[omniORB] Hanging connections

[Duncan Grisby]

On Wednesday 10 May, Serguei Kolos wrote:

> emon_push_sampler.trace - is the trace of the app, which accumulates
> hanging connections
> 
> emon_task.trace -  trace shows one last invocation from 5 of the
> emon_task app. This app
>                                subscribes to the emon_push_sampler,
> which starts pushing events to it.
>                                Events are stored in the buffer, from
> which emon_task read 1 event and
>                                exits.

OK, so the thing that's growing file descriptors is acting in the role
of a client when it pushes events. All the events are sent oneway, so it
never tries to read from the connections it uses. Therefore, when the
clients close the connections and go away, the closure is not noticed.
It's not until the connections are considered idle and timed out that
they are closed.

I think that if you reduce the value of the outConScanPeriod parameter
from 120 seconds to a much smaller number, you'll find that the
connections are closed much sooner.

Cheers,

Duncan.

-- 
 -- Duncan Grisby         --
  -- duncan at grisby.org     --
   -- http://www.grisby.org --