There is a CORBA connection between 2 HP-UX 11.11i hosts.
Then the LAN of the 2nd host is pulled.
On the 1st host all connections disappear, as expected.
But on the 2nd host all connections still are present, as established.
With lsof one can see that the CORBA processes are still connected to those ports.
My questions are about the port management.
Isn't this a HP-UX core issue ?
Aren't there checks that ensure that a port only can show established, if the other side shows the same ?
Can it be that an application keeps ports established, whilst on interface level this is not possible ?
It appears to be possible, but I would assume it to be the other way around.
I would assume that netstat would show no connection or perhaps time_wait.
That the application thinks the connection is still established, is the problem of the application.
I would expect that on interface level I get the valid port information.
Ive seen SUNs and AIXs acting the same way... So I doubt it being an HP-UX core issue, I always understood "whoever (application) opens a connection is responsible for closing it" but you do point out the fact "what about network partial failure?". I have no answer just like you but hope someone can enlighten us...
Thanks for replying.
It is always difficult to use the right technical terms and phrases, so my question may have come across a bit unclear.
First of all, I think it could be a general TCP/IP question, not really related to HP-UX.
It is just that it happens between 2 HP-UX boxes.
What I meant to say about it being a HP-UX core issue is that I expected the netstat information to come from the HP OS (for example the LAN adapter or so) and not from the applications that use ports/sockets.
Probably it is better to talk in TCP/IP terms, but one should know how
The assumption "whoever (application) opens a connection is responsible for closing it" makes sense, but in case of my issue, it forces me to take
into account that the netstat information is "unreliable".
Or better said: I have to keep in mind that this output is based on the information provided by the applications. A mind-shift.
In my case the LAN cable is pulled, so the first layer in the TCP/IP model, the link layer, is not present any more.
One box did the expected and released all related ports/sockets.
The other box is still pretending all is well.
As long as the application keeps the sockets in the established state, netstat will report them as established.
Which makes sense actually, while writing this, in principle the applications (application processes) are the best source for this, as they rely on this.
Anyway, bit by bit it is getting clearer, although I do not understand why the 2nd box is keeping the connections established.
I will read some TCP/IP information now, always useful
