Here is view of top showing how a server looks when it is affected, and a list of servers running 12c grid home below the Apr 2016 PSU patch level.
If the server has loads of cores assigned the overall effect will be small as seen here, CPU shows as 2.8% used
Tasks: 618 total,   3 running, 614 sleeping,   0 stopped,   1 zombie
Cpu(s):  2.8%us,  0.2%sy,  0.0%ni, 96.8%id,  0.2%wa,  0.0%hi,  0.0%si,  0.0%st
Mem:  198475784k total, 80283568k used, 118192216k free,  4491100k buffers
Swap: 41680876k total,        0k used, 41680876k free, 56391976k cached

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
44361 oracle    20   0 4225m 2.8g  18m R 99.6  1.5 753924:28 asm_gen0_+ASM
24347 root      20   0  448m 5536 1348 S  4.9  0.0   7480:21 /usr/sbin/collectd -C /etc/collectd.conf -f
68317 oracle    20   0 50.3g 182m 166m S  3.3  0.1   3052:05 ora_pr00_OEMPRD2A
66023 oracle    -2   0 50.3g  19m  17m S  2.3  0.0   1113:44 ora_vktm_OEMPRD2A
34357 oracle    -2   0 1405m  17m  15m S  2.0  0.0  17424:26 asm_vktm_+ASM
67399 oracle    20   0 50.4g  99m  38m S  2.0  0.1   1885:56 oracleOEMPRD2A (LOCAL=NO)

This SQL is a useful way to identify which servers require patching assuming you are using OEM

select
host_name,
home_name,
home_location
patch_id,
install_timestamp,
description</span>
from sysman."MGMT$SOFTWARE_ONEOFF_PATCHES"
where
host_name not in (
select host_name 
from sysman."MGMT$SOFTWARE_ONEOFF_PATCHES"
where patch_id in ('22291127','20373598'))
and home_location like '%12%grid'
and description like '%Database Patch Set Update%'
order by 1,4
;

Posted in Oracle | Tagged: , | Leave a Comment »

Creating a database link using a tnsnames entry

Posted by John Hallas on December 29, 2017

I have had this in my draft posts for a couple of weeks and I refrained from publishing because it was too simple.

However the main purpose of this post is to help others save time by providing useful information. The secondary purpose of to remind myself how I overcame whatever problem is being covered.

A user did not have permissions to edit the tnsnames.ora file on a server so I created the database link for him to use the connect description, thereby circumventing the need for an update to the tnsnames.ora file.

I prefer database links to be named to show the database they connect to, however in this case they wanted a standard name which is used in code and can be used across all the environments.

 create database link ebsdev connect to remote_username  identified by password using
‘EBSDEV =
(DESCRIPTION =
(ADDRESS = (PROTOCOL = TCP)(HOST = servername)(PORT = 1531))
(CONNECT_DATA =
(SERVER = DEDICATED)
(SERVICE_NAME = EBSDEV)
)
)’;
select ‘x’ from dual@ebsdev;

Posted in Oracle | Leave a Comment »

ORA-01450: maximum key length (3800) exceeded

Posted by John Hallas on December 13, 2017

This is an oddity as essentially I have an index on a table but I cannot rebuild it because it suggests the block size should be bigger than 8K – however both the database and tablespace block size are 16K already.

The maximum allowed index key length depends on your block size. So the minimum allowed size reported in ORA-01450 varies, depending on which block size your index is using:

ORA-01450 maximum key length (758) exceeded -> (2K Block)

ORA-01450 maximum key length (1578) exceeded -> (4K block)

ORA-01450 maximum key length (3218) exceeded -> (8K Block)

ORA-01450 maximum key length (6498) exceeded -> (16K Block)
See MOSC Note 136158.1 and MOSC Note 236329.1 for more details on the ORA-01450 error and key length.

Even more odd was that I could drop the index and recreate it with no issues and yet rebuild failed every time, even after the recreate. Studying the table and data there did not seem to be a value that was much bigger than the others and certainly nothing that would cause one or more rows to be much larger than the others. Read the rest of this entry »

Posted in Oracle | Tagged: | 3 Comments »

How to add a metric extension and use a ‘control’ table to test results

Posted by John Hallas on November 27, 2017

I have previously blogged about creating a User Defined Metric (UDM) before  (7 years ago!)  but the method I am using today demonstrates how you can test out the alert using specific data.

The aim was to see if integration data is flowing through a table.  If there is a backlog we want to know about it.

