1

I have a setup of 2 HAProxy servers using the KeepAlived process for failover, and each HAProxy is configured to load balance mysql connections to my MariaDB Galera cluster of 3 nodes. I'm trying to load test my environment, and before I even push traffic to one of my nodes, i'm looking at the netstat -an | grep 3306, and this is what I see on my first node:

ubuntu@mariadb1:~$ netstat -an |grep 3306
tcp    0      0 0.0.0.0:3306           0.0.0.0:*               LISTEN
tcp    0      0 10.60.61.21:3306       10.60.61.19:44563       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44521       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44536       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58622       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58637       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58649       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44533       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44557       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58616       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58628       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58619       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58646       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58625       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58631       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44551       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58652       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44524       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58658       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44545       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58655       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58640       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58634       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44539       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44554       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44527       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58661       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44518       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44548       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44560       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.19:44530       TIME_WAIT
tcp    0      0 10.60.61.21:3306       10.60.61.20:58643       TIME_WAIT

I want to know why the HAProxy's are filling up the connections on 3306 before I have even generated any traffic. I would have thought that the primary HAProxy would have to have at least one connection running so that it could pass traffic, but not both HAProxy's simultaneously, and not with this many concurrent connections. Please let me know what I need to look for to tune this, or just accept that this is normal behavior for a HA setup.

Here are the spec's on my mariaDB machines:

Ubuntu 12.04 8GB RAM 4 vCPU (each node is running in VMware)

If you need the /etc/mysql/my.cnf file please let me know, or the /etc/haproxy/haproxy.cnf file. Sorry if my post seems terribly formatted, not sure how to make this look all nice and clean like the rest of you guys.

Nathan C
  • 15,059
  • 4
  • 43
  • 62
Daniel Graves
  • 33
  • 5

1 Answers1

4

Haproxy does "heartbeats" to ensure the service is actually up. These can be quite frequent and should be to ensure quick failover.

All those connections are in a TIME_WAIT state so they're already closed and waiting to be cleaned up by the OS. They're nothing to be concerned about.

Nathan C
  • 15,059
  • 4
  • 43
  • 62
  • Thank you for the quick response and editing of my post, Nathan. I'll upvote this and mark as answer as soon as serverfault lets me. – Daniel Graves Jun 24 '14 at 15:17