Sokrates on Oracle

Issue with updatable views

Posted by Matthias Rogel on 7. January 2014

It’s sometimes amazing, how many bugs there are still with elementary SQL.

Here is one concerning updatable views:

sokrates@12.1 > create table t ( v varchar2(30) );

Table created.

sokrates@12.1 > create view v as
  2  select v as dontdothatman, v as canbelostwheninserted
  3  from t; 

View created.

sokrates@12.1 > insert /* this is fine */ into v 
  2  values('fine', 'fine');

1 row created.

sokrates@12.1 > select * from v;

------------------------------ ------------------------------
fine			       fine

sokrates@12.1 > insert /* exception expected because 1st value is lost */ into v
  2  values('this one is lost', 'why isnt that one lost ?');

1 row created.

sokrates@12.1 > select * from v;

------------------------------ ------------------------------
fine			       fine
why isnt that one lost ?       why isnt that one lost ?


Posted in 12c, Bug, sql | 4 Comments »

Best Practice in 12c

Posted by Matthias Rogel on 4. December 2013

Since PL/SQL now is closely integrated into SQL, we hence can happily state

sokrates@12.1 > with function bestpractice return varchar2
  2  is
  3  begin
  4     return 'Do not use PL/SQL when it can be done with SQL alone !';
  5  end bestpractice;
  6  select bestpractice() from dual
  7  /

Do not use PL/SQL when it can be done with SQL alone !

Posted in 12c, Allgemein, fun, sql | Tagged: | 2 Comments »

Strange ORA-14196

Posted by Matthias Rogel on 7. October 2013

It seems that sometimes you need a non-unique index to enforce a unique constraint even if this constraint is declared as not deferrable.

sokrates@11.2 > create table strange(i int not null, j int not null);

Table created.

sokrates@11.2 > alter table strange add constraint unique_i unique(i) not deferrable
  2  using index ( create unique index struix on strange ( i, j ) )
  3  /
alter table strange add constraint unique_i unique(i) not deferrable
ERROR at line 1:
ORA-14196: Specified index cannot be used to enforce the constraint.

We have to create a non-unique index here !

sokrates@11.2 > alter table strange add constraint unique_i unique(i) not deferrable
  2  using index ( create  index struix on strange ( i, j ) )
  3  /

Table altered.

Also reproduced on 12.1.
Who can explain this behaviour to me ( I suppose it is a bug ) ?

Posted in Allgemein | Tagged: | 3 Comments »

Partition Info in V$SESSION_LONGOPS

Posted by Matthias Rogel on 10. May 2013

Oracle’s advanced partitioning has some deficiencies. For example, partition info is missing in V$SESSION_LONGOPS for scan-operations ( full table scans, full index scans ). V$SESSION_LONGOPS.TARGET only shows OWNER.TABLE_NAME in these cases, even when the underlying table/index is partitioned, though the longop doesn’t refer to the whole segment but only to one (sub-)partition of it.
I filed an enhancement request several years ago concerning this matter, but never received any feedback.
However, there is a workaround to that. In many cases, we can find out on which (sub-) partition the longop is working on: V$SESSION_WAIT’s P1- and P2-info can be used for that in case the session is waiting mainly on I/O ( which might be most likely for many systems. )
Here is an extension to V$SESSION_LONGOPS which tries to figure out this additional info.

Update 27/02/2014
Note that the original version has been improved by Jonathan Lewis. I have marked the relevant part with a corresponding comment.
I haven’t observed so far that I wasn’t able to get the partition information from v$session.row_wait_obj# ( as suggested by him ), but from the part marked as “superfluous most likely” ( my original version ). However, I have no proof that this is not possible.

