Skip to main content

But... we already have tools for explain plan

There are already ways to see execution plan

This is very true. And for day to day use, they are more then sufficient.

But sometimes the tools don't show the functions/conditions correctly.
And some details are never present (e.g. in complex plans with multiple materialized subqueries, it's sometimes impossible to determine from which copy of a table/subquery a column comes).

(And out of scope of this series, you may want to see bind values, or to know which line in the plan you are on = which (copy of) table is oracle just reading.)

Examples in this blog are necessarily simple ones; and in such examples we see cases where the added value of such access to execution plan is not very high. But never forget that this can be extended to much more complex cases.

Existing tools

EXPLAIN PLAN

EXPLAIN PLAN is long time with us. The main problem is that it shows plan for a new parse - and this can be different than what a session is actually executing, as it depends on bind types and lengths, NLS settings, and SQL baselines, many optimizer settings... and every version of Oracle introduces more dynamics and adaptability, making it harder and harder to get exactly the same plan.
On the other hand, EXPLAIN PLAN runs "offline" and thus has time to do more parsing of the plan than other approaches, and thus can parse some cases that other ways cannot. E.g. this output of explain plan won't be shown with dbms_xplan:
filter("Q"."PROD_NAME"= (SELECT LISTAGG(TO_CHAR("PROD_ID"),NULL) WITHIN GROUP ( ORDER BY "PROD_ID") FROM "PRODUCTS" "PRODUCTS")) 
But even explain plan fails sometimes:
filter(INTERNAL_FUNCTION("TIME_ID")=TIMESTAMP' 2000-01-01 00:00:00.000000000')
And in other cases (like the multiple subqueries/tables), it simply cannot show the full info, as the plan table does not allow that level of detail.

dbms_xplan

With dbms_xplan, it is easier to get the right plan, as we can specify an existing plan id, or existing cursor.  But it does not parse some of the more complicated cases - for example the first filter from above is shown as:
filter("Q"."PROD_NAME"=)

v$sql_plan

The output is similar to dbms_xplan. What is nice that this is a v$ view... so it should be backed up by a x$ view, and that should have an address in memory, something we should be able to read from the SGA. Let's have a look.

An example SQL

Our example is based on the sample SH schema, with an added table FOOBAR (id number, key varchar2(30)):

SELECT prod_id, key FROM products CROSS JOIN foobar WHERE prod_id in (143,144,id) and id in (1,2,3);

So let's have a look at the v$sql_plan. This is backed by x$kqlfxpl:

select addr, kqlfxpl_oper, kqlfxpl_oopt from x$kqlfxpl where kqlfxpl_sqlid='b4hdxqwy614fa'
ADDR             KQLFXPL_OPER         KQLFXPL_OOPT
---------------- -------------------- -------------------
00007F9F384F2E68 SELECT STATEMENT
00007F9F384F2C10 NESTED LOOPS
00007F9F384F2990 TABLE ACCESS         FULL
00007F9F384F2698 INDEX                FULL SCAN
The important detail is the ADDR. As we've seen in the previous post, SGA base is 0x60_000_000, but e.g the first ADDR is 
0x7F9_F38_4F2_E68, somewhere very, very far. It's also close to 2^48, the size of current CPU address space (64-bit CPUs use 48-bit address space, to save on the address line cost; it's still more than enough). Process stack tends to grow down from the top of process address space. This indicates that this is process' private stack memory.

The truth is that v$sql_plan does not read the plan directly - instead it is using a helper function to first populate a temporary array, and then to show that array. We thus need to go deeper to get the raw, unparsed execution plan.

Next time we will look where the function gets the data from.

Comments

Popular posts from this blog

ORA-27048: skgfifi: file header information is invalid

I was asked to analyze a situation, when an attempt to recover a 11g (standby) database resulted in bunch of "ORA-27048: skgfifi: file header information is invalid" errors.

I tried to reproduce the error on my test system, using different versions (EE, SE, 11.1.0.6, 11.1.0.7), but to no avail. Fortunately, I finally got to the failing system:

SQL> recover standby database;
ORA-00279: change 9614132 generated at 11/27/2009 17:59:06 needed for thread 1
ORA-00289: suggestion :
/u01/flash_recovery_area/T1/archivelog/2009_11_27/o1_mf_1_208_%u_.arc
ORA-27048: skgfifi: file header information is invalid
ORA-27048: skgfifi: file header information is invalid
ORA-27048: skgfifi: file header information is invalid
ORA-27048: skgfifi: file header information is invalid
ORA-27048: skgfifi: file header information is invalid
ORA-27048: skgfifi: file header information is invalid
ORA-00280: change 9614132 for thread 1 is in sequence #208


Interestingly, nothing interesting is written to alert.log n…

Reading data from PGA and SGA

Overview For our investigation of execution plan as it is stored in memory, we need in the first place to be able to read the memory.

We have the options of
x$ksmmem, reading SGA using SQL. Personally I don't like it, it's cumbersome and slow.direct SGA read: obviously reading SGA only; it's fast and easy to doread process memory: can read PGA, process stack - and since the processes do map the SGA, too, you can read it as well. Unfortunately ptrace sends signals to the processes and the process is paused when reading it, but so far all my reads were short and fast and the processes did not notice. Some OS configurations can prevent you from using ptrace (e.g. docker by default), google for CAP_SYS_PTRACE.gdb: using your favorite debugger, you can read memory as well. Useful when investigating. Direct SGA read I always considered direct SGA read of some dark magic, but the fundamentals are actually very easy. It still looks like sorcery when actually reading the Oracle in…

Filter and access predicates

More than just column projections When we look around for further pointers in the tree nodes, we find more pieces resembling the column projections we have seen so far. With some experimenting, we will find out that these are access predicates and filters.

First of all, the location of these pointers is not always the same, it seems that the value at 0x34 is some kind of flag, indicating whether filters and/or access predicates are present, and where. Or whether there is just one, or more of them.  It probably also indicates what other info is present, but I have no idea what info that would be or what each value means.
Resembling, but different The data we see as predicates are not columns; after all, a predicate is a condition, not a single column. But the structure is similar to what we have seen with columns, and if we follow pointers further, we eventually build a tree, and some of the leaves are indeed just column projections.
After some contemplation, we realize it's all t…