Monday, May 27, 2013

The Adventures of the Trickster developer - Aliases

The Trickster is a mythological being who enjoys potentially dangerous counter-intuitive behaviour. You'll often find him deep within the source code of large systems.

The Alias Trap

Generally an alias in a query is there to make it easier to understand, either for the developer or the database. However the Trickster can reuse aliases within a query to make things more confusing.


desc scott.emp
       Name       Null?    Type
       ---------- -------- --------------------
1      EMPNO      NOT NULL NUMBER(4)
2      ENAME               VARCHAR2(10)
3      JOB                 VARCHAR2(9)
4      MGR                 NUMBER(4)
5      HIREDATE            DATE
6      SAL                 NUMBER(7,2)
7      COMM                NUMBER(7,2)
8      DEPTNO              NUMBER(2)

desc scott.salgrade
       Name      Null?    Type
       --------- -------- --------------------
1      GRADE              NUMBER
2      LOSAL              NUMBER
3      HISAL              NUMBER


desc scott.dept
       Name      Null?    Type
       --------- -------- --------------------
1      DEPTNO    NOT NULL NUMBER(2)
2      DNAME              VARCHAR2(14)
3      LOC                VARCHAR2(13)


The trickster can try queries such as

SELECT e.ename, e.grade
FROM scott.emp e 
       JOIN scott.salgrade e ON e.sal BETWEEN e.losal AND e.hisal;

and


SELECT x.ename, x.dname
from scott.emp x join scott.dept x using (deptno);



As long as any prefixed column names involved are unique to a table, the database can work out what to do. 

If you find this in a 'live' query, it is normally one with at least half a dozen tables where an extra join has been added without noticing that the alias is already in use. And you'll discover it when a column is added to one of those tables causing the database to throw up its hands in surrender. Sometimes you will find it in a dynamically constructed query, when it will fail seemingly at random. 

Once discovered, it isn't a mentally difficult bug to resolve. But first you have to get past the mental roadblock of "But all the columns already have an alias pointing to the table they come from".

Thursday, May 16, 2013

SCNs and Timestamps

The function ORA_ROWSCN returns an SCN from a row (or more commonly the block, unless ROWDEPENDENCIES has been used).

select distinct ora_rowscn from PLAN_TABLE;

But unless you're a database, that SCN doesn't mean much. You can put things in some sort of order, but not much more.

Much better is

select sys.scn_to_timestamp(ora_rowscn) from PLAN_TABLE;

unless it gives you

ORA-08181: specified number is not a valid system change number

which is database-speak for "I can't remember exactly".

That's when you might be able to fall back on this, plugging the SCN in place of the **** :

select * from 
  (select first_time, first_change# curr_change, 
          lag(first_change#) over (order by first_change#) prev_change,
          lead(first_change#) over (order by first_change#) next_change
  FROM v$log_history)
where **** between curr_change and next_change

It won't be exact, and it doesn't stretch back forever. But it is better than nothing.

PS. This isn't a perfect way to find when a row was really inserted/updated. It is probably at the block level, and there's 'stuff' that can happen which doesn't actually change the row but might still reset the SCN. If you're looking for perfection, you at the wrong blog :)