Simon Ryan, a colleague did all the preparatory work and created all the 18 !! screen shots which I had to copy and paste into this entry

select * from sm7.EVENTOUTM1; -- Transient table so data not usually present.

The data is processed quickly but occasionally you can see a handful of rows for a few seconds.

select DESCRIPTOR ,
EVTYPE ,
TO_CHAR(EVTIME, 'DD-MON-YYYY HH24:MI:SS') ,
EVSYSSEQ ,
EVUSRSEQ from sm7.EVENTOUTM1;
DESCRIPTOR EVTYPE EVTIME EVSYSSEQ EVUSRSEQ
 (BLOB) email 08-NOV-2017 11:40:01 0000000005254328

A typical query for the ME would be very simple:

select count(*) from sm7.EVENTOUTM1;

However, we are going to plugin a ‘control’ table in order for us to test the alert by artificially adjusting the results. Read the rest of this entry »

Posted in Oracle | Tagged: , , , | Leave a Comment »

DBA vacancy Bradford, West Yorkshire

Posted by John Hallas on November 24, 2017

I have a vacancy in my team and the details are in the link below

https://apply.morrisons.jobs/vacancies/2137/technology-specialist–database.html

Lots of interesting work going including 85Tb Data Warehouse on Exadata and developing Cloud capability using AWS and Google.

 

Posted in Oracle | 3 Comments »

Baselines – session creating privs v session running privs

Posted by John Hallas on November 21, 2017

A colleague Richard Wilkinson was telling me about an issue he had come across with baselines and I asked him to write it up as it was an interesting experience.

The following MERGE SQL runs once a day. It always uses the same plan and roughly takes between 30 and 80 minutes:

SQL_ID/ PHV 6zs5dk6t6pkfs / 3064471754

 MERGE INTO DWB_X_PLNGRM_ATTR_DAY t
 USING (
 SELECT ilv.bsns_unit_cd,
 ilv.sku_item_nbr,
 ilv.cntnt_cd ,
 ilv.sel_units_cntnt_status,
 ilv.plngrm_item_grp_cd,
 dwr.bay_cnt_key,
 dwr.bay_cnt
 FROM (
 SELECT DISTINCT bsns_unit_cd,
 sku_item_nbr,
 cntnt_cd ,
 sel_units_cntnt_status,
 plngrm_item_grp_cd,
 --bay_cnt_key,
 max(bay_cnt_key) bay_cnt_key
 --bay_cnt
 FROM dwb_x_sel_item_units_day
 WHERE (1=1)
 AND day_key >= :v_daywkey
 GROUP BY bsns_unit_cd, sku_item_nbr, cntnt_cd, sel_units_cntnt_status, plngrm_item_grp_cd
 ) ilv,
 dwr_x_plngrm_loc_bay_cnt dwr
 WHERE ilv.bay_cnt_key = dwr.bay_cnt_key
 ) s
 ON (
 t.CNTNT_CD = s.CNTNT_CD
 AND t.SKU_ITEM_NBR = s.SKU_ITEM_NBR
 AND t.BSNS_UNIT_CD = s.BSNS_UNIT_CD
 AND t.PLNGRM_CNTNT_STATUS = s.SEL_UNITS_CNTNT_STATUS
 AND t.PLNGRM_ITEM_GRP_CD = s.PLNGRM_ITEM_GRP_CD
 --and t.DAY_KEY >= :v_daywkey
 )
 WHEN MATCHED
 THEN UPDATE SET
 t.bay_cnt_key = s.bay_cnt_key,
 t.bay_cnt = s.bay_cnt;
Plan hash value: 3064471754 
 
