Re: [HTCondor-users] Do you NEED to share condor_config across multiple versions of HTCondor?

[Brian Bockelman <bbockelm@xxxxxxxxxxx>]

On Jan 7, 2014, at 3:16 PM, John (TJ) Knoeller <johnkn@xxxxxxxxxxx> wrote:

> If the HTCondor team were to add new features to the condor_config file that would
> cause older versions of HTCondor to regard the condor_config as invalid -
> Would this prevent you from using the new feature?
> 
> Say, for instance that you could have lines in the config file that were neither comments
> nor valid  key = value  statements; and that these new statements were necessary
> for configuring a STARTD to use GPUs.
> 
> Would the fact that HTCondor 8.0.x and earlier would core dump when trying to read
> this configuration prevent you from using these new statements to configure
> other machines running HTCondor 8.1.x that could read and use this configuration?
> 
> We generally try and preserve backward compatibility, but there are real advantages to
> new syntax in the config file for solving certain STARTD configuration problems.
> 
> So the question is - is backward compatibility so important to you that it would prevent
> you from using 8.1 features on 8.1 machines because your installation needs for those
> config files to look valid to older HTCondor versions.
> 

$0.02 from a site that is admittedly "aggressive" in these matters.  The following would be acceptable to me:

1) HTCondor dev series N+1 introduces new feature X
2) Feature X can only be enabled via config syntax that can be understood by N+1 and newer.

I would have a problem if the new syntax prevented me from accessing features in prior versions.  That is, I would always expect that doing "yum upgrade" on HTCondor would have a reasonable chance of success.

In general, we use some other templating language to generate our config files across multiple machines.  Having syntax specific to certain machines (or, I guess, certain versions) doesn't seem to be a large problem.

Brian