[omniORB] Abrupt unbinding of OMNI name servers

[vishnu vardhan]

Hi All,

We have a scenario explained below with the name server.

We have a product built upon the omni name servers as the middleware
connecting between different processes. All of the processes are getting
binded to the omniNames and uses these sessions for communicating between
them. In some of such setup, we are frequently observing that the sessions
are getting unbinded abruptly even though the associated processes are still
running. We have around 8 application processes and all of them are getting
unbinded abruptly at the same time...!!!Also, we are able to see the
unbinding log entries are the omniname.log.bak.

a) unbind call is made from the process registered to the name server is
made only when it comes down. In this case, even though the processes are
running, we are seeing that the unbind happens abruptly.
b) Another possibility we thought of is the use of nameclt to unbind the
sessions. We are sure that this not being done.
c) Another possibility we heard of is the file system where the log files
are situated become full. There are no such problems here.

Q1) What are the possible reasons for such a behaviour? We have seen the
similar scenario in HP-UX and Linux Os's using omniORB. Apart from the
processes getting binded to name server AND the nameclt tool, nny other
tools/processes can do this unbinding?

Q2) Will any system resource like memory/network/disks problem result in
this type of this issue?

Q3) How do we identify which application/process is doing this unbinding
activity if its not happening abruptly?

Q4) Did anybody observe any such issues even in older versions of omni?

It will be great if somebody can help us in this scenario as we are in dark
identifying the root-cause.

[Duncan Grisby]

On Friday 2 May, "vishnu vardhan" wrote:

[...]

Post by vishnu vardhan
a) unbind call is made from the process registered to the name server is
made only when it comes down. In this case, even though the processes are
running, we are seeing that the unbind happens abruptly.
b) Another possibility we thought of is the use of nameclt to unbind the
sessions. We are sure that this not being done.
c) Another possibility we heard of is the file system where the log files
are situated become full. There are no such problems here.
Q1) What are the possible reasons for such a behaviour? We have seen the
similar scenario in HP-UX and Linux Os's using omniORB. Apart from the
processes getting binded to name server AND the nameclt tool, nny other
tools/processes can do this unbinding?

Any process is in a position to call unbind on the naming service and
remove the bindings. There is nothing in omniNames that will
automatically remove a binding, so something in your environment is
doing it.

Post by vishnu vardhan
Q2) Will any system resource like memory/network/disks problem result in
this type of this issue?

No.

Post by vishnu vardhan
Q3) How do we identify which application/process is doing this unbinding
activity if its not happening abruptly?

Run omniNames with command line options

-ORBtraceLevel 25 -ORBtraceInvocations 1 -ORBtraceThreadId 1 -ORBtraceTime 1

That will show you loads of data about what's going on, including
showing the incoming address and port of calls to unbind. That should
help you track it down.

Cheers,

Duncan.