---------------------------------------------------------------------------------- 
| Id | Operation | Name | 
---------------------------------------------------------------------------------- 
| 0 | MERGE STATEMENT | | 
| 1 | PX COORDINATOR | | 
| 2 | PX SEND QC (RANDOM) | :TQ10007 | 
| 3 | INDEX MAINTENANCE | DWB_X_PLNGRM_ATTR_DAY | 
| 4 | PX RECEIVE | | 
| 5 | PX SEND RANGE | :TQ10006 | 
| 6 | MERGE | DWB_X_PLNGRM_ATTR_DAY | 
| 7 | PX RECEIVE | | 
| 8 | PX SEND HYBRID (ROWID PKEY) | :TQ10005 | 
| 9 | VIEW | | 
| 10 | HASH JOIN BUFFERED | | 
| 11 | PX RECEIVE | | 
| 12 | PX SEND HASH | :TQ10003 | 
| 13 | HASH JOIN BUFFERED | | 
| 14 | PX RECEIVE | | 
| 15 | PX SEND HASH | :TQ10001 | 
| 16 | PX BLOCK ITERATOR | | 
| 17 | TABLE ACCESS STORAGE FULL | DWR_X_PLNGRM_LOC_BAY_CNT | 
| 18 | PX RECEIVE | | 
| 19 | PX SEND HASH | :TQ10002 | 
| 20 | VIEW | | 
| 21 | SORT GROUP BY | | 
| 22 | PX RECEIVE | | 
| 23 | PX SEND HASH | :TQ10000 | 
| 24 | PX BLOCK ITERATOR | | 
| 25 | TABLE ACCESS STORAGE FULL| DWB_X_SEL_ITEM_UNITS_DAY | 
| 26 | PX RECEIVE | | 
| 27 | PX SEND HASH | :TQ10004 | 
| 28 | PX BLOCK ITERATOR | | 
| 29 | TABLE ACCESS STORAGE FULL | DWB_X_PLNGRM_ATTR_DAY | 
----------------------------------------------------------------------------------

One day this sql changes execution plan and runs for over 5 hours before being killed.

The new plan is 4185795484:

----------------------------------------------------------------------------- 
| Id | Operation | Name | 
----------------------------------------------------------------------------- 
| 0 | MERGE STATEMENT | | 
| 1 | MERGE | DWB_X_PLNGRM_ATTR_DAY | 
| 2 | PX COORDINATOR | | 
| 3 | PX SEND QC (RANDOM) | :TQ10005 | 
| 4 | VIEW | | 
| 5 | HASH JOIN BUFFERED | | 
| 6 | PX RECEIVE | | 
| 7 | PX SEND HASH | :TQ10003 | 
| 8 | PX BLOCK ITERATOR | | 
| 9 | TABLE ACCESS STORAGE FULL | DWR_X_PLNGRM_LOC_BAY_CNT | 
| 10 | PX RECEIVE | | 
| 11 | PX SEND HASH | :TQ10004 | 
| 12 | HASH JOIN BUFFERED | | 
| 13 | PX RECEIVE | | 
| 14 | PX SEND HASH | :TQ10001 | 
| 15 | VIEW | | 
| 16 | SORT GROUP BY | | 
| 17 | PX RECEIVE | | 
| 18 | PX SEND HASH | :TQ10000 | 
| 19 | PX BLOCK ITERATOR | | 
| 20 | TABLE ACCESS STORAGE FULL| DWB_X_SEL_ITEM_UNITS_DAY | 
| 21 | PX RECEIVE | | 
| 22 | PX SEND HASH | :TQ10002 | 
| 23 | PX BLOCK ITERATOR | | 
| 24 | TABLE ACCESS STORAGE FULL | DWB_X_PLNGRM_ATTR_DAY | 
-----------------------------------------------------------------------------

The main difference I can see is that the tables are being joined in a different order.

3064471754 (good): DWB_X_SEL_ITEM_UNITS_DAY joined to DWR_X_PLNGRM_LOC_BAY_CNT then on to the target of the MERGE, DWB_X_PLNGRM_ATTR_DAY.
4185795484 (bad): DWB_X_SEL_ITEM_UNITS_DAY joined to DWB_X_PLNGRM_ATTR_DAY then to DWR_X_PLNGRM_LOC_BAY_CNT.

The fix is simple, fix the good plan that has always been used (3064471754) using SQL baselines.

It looks like this statement already has a few couple of baselined plans , neither are fixed.

sql handle: SQL_0e4f45d7832dd5ab
SQL_PLAN_0wmu5uy1kvpdb8c1d711e 10-NOV-17 10.38.07.000000000 MANUAL-LOAD
SQL_PLAN_0wmu5uy1kvpdb1eaf5716 29-SEP-17 23.16.31.000000000 AUTO-CAPTURE

Looking at the detailed view of the baseline shows that one of the plans is the bad plan and the other has no history of use in ASH:

select * 
 from table(dbms_xplan.display_sql_plan_baseline(sql_handle => 'SQL_0e4f45d7832dd5ab',
 format => 'basic')); --detailed view


