Gary Smith wrote:
> It's not a connection pooling issue per say. We have several boxes running spam
> assassin, sqlgrey and postfix (via mysql). Normally these components work great. SA and
> sqlgrey both have a fixed number of connections, around 16, that they are generally
> actively using unless we get a burst of email, at which time they will increase by a few.
> The problem is that Postfix has been receiving a higher level than normal of emails as we
> have taken 50% of our servers offline at this location (setting them up at a new
> location). We're also have this bouncing across a couple different firewalls, so for some
> reason, the conneciton to mysql is generating a larger number of these:
> 090407 12:26:42 [Warning] Aborted connection 972479 to db: 'db' user: 'user' host:
> 'host' (Got an error reading communication packets)
> We do know the network isn't optimal right now and are working to fix the issues but
> we are hoping to get by just for the short term.
> But that leads back to the original question about increase the connection error
> cutoff before banning a host.
> We are using 5.1.32 with INNODB tables.
> From: mos [mos99@stripped]
> Sent: Tuesday, April 07, 2009 9:18 AM
> To: mysql@stripped
> Subject: Re: flush-host problem
> At 10:39 AM 4/7/2009, Gary Smith wrote:
>> I have system that is generating a larger than normal number of connection
>> errors. We know why the errors are occuring and are working to resolve
>> them (connectivity and load issue on the client). The question is, how
>> can I tweak mysql to tolerate a higher level than normal of bad
>> connections before banning the host.
>> What happens is that when we have 300-500 connections a few random ones
>> will get mucked up during a heavier than normal load on the client. I
>> have set the max connections to 3000 (which we never get close to).
>> So, if there a config/startup setting to tweak to ease the banning of bad
>> connetions thus reducing the need for me to continually "mysqladmin
>> flush-host" on the server?
The server generates those "Aborted connection" messages under the
1) The server was trying to pass information to the client and the
client stopped being there.
2) The client sat idle beyond the wait_timeout setting.
3) There was some kind of networking interference (such as a VPN or
proxy server closing an idle connection) or bad connections.
Look at your SHOW PROCESSLIST report. If it shows many connections that
have been idle for more than 30 seconds or so, then you need to change
your client software behavior. If these are pooled connections, make
sure your idle connection timeout for the pool (part of the pool
configuration) is set to a value smaller than wait_timeout. Otherwize
you need to audit your client code to ensure that it is calling the
appropriate close() function for the library it is using to connect to
the MySQL server. That will release the connection and allow the server
to reallocate those resources to handling the commands for the live (and
not idle) connections.
For additional reasons for these errors, please peruse:
Shawn Green, MySQL Senior Support Engineer
Sun Microsystems, Inc.
Office: Blountville, TN