dbKoda for MongoDB

dbKoda for MongoDB - a modern open source database IDE, now available for MongoDB. Download it here!

Next Generation
Databases: NoSQL,
NewSQL and Big Data

Buy at Amazon

Read sample at Amazon

Buy at Apress

Latest Postings:

Search

Oracle Performance Survival Guide

Buy It
Read it on Safari
Scripts and Examples
Sample Chapter

                                    

Powered by Squarespace

 MySQL Stored procedure programming

Buy It
Read it on Safari
Scripts and Examples 

                                                

Wednesday
Sep272006

10g tracing quick start

Oracle’s released a few new facilities to help with tracing in 10g,  here’s a real quick wrap up of the most significant:

Using the new client identifier

You can tag database sessions with a session identifier that can later be used to identify sessions to trace.  You can set the identifier like this:

begin
    dbms_session.set_identifier('GUY1');
end;


You can set this from a login trigger if you don’t have access to the source code.  To set trace on for a matching client id, you use DBMS_MONITOR.CLIENT_ID_TRACE_ENABLE:

BEGIN

   DBMS_MONITOR.client_id_trace_enable (client_id      => 'GUY1',
                                        waits          => TRUE,
                                        binds          => FALSE
                                       );
END;

You can add waits and or bind variables to the trace file using the flags shown.

Tracing by Module and/or action

Many Oracle-aware applications set Module and action properties and you can use these to enable tracing as well.  The serv_mod_act_trace_enable method allows you to set the tracing on for sessions matching particular service, module, actions and (for clusters) instance identifiers.  You can see current values for these usng the following query:

SELECT DISTINCT instance_name, service_name, module, action
           FROM gv$session JOIN gv$instance USING (inst_id);



INSTANCE_NAME    SERVICE_NA MODULE                         ACTION

Wednesday
Sep272006

10g time model query

Joining the 10g time model to the traditional wait interface views and taking advantage of the wait_class data is something most of us have probably done.  Here's my standard queries that do that thing....

fCOLUMN wait_class format a20
COLUMN name format a30
COLUMN time_secs format 999,999,999,999.99
COLUMN pct format 99.99

SELECT   wait_class, NAME, ROUND (time_secs, 2) time_secs,
         ROUND (time_secs * 100 / SUM (time_secs) OVER (), 2) pct
    FROM (SELECT n.wait_class, e.event NAME, e.time_waited / 100 time_secs
            FROM v$system_event e, v$event_name n
           WHERE n.NAME = e.event AND n.wait_class <> 'Idle'
                 AND time_waited > 0
          UNION
          SELECT 'CPU', 'server CPU', SUM (VALUE / 1000000) time_secs
            FROM v$sys_time_model
           WHERE stat_name IN ('background cpu time', 'DB CPU'))
ORDER BY time_secs DESC;

Which generates CPU and wait event times broken down to the event name:

WAIT_CLASS           NAME                                     TIME_SECS    PCT

Monday
May082006

Unit testing stored procedures with Junit

Anyone who has used an automated unit testing framework such as Junit knows just how life-changing an automated test suite can be.   Once you've  experienced validating that recent changes have not broken old code, or discovering subtle bugs via junit that would otherwise have remained undetected , you naturally want to have this capability in all your programming environments.

Guisseppe Maxia has written a few stored procedure snippets to assist with automated unit testing of MySQL routines.  Unfortunately, the MySQL stored procedure language itself does not have the necessary abilities to fully implement the sort of unit testing we would like.  In particular, the inability for a stored procedure to capture the result sets generated by another stored procedure prevents a stored procedure from fully unit testing another. 

So I decided that - for me - Junit offered the best solution.  I created an extension to the Junit class that contains some assertions useful when unit testing stored procedure and extended my PlainSql JDBC wrapper classes to allow a stored procedure to return a single object that contains all its result sets and the values of OUT or INOUT parameters.   This object can be made the target of the various assertions.

If you're not familiar with Java and/or, you might feel that this solution is not for you.  However, the amount of java programming you need to do is very minimal and GUI environments such as Eclipse make it very easy to set up.  The rest of this article is available here;  it contains instructions, my Java classes and examples for setting up Junit test cases for MySQL stored procedures.

Monday
Apr242006

MySQL stored procedures with Ruby

Ruby's getting an incredible amount of attention recently, largely as the result of Ruby on Rails.  I've played a little with Ruby on Rails and it certainly is the easiest way I've seen so far to develop  web interfaces to a back-end database.

At the same time,  I've been shifting from perl to Java as my language of choice for any serious database utility development.  But I still feel the need for something dynamic and hyper-productive when I'm writing something one-off or for my own use.  I've been playing with Python, but if Ruby has the upper ground as a web platform then maybe I should try Ruby. 

So seeing as how I've just finished the MySQL stored procedure book, first thing is to see if I can use Ruby for MySQL stored procedures.

Database - and MySQL - support for Ruby is kind of all over the place.  There's a DBI option (similar to perl) which provides a consistent interface and there's also native drivers.  For MySQL there are pure-ruby native drivers and drivers written in C.  Since the DBI is based on the native driver, I thought I'd try the native driver first.  The pure-ruby driver gave me some problems so I started with the C driver on Linux (RHAS4). 

Retrieving multiple result sets

The main trick with stored procedures is that they might return multiple result sets. OUT or INOUT parameters can be an issue too, but you can always work around that using session variables. 

