Location service

[timruijs]

In order to simplify maintenance of our applications I would like to remove the portnumbers of our servants. To this extend I use the location service scheme.

Our applications are distributed. Each host has its own locator. Services running on that host register at that locator .
Whenever a service requires access to another service at a different host, I provide the proxy of the locator at the destination host in the proxy for the intended object (using ice_locator).
This way objects can subscribe to events of any other object in the network. Note: our own publish/subscibe mechanism, not Ice's.

And this all works great. However, when a host is disconnected from the network (either cable disconnect, router crash, etc), the local requests are delayed for seconds. Each subscriber has its own thread with queue, but still local requests are delayed.

Reading the doc/forums do not shed much light on what happens when connections fail, especially with respect to locator functionality.

Before we try to minimise our system to get a usable network trace, perhaps you can give a few pointers.

[benoit]

Hi,

It's not clear to me where these delays occur. In theory, local connections using the loopback interface shouldn't be affected by network issues. However, without more information, it's difficult to help. Adding tracing to try to figure out where the delays occur will hopefully help. If you need further help with this, please specify which Ice version and operation system you're using.

Cheers,
Benoit.

[slypete]

Perhaps your application is trying to resolve a domain?

Pete

[timruijs]

Thanx for your quick response.

We have found the culprit! Our Name Server Cache Daemon (nscd) wasn't running . Something went wrong in generating our system configs.
We did not investigate this earlier because for two (known) missing hosts the DNS exceptions were actually logged (while the network is up). When the network was disconnected no additional errors were logged for any of the other hosts putting us off track.
We will look into the missing logs, but for now everything works great (even when the network is down ).

So, thanx for your time.