            select 'does not apply'
            from dual
            where slo.TARGET not like '%.%'
            or slo.TARGET is null
            select 'does not apply'
            from dba_tables dt
            where dt.OWNER=substr(, 1, instr(, '.') - 1)
            and dt.TABLE_NAME=substr(, instr(, '.') + 1)
            and dt.PARTITIONED='NO'       
            -- Jonathan Lewis, see
               ob.subobject_name || ' (' || ob.object_type || ')'
            from v$session s, dba_objects ob
              ob.object_id = s.row_wait_obj#
            and s.sid = slo.sid
            and ob.OBJECT_TYPE like '%PARTITION%'            
            -- superfluous most likely
               de.partition_name || ' (' || de.segment_type || ') NOT SUPERFLUOUS IF YOU SEE THAT'
            from v$session_wait sw, dba_extents de
            and slo.opname like '%Scan%'
            and sw.P1TEXT like 'file%'
            and sw.P1 = de.FILE_ID and sw.P2 between de.BLOCK_ID and de.BLOCK_ID + de.BLOCKS - 1
            and de.owner = substr(, 1, instr(, '.') - 1)
            and de.segment_type in
               'INDEX PARTITION', 'INDEX SUBPARTITION'            
            and de.segment_name in
                 -- table
                    substr(, instr(, '.') + 1)
                 from dual
                 union all
                 -- index
                 select di.index_name
                 from dba_indexes di
                 where di.owner=substr(, 1, instr(, '.') - 1)
                 and di.TABLE_NAME = substr(, instr(, '.') + 1)
   as partition_info,     
from v$session_longops slo
where slo.TIME_REMAINING > 0

Note that this might take a bit longer than a simple

select slo.*
from v$session_longops slo
where slo.TIME_REMAINING > 0

, though due to coalesce’s short circuiting it is quite efficient.

Posted in Allgemein, sql | Tagged: | 2 Comments »


Posted by Matthias Rogel on 29. April 2013

Tim wrote
… the UTL_RAW package has a bunch of casting functions for RAW values (CAST_TO_BINARY_DOUBLE, CAST_TO_BINARY_FLOAT, CAST_TO_BINARY_INTEGER, CAST_TO_NUMBER, CAST_TO_NVARCHAR2, CAST_TO_VARCHAR2). Note the absence of a CAST_TO_DATE function.

Bertrand Drouvot also misses it, see Bind variable peeking: Retrieve peeked and passed values per execution in oracle 11.2

Here is a try to write one, fixes and improvements are welcome !

create or replace function CAST_TO_DATE(bdr in raw) return date deterministic is
         100 * (to_number(substr(bdr,1,2), 'xx') - 100) + 
         to_number(substr(bdr,3,2), 'xx') - 101, 
     + NUMTOYMINTERVAL(to_number(substr(bdr,5,2), 'xx')-1, 'month')
     + NUMTODSINTERVAL(to_number(substr(bdr,7,2), 'xx')-1, 'day')
     + NUMTODSINTERVAL(to_number(substr(bdr,9,2), 'xx') - 1, 'hour')   
     + NUMTODSINTERVAL(to_number(substr(bdr,11,2), 'xx') - 1, 'minute')   
     + NUMTODSINTERVAL(to_number(substr(bdr,13,2), 'xx') - 1, 'second');
  exception when others then return to_date(1, 'J');
end CAST_TO_DATE;   

Posted in Allgemein | Tagged: | 4 Comments »

A simple pipelined version of print_table

Posted by Matthias Rogel on 10. April 2013

Tom Kyte’s print_table procedure, available on
seems to be very popular and there exist tricky variations on the theme, for example the following nice xml-trick by Sayan Malakshinov.

Please note that it is very easy to use the existing print_table-code to generate a pipelined version which can be used in SQL.
I use the following code since ages and it always does me a great job, so probably it is worth sharing.

create or replace function fprint_table
( p_query in varchar2,
  p_date_fmt in varchar2 default 'dd-mon-yyyy hh24:mi:ss' )
return sys.odcivarchar2list 
authid current_user
l varchar2(4000);
s integer default 1;
  dbms_output.enable(buffer_size => null);
     p_query => p_query,
     p_date_fmt => p_date_fmt

     dbms_output.get_line(line => l, status => s);
     exit when s != 0;
        pipe row(l);
     exception when no_data_needed then exit;
  end loop;

end fprint_table;

sokrates@11.2 > select * from table(fprint_table('select user,sysdate from dual'));

USER                          : SOKRATES
SYSDATE                       : 10-apr-2013 12:27:50
1 row selected.

