topic Reset / reclaim port in Operating System - HP-UX

Reset / reclaim port

yulianto piyut — Wed, 29 Apr 2009 08:06:56 GMT

all,

I have an application running on my server rp4440, run hp-ux 11.11 with port 5501. The application is become hang and cannot be shutdown and kill.

# netstat -an |grep 5501
tcp 0 0 10.2.245.13.63151 10.2.224.49.5501 FIN_WAIT_2
55013540 stream 0 0 0 5b5de9c0 0 0
550139c0 stream 0 0 0 54e14680 0 0
55013b40 stream 0 0 0 54e14600 0 0 /tmp/.AgentSockets/A
# lsof -i:5501
#

how to reset this port ? should i reboot my server ?

Re: Reset / reclaim port

V. Nyga — Wed, 29 Apr 2009 08:52:38 GMT

Hi,

reboot is always a clean solution.
You also can try to change boot level to 'init 2' or 'init 1', then go back to 'init 3'.

HTH
Volkmar

Re: Reset / reclaim port

yulianto piyut — Wed, 29 Apr 2009 08:57:56 GMT

oo...only reboot server to reset the port ? hmm...not good for my production server.

Re: Reset / reclaim port

Matti_Kurkela — Wed, 29 Apr 2009 10:01:07 GMT

Your socket is in FIN_WAIT_2 state and the process associated with the socket has already died. This means the socket is in "detached" state.

With default settings, the solution should be simple: just wait 2 minutes. The system should automatically remove this socket in that time.

This behaviour is controlled by the ndd parameter "tcp_keepalive_detached_interval". To check it, run:

ndd -get /dev/tcp tcp_keepalive_detached_interval

The default value for this parameter is 120000, which means 120 seconds (=2 minutes).

If the socket does not go away and/or your tcp_keepalive_detached_interval is currently set to an unreasonably long time, you might use the attached Perl script. It will use an *unsupported* ndd parameter "tcp_discon_by_addr" to force disconnect *all* connections that are currently in state FIN_WAIT_2. It was originally created for HP-UX 11.00, but should work on 11.11 too.

MK

Re: Reset / reclaim port

yulianto piyut — Wed, 29 Apr 2009 10:17:56 GMT

hi matti, how to use tcp_discon_by_addr?
the port state of 5501 still FIN_WAIT_2. it's happened for about 1 month ago.

Re: Reset / reclaim port

Matti_Kurkela — Wed, 29 Apr 2009 10:42:14 GMT

The easiest way would be to just run the script I attached: it does not require any parameters.

To use tcp_discon_by_addr directly, you would have to convert the local IP address, local port number, remote IP address and remote port number to hexadecimal and concatenate all of them together in this order. (always 2 hex digits for each part of an IP address, and always 4 hex digits for the port number; add leading zeroes if necessary)

For your specific case, the command line would be:

ndd -set /dev/tcp tcp_discon_by_addr 0a02f50df6af0a02e031157d

The conversion:
10.2.245.13 = 0a 02 f5 0d
63151 = f6af
10.2.224.49 = 0a 02 e0 31
5501 = 157d

(I was hoping I would not have to do this conversion by hand by providing the script. Oh well, slow day... :-)

MK

Re: Reset / reclaim port

yulianto piyut — Thu, 30 Apr 2009 06:25:01 GMT

hi, after i run script in your attachment:

# netstat -an |grep 5501
tcp 0 0 10.2.245.13.1556 10.2.245.13.55010 ESTABLISHED
tcp 0 0 10.2.245.13.55010 10.2.245.13.1556 ESTABLISHED
550139c0 stream 0 0 0 54e14680 0 0
55013b40 stream 0 0 0 54e14600 0 0 /tmp/.AgentSockets/A

hmm...i'm confuse, how to use port 5501 again ?

Re: Reset / reclaim port

V. Nyga — Thu, 30 Apr 2009 07:42:00 GMT

Hi,

not all ports are used every time.
The port will be used again after restart of your application.

Volkmar

Re: Reset / reclaim port

yulianto piyut — Thu, 30 Apr 2009 07:54:58 GMT

hi nyga,

the applicaiton cannot be killed or shutdown and port 5501 is used for it.

Re: Reset / reclaim port

V. Nyga — Thu, 30 Apr 2009 08:18:57 GMT

Hi again,

yes, but you says it's hanging ....
so it can't (and don't need to) use the port now.
It seems like you have to investigate how to stop the application.
Have you tried 'kill -9 '?
Any error message?

What kind of application?
Have you tried to kill the parent process?

Volkmar

Re: Reset / reclaim port

yulianto piyut — Thu, 30 Apr 2009 08:36:04 GMT

I have tried to kill -9 (PID) and parent process, but the application daemon still run. The application is replication software, Repx, called topio software from netapp. Netapp engineer said this application use port 5501 and this port is going down.

Re: Reset / reclaim port

V. Nyga — Thu, 30 Apr 2009 08:44:41 GMT

Well, Netapp should be able to tell you how to stop it :-)
Is there a replication hardware involved?
Can you switch it off?
Or:
Can you restart the application?

Else I think your only option is a restart of your server during a maintenance window, or at least change to init 2, but this if of course similar to a reboot, only shorter.

HTH
Volkmar

Re: Reset / reclaim port

yulianto piyut — Thu, 30 Apr 2009 09:14:12 GMT

hehehe...no hardware involved in this application. I think, application going to hang and should reboot my server. I can not kill, restart, & shutdown this application and in hp-ux, there is no command to kill zombie process. in sun, there is command "preape" to kill zombie process. isn't it ?

Re: Reset / reclaim port

Dennis Handly — Fri, 01 May 2009 03:39:16 GMT

>there is no command to kill zombie process.

You can kill zombies by killing the zombie master.

>I have tried to kill -9 (PID) and parent process,

Then this isn't a zombie, it is hung.