PLAN_TABLE_OUTPUT 
----------------------------------------------------------------------------- 
SQL handle: SQL_0e4f45d7832dd5ab 
SQL text: MERGE INTO DWB_X_PLNGRM_ATTR_DAY T USING ( SELECT ILV.BSNS_UNIT_CD, 
ILV.SKU_ITEM_NBR, ILV.CNTNT_CD , ILV.SEL_UNITS_CNTNT_STATUS, 
ILV.PLNGRM_ITEM_GRP_CD, DWR.BAY_CNT_KEY, DWR.BAY_CNT FROM ( SELECT 
DISTINCT BSNS_UNIT_CD, SKU_ITEM_NBR, CNTNT_CD , SEL_UNITS_CNTNT_STATUS, 
PLNGRM_ITEM_GRP_CD, MAX(BAY_CNT_KEY) BAY_CNT_KEY FROM 
DWB_X_SEL_ITEM_UNITS_DAY WHERE (1=1) AND DAY_KEY >= :B1 GROUP BY 
BSNS_UNIT_CD, SKU_ITEM_NBR, CNTNT_CD, SEL_UNITS_CNTNT_STATUS, 
PLNGRM_ITEM_GRP_CD ) ILV, DWR_X_PLNGRM_LOC_BAY_CNT DWR WHERE 
ILV.BAY_CNT_KEY = DWR.BAY_CNT_KEY ) S ON ( T.CNTNT_CD = S.CNTNT_CD AND 
T.SKU_ITEM_NBR = S.SKU_ITEM_NBR AND T.BSNS_UNIT_CD = S.BSNS_UNIT_CD AND 
T.PLNGRM_CNTNT_STATUS = S.SEL_UNITS_CNTNT_STATUS AND 
T.PLNGRM_ITEM_GRP_CD = S.PLNGRM_ITEM_GRP_CD ) WHEN MATCHED THEN UPDATE 
SET T.BAY_CNT_KEY = S.BAY_CNT_KEY, T.BAY_CNT = S.BAY_CNT 
-------------------------------------------------------------------------------- 
 
-------------------------------------------------------------------------------- 
Plan name: SQL_PLAN_0wmu5uy1kvpdb1eaf5716 Plan id: 514807574 
Enabled: YES Fixed: NO Accepted: NO Origin: AUTO-CAPTURE 
-------------------------------------------------------------------------------- 
 
Plan hash value: 4185795484 /******BAD PLAN*******/ 
 
----------------------------------------------------------------------------- 
| Id | Operation | Name | 
----------------------------------------------------------------------------- 
| 0 | MERGE STATEMENT | | 
| 1 | MERGE | DWB_X_PLNGRM_ATTR_DAY | 
| 2 | PX COORDINATOR | | 
| 3 | PX SEND QC (RANDOM) | :TQ10005 | 
| 4 | VIEW | | 
| 5 | HASH JOIN BUFFERED | | 
| 6 | PX RECEIVE | | 
| 7 | PX SEND HASH | :TQ10003 | 
| 8 | PX BLOCK ITERATOR | | 
| 9 | TABLE ACCESS STORAGE FULL | DWR_X_PLNGRM_LOC_BAY_CNT | 
| 10 | PX RECEIVE | | 
| 11 | PX SEND HASH | :TQ10004 | 
| 12 | HASH JOIN BUFFERED | | 
| 13 | PX RECEIVE | | 
| 14 | PX SEND HASH | :TQ10001 | 
| 15 | VIEW | | 
| 16 | SORT GROUP BY | | 
| 17 | PX RECEIVE | | 
| 18 | PX SEND HASH | :TQ10000 | 
| 19 | PX BLOCK ITERATOR | | 
| 20 | TABLE ACCESS STORAGE FULL| DWB_X_SEL_ITEM_UNITS_DAY | 
| 21 | PX RECEIVE | | 
| 22 | PX SEND HASH | :TQ10002 | 
| 23 | PX BLOCK ITERATOR | | 
| 24 | TABLE ACCESS STORAGE FULL | DWB_X_PLNGRM_ATTR_DAY | 
----------------------------------------------------------------------------- 
 
-------------------------------------------------------------------------------- 
Plan name: SQL_PLAN_0wmu5uy1kvpdb8c1d711e Plan id: 2350739742 
Enabled: YES Fixed: NO Accepted: YES Origin: MANUAL-LOAD 
-------------------------------------------------------------------------------- 
 
