Dirk Manske wrote:
> On Wednesday 05 January 2011 13:55:48 Johan van Selst wrote:
> > The attached patch adds an option to blacklist certain IP addresses for
> > a specific connection.
> When that randomly broken Server becomes really dead, then your testing
> will not notice this breakdown, but real users could notice that.

Yes, of course broken servers need to be fixed a.s.a.p.. But that is not
really a job for the HTTP client application. Note that an undesired
result need not even indicate a broken server - it can also be that the
server is just too slow for this client.

> I would test all servers directly (URL with IP, and host entry in the
> header). And to calculate the overall availabilty you could cummulate
> the single host test results.

I am confident that there is proper monitoring for the servers. And
that in theory it really should not matter which one sends an answer.
Unfortunately theory doesn't always works just as well in practice.
Good monitoring doesn't change the fact that there are sometimes servers
that my application should avoid.

Furthermore, at the moment DNS is completely handled internally by
libcurl - with features for DNS caching, asynchronous resolving etc.
I'm not really looking forward to moving this burden to the application
itself without good reason.

Regards,
Johan

-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html