[distcc] Remote Fallback

[Martin Pool]

On 20 May 2003, Thomas Walker <Thomas.Walker at morganstanley.com> wrote:
> What is the working status on implementing remote fallbacks?  i.e. if I
> fail to contact a remote host, have it try another remote host rather
> than fallback to localhost or fail (as per DISTCC_FALLBACK).  I see that
> there is space in the 2.3 code for this to be implemented but that its
> masked at the moment.  Is this something that anyone has bothered to
> implement that just hasn't made it into the mainstream release yet or
> has it not been done?
> I'm in need of this functionality and am willing to do it myself but
> don't want to duplicate effort.

Can you tell us more about why it's needed for your situation?  

I'm guessing it's because you have a lot of servers, and machines in
the list that are sometimes inaccessible.

It should be straightforward to implement.  The only subtlety is
making sure that it tries each remote host only once.  

It might in fact be better after trying a few the remote hosts to just
give up and assume that the client is in fact disconnected from the
network.  Trying to connect to unavailable machines has a cost.

I'll add intelligence about remembering unusable hosts shortly and
that should mitigate some problems.

> I could add options to DISTCC_FALLBACK (but leave it backwards
> compatable with old options) or throw in another switch, any
> preferences? (distcc seems to lean towards a whole bunch of boolean
> switches rather than ternary or more way options).

If it's implemented properly I think it should be OK to leave it
always on.

-- 
Martin 

linux.conf.au 2004: Adelaide, Australia         http://lca2004.linux.org.au/