Plan hash value: 1932271691 /**********no record of use in ASH*********/ 
 
----------------------------------------------------------------------------- 
| Id | Operation | Name | 
----------------------------------------------------------------------------- 
| 0 | MERGE STATEMENT | | 
| 1 | MERGE | DWB_X_PLNGRM_ATTR_DAY | 
| 2 | PX COORDINATOR | | 
| 3 | PX SEND QC (RANDOM) | :TQ10005 | 
| 4 | VIEW | | 
| 5 | HASH JOIN BUFFERED | | 
| 6 | PX RECEIVE | | 
| 7 | PX SEND HASH | :TQ10003 | 
| 8 | HASH JOIN BUFFERED | | 
| 9 | PX RECEIVE | | 
| 10 | PX SEND HASH | :TQ10001 | 
| 11 | PX BLOCK ITERATOR | | 
| 12 | TABLE ACCESS STORAGE FULL | DWR_X_PLNGRM_LOC_BAY_CNT | 
| 13 | PX RECEIVE | | 
| 14 | PX SEND HASH | :TQ10002 | 
| 15 | VIEW | | 
| 16 | HASH GROUP BY | | 
| 17 | PX RECEIVE | | 
| 18 | PX SEND HASH | :TQ10000 | 
| 19 | PX BLOCK ITERATOR | | 
| 20 | TABLE ACCESS STORAGE FULL| DWB_X_SEL_ITEM_UNITS_DAY | 
| 21 | PX RECEIVE | | 
| 22 | PX SEND HASH | :TQ10004 | 
| 23 | PX BLOCK ITERATOR | | 
| 24 | TABLE ACCESS STORAGE FULL | DWB_X_PLNGRM_ATTR_DAY | 
-----------------------------------------------------------------------------

Plan 1932271691 joins the table in the same order as 3064471754 and looks logically the same, so I think this is preferable to the bad plan (which is also in the baseline).
I’m not sure why the optimizer isn’t choosing plan 1932271691.

I created an SQL Tuning Set from a snapshot range containing yesterday’s good run and confirmed that the only plan in there was 3064471754.

I uploaded the STS to the SQL baseline, however it does not add a new plan, it still shows:

sql handle: SQL_0e4f45d7832dd5ab
SQL_PLAN_0wmu5uy1kvpdb8c1d711e
SQL_PLAN_0wmu5uy1kvpdb1eaf5716

The only difference is that the timestamp of SQL_PLAN_0wmu5uy1kvpdb8c1d711e is now current so it looks like it’s been updated.

When querying the baseline in detail I expect to see SQL_PLAN_0wmu5uy1kvpdb8c1d711e now contains 3064471754, however it doesn’t.
It still shows 1932271691, nothing has changed except the timestamp, it looks like we can’t fix the good plan that has been used every day for months.

I fixed the potentially good plan and flushed the bad plan from all instances, however upon retrying the MERGE the bad plan was still chosen.

As an experiment I ran an explain plan of the MERGE in SQL Developer replacing the bind variable with the literal that was being passed.
The explain plan returns the good plan (same as 3064471754).

As a workaround I ask the team to run the MERGE from SQL Developer with the literal (after confirming their explain plan also returns the good plan).

Upon monitoring the SQL Developer session it is confirmed that it is using the good plan with the exact same plan ID (3064471754) and completes in 1 hour 15 minutes.

As it is Friday afternoon we agree on this workaround for the weekend (support team is 24 hours / shifts).

The next day the batch job MERGE has picked a different bad plan.
The support team kill it and run the workaround.

The workaround then stops working well as it has picked the same bad plan as the batch job.
The SQL Developer MERGE runs for 5 hours before being killed.

The new bad plan is 1501257745.

1501257745 looks logically the same as the good plan (3064471754) and the potentially good plan in the baseline (1932271691) as it has the same join order.

However we now have a breakthrough – the new bad plan has now given us a new angle to look at.

1501257745 starts by spawning many parallel processes, however it soon drops off to one serial process.
This one serial process is performing the top-level MERGE operation which flatlines for hours before being killed.

I decide to retry fixing the original good plan.
I start with a clean slate and drop the SQL baseline and both plans contained within.
I create a new STS from a different snapshot period than before which only contains 3064471754.
I load the STS into the baseline and a new baseline and plan are created.

When querying the contents of the baseline it still just contains the potentially good plan from before (1932271691).

Oracle is now deliberately and consistently changing my plan in transit from 3064471754 to 1932271691.

With the new insight given by the new bad plan (1501257745) I now look at the top-level DML parts of the plans.

Logically the potentially good plan (1932271691) and the new bad plan (1501257745) are the same statement as 3064471754 but without the parallel MERGE.