Posted in Allgemein | Tagged: | Leave a Comment »

An undocumented restriction in Workspace Manager – exporting tables with valid time support

Posted by Matthias Rogel on 7. February 2013

If you are using Workspace Manager, it could be probably useful to know, that there is an undocumented restriction concerning import/export.
Due to Import and Export Considerations,
…Workspace Manager supports the import and export of version-enabled tables in one of the following two ways: a full database import and export, and a workspace-level import and export through Workspace Manager procedures. No other export modes, such as schema, table, or partition level, are currently supported….

However, this does not hold for tables with valid time support:

sokrates@11.2 > CREATE TABLE d (id NUMBER PRIMARY KEY);

Table created.

sokrates@11.2 > EXECUTE DBMS_WM.EnableVersioning (table_name=>'D', validTime=>TRUE, hist => 'NONE');

PL/SQL procedure successfully completed.

sokrates@11.2 >  EXECUTE DBMS_WM.Export(table_name => 'D',staging_table => 'D_STG', workspace => 'LIVE');
BEGIN DBMS_WM.Export(table_name => 'D',staging_table => 'D_STG', workspace => 'LIVE'); END;

ERROR at line 1:
ORA-20171: WM error: Export not supported on a table with valid time
ORA-06512: at "WMSYS.LT", line 13185
ORA-06512: at line 1

Support confirmed, that in this case only full db import/export (!) is supported, documentation would be updated somewhen.

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


Posted by Matthias Rogel on 4. February 2013

The biggest advantage of being developer and DBA at the same time in my eyes is: tuning lies in one hand.
Peter Scott twittered about bringing a query down from 25 hours to 83 seconds by rewriting a query using MINUS.
Funny, in February 1996 I started a new job and my first task was tuning a query running for several hours – basically

select ...
from t
where not exists
( select ...
  from r@remote r
  where <join t and r>

which could be tuned down to a few seconds using MINUS, so quite similar to Peter’s job ( Version was 7.0.something at that time as far as I remember ).
In my experience, most performance gains are achieved by rewriting SQL or even by restructuring your entire application logic.

Posted in Allgemein | Leave a Comment »

what’s in my buffer cache ?

Posted by Matthias Rogel on 4. February 2013

The following SQL shows me what is currently in my buffer cache and runs in a reasonable amount of time ( never longer than 30 seconds with some dozens of GB buffer cache and around 5 million entries in v$cache ), it also shows me the cached percentage of each segment which is currently part of the cache.
Version is 11.2.

