From my holiday reading and from watching the most excellent DB2 Night Show and specifically an episode that was done a while back by Iqbal Goralwalla of Triton Consulting (@iqbalgoralwalla) on “DB2 LUW 10.1 Cool Features No One is Talking About” I have come across Usage Lists for tables and indexes in DB2.
Why use – Usage Lists
Have you got a table or an index and you never know when or how it is used; Stored Procedures, screens, systems or dynamic SQL, or do you want to monitor the SQL that runs against a table or index then and what work is done then this could save you ploughing through a lot of code, but means it won’t be an instant fix as the code has to run.
Usage Lists
Usage Lists in DB2 essentially allow you to monitor the SQL that runs against a tables or indexes that you have identified that you want monitoring. This does not come without costs and a list of the GOTCHA’s can be found on the “Notes” section of the page here and in “Chapter 26. Usage lists” of the “Preparation Guide for Exam 611” (not sure on how much these will come up in the exam?)
GOTCHA
- Please note in the above paragraph the words “you have identified that you want monitoring” as you will see you will only get the stats if the table is monitored and you have set up the individual monitor!
Usage Lists – Creation
Not going to lie there is a page of the IBM Info Centre that has a version of this information but it is a little hard to find unless you type in the exact words but it can be found here, as you can see from the title then it is not really close to usage lists!
First you need to set a database configuration parameter MON_OBJ_METRICS:
db2 UPDATE DATABASE CONFIGURATION USING MON_OBJ_METRICS EXTENDED
On the page mentioned above then it says you need to set this so that "statistics are collected for each entry in the usage list" but on the small scale of the testing that I did I have not found any difference in captured data.
Then for each table that you want to monitor then you need to run at a minimum:
db2 CREATE USAGE LIST {Some Memorable Name} FOR TABLE {Schema}.{Table}
There are other parts to this command that can be found here and it has some useful parts like the ability too "turn its self off" when a certain number of different statements have been run by doing something like:
db2 CREATE USAGE LIST {Some Memorable Name} FOR TABLE {Schema}.{Table} LIST SIZE {Some Number} WHEN FULL DEACTIVATE
Or a rolling list, but this might create difficulties if you want repeatability:
db2 CREATE USAGE LIST {Some Memorable Name} FOR TABLE {Schema}.{Table} LIST SIZE {Some Number} WHEN FULL WRAP
From testing if you unless you specify a LIST SIZE then the collection will continue for as long as the list is active, which is the next statement to run to get it too work.
db2 SET USAGE LIST {Some Memorable Name} STATE = ACTIVE
And to disable it again:
db2 SET USAGE LIST {Some Memorable Name} STATE = INACTIVE
So above is a quick look at how to get this to work and the links to get a better lets move on to look at what it collects.
Usage Lists - The output
The output is quite useful and the full output of the MON_GET_TABLE_USAGE_LIST table function can be found here. It is also a little disappointing because this does not return the statement only an identifier (EXECUTABLE_ID) that you can supply to the MON_GET_PKG_CACHE_STMT table function which info for this can be found at here.
You can do something like this and potentially get a lot of data on what your MON_GET_TABLE_USAGE_LIST captured and the statements from MON_GET_PKG_CACHE_STMT when joined together:
SELECT *
FROM TABLE(MON_GET_TABLE_USAGE_LIST(NULL,{Some Memorable Name},0)) A
LEFT JOIN
TABLE(MON_GET_PKG_CACHE_STMT(NULL, NULL, NULL, -2)) B
ON A.EXECUTABLE_ID = B.EXECUTABLE_ID
Looking at a version that yields some more focused information:
SELECT B.STMT_TEXT AS SQL_STATEMENT,
A.LAST_UPDATED AS LAST_RUN,
A.NUM_REF_WITH_METRICS AS NO_TIMES_RUN,
A.ROWS_READ,
A.ROWS_INSERTED,
A.ROWS_UPDATED,
A.ROWS_DELETED,
A.LOCK_WAIT_TIME,
A.OBJECT_DATA_L_READS AS BUFFERPOOL_READS,
A.OBJECT_DATA_P_READS AS NON_BUFFERPOOL_READS
FROM TABLE(MON_GET_TABLE_USAGE_LIST(NULL,{Some Memorable Name},0)) A
INNER JOIN
TABLE(MON_GET_PKG_CACHE_STMT(NULL, NULL, NULL, -2)) B
ON A.EXECUTABLE_ID = B.EXECUTABLE_ID
This enables you to see how efficient the query is in terms of how often it is run and the number of times, with the work that they did and how much of the data resides in the bufferpools (BUFFERPOOL_READS) and how much has to come from disk (NON_BUFFERPOOL_READS). As you can see from my not very good test system query tracking below:
SQL_STATEMENT LAST_RUN NO_TIMES_RUN ROWS_READ ROWS_INSERTED ROWS_UPDATED ROWS_DELETED LOCK_WAIT_TIME BUFFERPOOL_READS NON_BUFFERPOOL_READS
------------------------------------------------------------------------------------------------------------ ------------------- --------------- ------------ ---------------- --------------- --------------- ----------------- ------------------- -----------------------
insert into {schema}.{table} ({field},{field},{field}) VALUES ({value},{value},{value}) 21/04/2013 10:10:54 1 0 1 0 0 0 1 0
As you can see here this update all happened inside the bufferpool as it was very small on a table with no data. If you can find a statement that you are interested in because it has a large amount of non logical data reads you can use the captured code and pass it through db2advis to get suggestions on how to make the query better with indexes etc. Please see my blog post on db2advis if you are un-familar with it.
The future
I am currently looking at automating db2advis and monitoring its suggestions. Which once you are capturing the SQL becomes a lot easier.