Plan hash value: 1932271691 and 1501257745 
----------------------------------------------------------------------------- 
 | Id | Operation | Name | 
----------------------------------------------------------------------------- 
 | 0 | MERGE STATEMENT | | 
 | 1 | MERGE | DWB_X_PLNGRM_ATTR_DAY |


Plan hash value: 3064471754 
---------------------------------------------------------------------------------- 
| Id | Operation | Name | 
 ---------------------------------------------------------------------------------- 
| 0 | MERGE STATEMENT | | 
| 1 | PX COORDINATOR | | 
| 2 | PX SEND QC (RANDOM) | :TQ10007 | 
| 3 | INDEX MAINTENANCE | DWB_X_PLNGRM_ATTR_DAY | 
| 4 | PX RECEIVE | | 
| 5 | PX SEND RANGE | :TQ10006 | 
| 6 | MERGE | DWB_X_PLNGRM_ATTR_DAY |

Now we are concentrating on the parallelism of DML, I remember working at a site previously and having to set ENABLE PDML at the session level.

I ran an explain plan of the MERGE before and after issuing the following statement:

ALTER SESSION ENABLE PARALLEL DML;

Afterwards the plan changed to include parallel steps for the MERGE.

I told the support team to do this in their SQL Developer session and the MERGE began running with the usual good plan and even had the same plan id (3064471754).

Now that PDML is enabled in my session, I reloaded the same STS as before into the same baseline.

Once again the only high-level change was the changing of the timestamp on the plan, however when querying the detail of the plan it no longer contained 1932271691 but a new plan, 2197295520.

Plan hash value: 2197295520 
 
---------------------------------------------------------------------------------- 
| Id | Operation | Name | 
---------------------------------------------------------------------------------- 
| 0 | MERGE STATEMENT | | 
| 1 | PX COORDINATOR | | 
| 2 | PX SEND QC (RANDOM) | :TQ10007 | 
| 3 | INDEX MAINTENANCE | DWB_X_PLNGRM_ATTR_DAY | 
| 4 | PX RECEIVE | | 
| 5 | PX SEND RANGE | :TQ10006 | 
| 6 | MERGE | DWB_X_PLNGRM_ATTR_DAY | 
| 7 | PX RECEIVE | | 
| 8 | PX SEND HYBRID (ROWID PKEY) | :TQ10005 | 
| 9 | VIEW | | 
| 10 | HASH JOIN BUFFERED | | 
| 11 | PX RECEIVE | | 
| 12 | PX SEND HASH | :TQ10003 | 
| 13 | HASH JOIN BUFFERED | | 
| 14 | PX RECEIVE | | 
| 15 | PX SEND HASH | :TQ10001 | 
| 16 | PX BLOCK ITERATOR | | 
| 17 | TABLE ACCESS STORAGE FULL | DWR_X_PLNGRM_LOC_BAY_CNT | 
| 18 | PX RECEIVE | | 
| 19 | PX SEND HASH | :TQ10002 | 
| 20 | VIEW | | 
| 21 | HASH GROUP BY | | /****only difference from original good plan (which has SORT GROUP BY instead)****/ 
| 22 | PX RECEIVE | | 
| 23 | PX SEND HASH | :TQ10000 | 
| 24 | PX BLOCK ITERATOR | | 
| 25 | TABLE ACCESS STORAGE FULL| DWB_X_SEL_ITEM_UNITS_DAY | 
| 26 | PX RECEIVE | | 
| 27 | PX SEND HASH | :TQ10004 | 
| 28 | PX BLOCK ITERATOR | | 
| 29 | TABLE ACCESS STORAGE FULL | DWB_X_PLNGRM_ATTR_DAY | 
----------------------------------------------------------------------------------

The tables are joined in the correct order and the MERGE is parallel.
The only difference is line 21 which has HASH GROUP BY instead of SORT GROUP BY.

I fixed this plan hoping that the batch version of the MERGE will pick it up tomorrow.
It doesn’t, however it now runs with the original good plan every day (3064471754).

Whilst I’m still looking into this a key point to take away is that when using SQL baselines the environment of the session creating the baseline and the of the session accessing the baseline counts toward the eligibility of the plan.

Posted in Oracle | Tagged: , , | 3 Comments »

HP Systems Management vacancy

Posted by John Hallas on August 8, 2017

I know I do not have the right readership on this blog for a Systems Management vacancy but if any readers have colleagues who have experience using any of HP Openview, OpsBridge, HP Service Management and HP UCMDB then I have a vacancy at our Head Office in Bradford for a permanent position

Details below

https://apply.morrisons.jobs/vacancies/491/technology-specialist–system-management.html

Posted in Oracle | Leave a Comment »