A zombie session makes the server response bad. Is this a bug?

James_Lau · November 6, 2007, 2:01am

Hi all,

I made a stress test using the OpenfireClusterStressTest.zip. The openfire version chosen was 3.3.3. The server was P4 2.5G with 512MB mem, running Fedora 7. Testing db was postgresql, which installed on another machine. No cluster was configured. Only one client machine, also P4 2.5G with 512MB mem, running CentOS.

20k connections were established to the server, and the cpu usage showed about 60_{70%. While I shutdown the test client, the sessions decreased respectively, and the cpu usage went down to 0}1% soon. All the tcp connections were shutdown both the server and the client finally, but there’s a xmpp session still available on the admin console even after several hours. The cpu usage was still 0~1%, but the server responsed really slow. I think this session is a zombie. So I deleted this session, and the server became fine.

Why there’s a zombie xmpp session while all tcp connections were closed, and maked the system slow? Is ths a bug?

James_Lau · November 6, 2007, 3:36am

I’m sorry. I made a mistake. The cpu usage was not went down to 0~1%. It seemed like:

Cpu(s): 0.2%us, 3.1%sy, 0.0%ni, 0.0%id, 96.7%wa, 0.0%hi, 0.0%si, 0.0%st

why so many cpu time wasted on iowait?

LG1 · November 6, 2007, 8:41pm

Hi James,

I guess that your server was paging, “vmstat 1” could show you that this is the case. “cat /proc/meminfo” should display very low values for MemFree, Buffers and Cached.

Did you set Xmx to something larger than 350 MB on your 500 MB server?

LG