Not sure I follow what you're doing here?
Are those extra connections running queries? Why is it killing your
performance?
Do this:
Pick through 'show processlist' and find the source ip:port of the
_source_ address.
Log into the host by that IP, and use netstat -np and map the 'port'
from that entry in show processlist to a specific process. Find out what
that process is, what it's doing, if it's actually a tracker, and what
type of worker. (which you can find in ps).
-Dormando
this matches with the number of DB connections I see on the tracker, but
in the DB server there are many more connections coming from the tracker
server. The total number of connections I saw directly on MySQL is equal
to the number of connections I saw with netstat on the DB server. All
the connections are in state ESTABLISHED. Do you have any idea on what
can be happening?
Thanks again!
Fernando
On Thu, 2008-04-10 at 16:04 -0700, dormando wrote:
telnet to the tracker's management port and run !jobs.
you should have one DB connection per job type that exists. I figure
you're probably sending most of your traffic to one of the trackers?
If you have fewer workers than connections something else might be up.
Are those connections busy with queries? Can you tell what kind of
queries? etc.
-Dormando
Fernando Gomes wrote:
Hello
I'm using MogileFS with mysql DB, with two trackers and two storage
nodes. There are around 65000 files stored, and as they are replicated,
about 130000 total, distributed by the different devices.
All was working as expected until recently I noticed two problems
(perhaps related). One is that from time to time the files I got from
the client (java) wasn't the requested file (I suppose that might be
something like reported here:
http://davidrasch.com/2008/01/29/mogilefs-and-race-condition). I'll try
to look to the client java code in order to see what can be done.
The other problem is that today the trackers started creating a lot of
database connections (one tracker more than the other), almost taking
the database down. After finding that the database performance problem
was caused by one of the trackers I restarted it and things got a bit
better, but I still see many connections on the database from the
trackers.
When trying to diagnose the problem also found something that seems
strange to me (perhaps caused by the hours spent on this problem) - the
number of connections from one tracker to the database, evaluated by
netstat -n |grep 3306, is not the same if I execute netstat on the
tracker server or on the database server (I am now having 8 connections
from tracker 2 to the database server if I run netstat on the tracker,
but I see 68 connections from tracker2 to the database if i run it on
the database server.
If you can give me some tip about what might be the problem it will be
very useful to me!
Thanks!
Fernando
