in our application server, after we start a transaction starts – we make sure we execute the following query using the connection object obtained from our pool. [i.e. the following query is the first query to be executed]

“select CONNECTION_ID() ”
refer to http://dev.mysql.com/doc/refman/5.0/en/information-functions.html#function_connection-id

This gives us the thread id which mysql would assign to this transaction.

we then log the following details – thread id & name of the application module which wants to start the db transaction.

when we see threads in sleep state(from show processlist) we note down the id & immediately look at our logs to identify the application module which has the corresponding id & debug the module code to see what it does. in most cases – we found our application thread starts of a database transaction and does a few queries and then makes a call to some other external system like a MQ server and gets stuck, & since the transaction is still active(i.e. not commited/rollbacked) – the thread in mysql goes to sleep state.

hope this helps

-vishnu rao
@flipkart.com

thank you for your valuable posts

Anyone interested in this feature I have requested?

http://bugs.mysql.com/bug.php?id=60476

However, as it is at the moment, tracking info in an SQL comment is useless in sleep state as the INFO column in processlist is NULL when connection is sleeping (understandable as there is actually no query executing)

Therefore, I would like to see a “Last Query” in the processlist, which simply maintains the last query name that was executed on that connection. Then tracking info in an SQL comment becomes much more useful in these “sleeping thread”, and other scenarios…

This would be a VERY powerful feature if you ask me.

I have seen it mentioned before, but it was closed as they did not consider it to be worthy of their attention…

http://forge.mysql.com/worklog/task.php?id=384

Cheers

//Steve

Not sure if it’s just me but on Mac with Safari and Chrome the code section is a bit screwy. The closing tegs don’t seem to work. After the line “This port can be used to identify the process which owns connection in question:” the rest of the article is in a code block

1. apache requests increase
2. mysql queries increase
3. mysql slow queries increase
4. apache connections start to wait on slow mysql queries
5. apache connections take longer to close
6. more apache connections are opened
7. with each apache request a new sleeping mysql connection is made
8. mysql & apache reach max_connections
9. server slows to a crawl

Solution:
- Reduce the time a mysql connection is open during script execution – don’t open a mysql connection in the header and close it in the footer. One approach is to create a query wrapper function and open a connection to the database only when the first query occurs – e.g:
function query ($query) {
if (!mysql_connection) mysql_connect();
return mysql_query();
}
- Reduce slow queries (always a good strategy!)

$link = mysql_connect (“localhost”, “root”, “password”);

$query = “show processlist;”;

$result = mysql_query($query) or die (“failed $query”);

while ($row = mysql_fetch_array($result))
{
if (($row['Time'] >

my.cnf:
wait_timeout=180 # = 30min

php cron script:

#!/usr/bin/php -q
1000 && $row['Command'] == “Sleep”) && ($row['User'] != “Root”))
{
$killed = 0;
mysql_query(“kill “.$row['Id']) or ($killed = 1);
if ($killed = 1)
{
echo
“Killed process number “.$row['Id'].”, with a User of “.$row['User']
.” in database “.$row['db'].”, which had a time of “.$row['Time']
.”, and a command of “.$row['Command'].”\r\n\r\n”;
}
}

}

@mysql_close($link);

?>