NIS, automounting, Solaris and headaches

[Mark Komarinski]

On Mon, Sep 16, 2002 at 12:47:43PM -0400, Derek D. Martin wrote:
> At some point hitherto, Mark Komarinski hath spake thusly:
> > I figured it out.  They were direct maps, and I've started converting
> > to indirect.  This is one of the places where Linux is really lacking
> > in "prime time" capabilities.  NFS encryption support would be
> > nice too.
> 
> Well, as Tom B. pointed out, indirect maps are a bit more efficient
> than direct maps, and should be supported by any autofs
> implementation...  You might as well just convert everything over
> (assuming you don't have other constraints preventing you from doing
> so).
 
Indirect maps require you to do things like:

/scratch/host1
/scratch/host2
...and so on (assuming I wanted to share scratch space)

But the maps as they exist now are:

/host1/scratch
/host2/scratch

Which doesn't translate well into indirect mapping unless I have every
user update their scripts, etc.  With two sets of maps, I can migrate
people slowly.

> The Linux autofs team does (or at least did) eventually plan to add
> direct map support; however there's no great rush to do so for three
> main reasons:
> 
>   - direct maps are hard to implement (required significant kernel
>     changes, IIRC)

I don't get this.  It should be easier than indirect, as all you're
doing is mounting an NFS share onto a directory.  It's pretty
easily faked using a few shell scripts.  The only reason I'm going
through the effort is to minimize the amount of post-install
configuration for each machine.

>   - you can generally achieve the same effect with indirect maps

Unless you have a messed up map structure, as I have.

>   - indirect maps are better
> 
> I'd personally switch over entirely to indirect maps, but be sure to
> document WHY directly in the map files.  This will help future
> administrators who've been accustomed to using direct maps to not make
> the mistake of trying to switch back.

The researchers used to be the admins of the sytem.

-Mark