If you try to call a stored procedure that returns a result set, you'll at first get a "procedure foo() can't return a result set in the given context error".  This is because the CLIENT_MULTI_RESULTS flag is not set by default when the connection is created.  Luckily we can set that in our own code:

dbh=Mysql.init
dbh.real_connect("127.0.0.1", "root", "secret", "prod",3306,nil,Mysql::CLIENT_MULTI_RESULTS)

The "query" method returns a result set as soon as it is called, but I found it easier to retrieve each result set manually, so i set the query_with_result attribute to false:

dbh.query_with_result=false

The next_result and more_results methods are implemented in the Ruby MySql driver, but there's some weird things about the more_results C API call that causes problems in python and PHP.  In Ruby, the more_results call returns true whether or not there is an additional result.   The only reliable way I found to determine if there is another result set is to try and grab the results and bail out if an exception fires (the exception doesn't generate an error code, btw);
      
    dbh.query("CALL foo()")
    begin
      rs=dbh.use_result
    rescue Mysql::Error => e 
      no_more_results=true
    end

.
We can then call more_results at the end of each rowset loop.  So here's a method that dumps all the result sets from a stored procedure call as XML using this approach (I'm know the Ruby is probably crap, it's like my 3rd Ruby program):

def procXML(dbh,sql)
  connect(dbh)
  no_more_results=false
  dbh.query(sql)
  printf("<?xml version='1.0'?>\n");
  printf("<proc sql=\"%s\">\n",sql)
  result_no=0
  until no_more_results
    begin
      rs=dbh.use_result
    rescue Mysql::Error => e 
      no_more_results=true
    end 
     if no_more_results==false
      result_no+=1
      colcount=rs.fetch_fields.size
      rowno=0
      printf("\t<resultset id=%d columns=%s>\n",result_no,colcount)
      rs.each do |row|
        rowno+=1
        printf "\t\t<row no=%d>\n",rowno
        rs.fetch_fields.each_with_index do |col,i|
          printf("\t\t\t<colvalue column=\"%s\">%s</colvalue>\n",col.name,row[i])
        end
        printf("\t\t</row>\n")
      end
      printf("\t</resultset>\n");
      rs.free
      dbh.next_result
    end
  end
  printf("</proc>\n")
end

No C programming required!

Whew!  No need to hack into the C code.  So you can use MySQL stored procedures in Ruby with the existing native C driver. The problem is that the C driver is not yet available as a binary on Windows yet and trying to compile it turns out to be beyond my old brain (and yes, I used minGW and all the other "right" things).   Hopefully a copy of the MySQL binary driver it will be available in the one-click installer Ruby installer eventually.

The above code doesn't work using the pure-Ruby driver on windows by the way -  there's an "out of sequence" error when trying to execute the stored proc.  I might hack around on that later (at the moment I'm 35,000 ft with 15 minutes of battery left on the way to the MySQL UC).  For now if you want to use MySQL stored procedures in a ruby program on windows I can't help.

Note that ruby seems to hit a bug that causes MySQL to go away if there are two calls to the same stored proc in the same session and the stored proc is created using server-side prepared statements.  Fixed soon hopefully, but for now if you get a "MySQL server has gone away error" you might be hitting the same problem.   Wez posted on this problem here.

I suppose the end of this investigation will probably be to see if there's any way to use stored procedure calls to maintain a Rails AcitveRecord object.  Not that I think you'd necessarily want to, but it would probably be a good learning exercise.

Sunday
Apr092006

Plain Old SQL Statements in Java

Lot's of Java developers want to avoid writing SQL or even avoid directly accessing with a relational data store.  I, on the other hand, want to use SQL in my Java programs (which are mostly database utilities) but I want it to be as easy to use SQL in Java as it is in Python, Perl or PHP. 

Recently, I decided to use Java rather than Perl as my language of choice for database utilities.  Perl is quick to write, but tends to be easier to write than to read.  I played with Python for a while, but in the end decided that by using Java I would have an easier time distributing by stuff and it could more easily be re-used inside of my company, which has Java developers but not many Perl people.  So developed a set of JDBC wrappers that would let me use SQL easily within my utilities.   

My Design :

  • Require no configuration files (no XML mappings for instance).  All the data access logic should be right there in the code.  Although I did end up allowing SQL statement to be held in an XML file just to avoid the whole messy string handling involved in really long SQL statements.   
  • Easy processing of result sets, leveraging Java collections (which were not around when JDBC was first speced).
  • Make it easiest to follow best practices.  For instance, make it very easy to use bind variables, re-use cursors, etc.
  • Allow interoperability with the underlying JDBC objects so that I could use these classes without worrying that I would run into a brick wall.
  • Work well with dynamic SQL.
  • Where appropriate, avoid some of the tedium involved with DML and DDL.
  • Cache prepared statements so that you don't have to worry about which prepared statements to keep open and when to close.
  • Be RDBMS neutral. 

This was a bit of a learning experience - had not programmed in Java for quite a while and I tried to follow best Java practice such as creating Junit tests, JavaDoc and ant builds.  If anyone's interested,  here's the latest versions:

I don't really expect anyone else to use this, posting it was I guess one of the ways to enforce a bit of discipline on myself as regards quality. However, it will be embedded in most of my database utilities so I guess it will see some use in our internal benchmarking routines and the like.