with cache_raw 
     c.owner#, c.kind,, c.partition_name, 
     c.status, c.file#, 
     count(*) co
  from v$cache c
  group by 
     c.owner#, c.kind,, c.partition_name,
     c.partition_name, file#, c.status
), cache_raw2
     du.username, c.kind, 
     case when c.partition_name is null
            else || ' partition ' || c.partition_name
     end as name,
     case c.status
        when 'free' then 'not currently in use'
        when 'xcur' then 'exclusive'
        when 'scur' then 'shared current'
        when 'cr' then 'consistent read'
        when 'read' then 'being read from disk'
        when 'mrec' then 'in media recovery mode'
        when 'irec' then 'in instance recovery mode'
     end as status,
     co * (select value from v$parameter where name='db_block_size') as anzb
  from cache_raw c, dba_users du
  where du.user_id(+)=c.owner#
), cache_segm as
     c.username as owner, c.kind, 
       when c.kind in ('INDEX', 'INDEX PARTITION')
       then || ' index on '|| ind.table_name
     end as name, 
     round(100 * c.anzb / sum(c.anzb) over (), 2) percentage,
     round(c.anzb / (1024 * 1024 * 1024), 2) as gbytes_in_cache, 
     round(100 * c.anzb / seg.bytes, 2) perc_of_segment_in_cache, 
     round(seg.bytes / (1024 * 1024 * 1024), 2) as gbytes_in_segment,   
     round(c.anzb / (1024 * 1024), 2) as mbytes_in_cache,
     round(seg.bytes / (1024 * 1024), 2) as mbytes_in_segment,   
     round(c.anzb / (1024), 2) as kbytes_in_cache,   
     round(seg.bytes / (1024), 2) as kbytes_in_segment,            
     c.anzb as bytes_in_cache, 
     seg.bytes as bytes_in_segment,   
     seg.segment_subtype, seg.tablespace_name
  from cache_raw2 c, dba_segments seg, dba_indexes ind
     case when seg.partition_name(+) is null
            then seg.segment_name(+)
            else seg.segment_name(+) || ' partition ' || seg.partition_name(+)
     end =
     and seg.segment_type(+)=c.kind
     ind.owner(+)=c.username and ind.index_name(+)=nvl(substr(, 1, instr(, ' ', 1) - 1),
     case row_number() over(order by c.bytes_in_cache desc) 
       when 1 then round(sum(c.bytes_in_cache) over () / (1024 * 1024 * 1024), 2) 
     end as gb_total,
     c.owner, c.kind,, 
     c.gbytes_in_cache, c.perc_of_segment_in_cache, 
     c.gbytes_in_segment, c.status, 
     c.mbytes_in_cache, c.mbytes_in_segment,   
     c.kbytes_in_cache, c.kbytes_in_segment,            
     c.bytes_in_cache, c.bytes_in_segment,   
     c.segment_subtype, c.tablespace_name
  from cache_segm c
  order by 
     c.bytes_in_cache desc

Posted in Allgemein | Tagged: | 2 Comments »

Mount an xml-file as “External Table”

Posted by Matthias Rogel on 3. January 2013

“Mounting” a csv-file as external-table is very handy.

Here a quick note to show, that we can “mount” an xml-file as well ( and query it via SQL )


sokrates@11.2 > create directory t as '/tmp';
Directory created.
sokrates@11.2 > !wget -O /tmp/t.xml
Connecting to||:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 1135 (1.1K) [text/xml]
Saving to: `/tmp/t.xml'
100%[============================================================================================================================>] 1,135 --.-K/s in 0s
17:31:25 (216 MB/s) - `/tmp/t.xml' saved [1135/1135]
sokrates@11.2 > !cat /tmp/t.xml
<?xml version="1.0" encoding="ISO-8859-1"?>
<!-- Edited by XMLSpy? -->
 <name>Belgian Waffles</name>
 <description>two of our famous Belgian Waffles with plenty of real maple syrup</description>
 <name>Strawberry Belgian Waffles</name>
 <description>light Belgian waffles covered with strawberries and whipped cream</description>
 <name>Berry-Berry Belgian Waffles</name>
 <description>light Belgian waffles covered with an assortment of fresh berries and whipped cream</description>
 <name>French Toast</name>
 <description>thick slices made from our homemade sourdough bread</description>
 <name>Homestyle Breakfast</name>
 <description>two eggs, bacon or sausage, toast, and our ever-popular hash browns</description>
sokrates@11.2 > select
 2 x.*
 3 from XMLTable(
 4 '/breakfast_menu/food'
 5 passing
 6 xmltype(
 7 bfilename('T','t.xml'), nls_charset_id('WE8ISO8859P1')
 8 )
 9 columns
 10 name varchar2(50) path 'name',
 11 price varchar2(20) path 'price',
 12 description varchar2(200) path 'description',
 13 calories number path 'calories'
 14 ) x;
------------------------------ ------- ------------------------------------------------------------------------------------------ ------------
Belgian Waffles $5.95 two of our famous Belgian Waffles with plenty of real maple syrup 650
Strawberry Belgian Waffles $7.95 light Belgian waffles covered with strawberries and whipped cream 900
Berry-Berry Belgian Waffles $8.95 light Belgian waffles covered with an assortment of fresh berries and whipped cream 900
French Toast $4.50 thick slices made from our homemade sourdough bread 600
Homestyle Breakfast $6.95 two eggs, bacon or sausage, toast, and our ever-popular hash 

Posted in Allgemein | Tagged: | 2 Comments »