Number of file descriptors in creases

Sai-Lai Lo S.Lo@orl.co.uk
Fri, 13 Mar 1998 16:51:34 GMT


>>>>> Jan Lessner writes:

> I'm currently walking through my code with purify to see wether there's
> anything dangorous happening. The most serious thing I doscovered was an
> increasing number of file descriptors being in use. They are allocated
> when ever a new connection is established and it seems they don't get
> cleaned up automatically.
> Is that me who is using omniORB wrong? Is there a way of cleaning up
> stale file descriptors from the API (either implcite or explicite). I
> guess there is because I would expect a server to run into problems
> after a while when the operating system's limit of FDs is reached. I
> actually have server crashes which may be caused by that. However,
> omniNames absolutely never crashes at all though being contacted quite
> often.
> I'm still using version 2.2.

In omniORB 2.2.0, connections are not shutdown until process exit.

Since omniORB 2.4.0, connections are shutdown after they are unused for a
period of time. There is API to control this period and whether outgoing
and/or incoming connections are to be monitored. The details are documented
in the User Guide. If you have to interoperate with other ORBs, you may
want to disable the scan for idle *incoming* because most of the other ORBs
tend to throw COMM_FAILURE when they see this, rather than quietly retry a
request (this is the compliant behaviour).


Sai-Lai