Re: Providing temporary storage space to a VM: qcow, nfs etc

[Mayuresh <mayuresh%acm.org@localhost>]

On Fri, Jun 26, 2020 at 11:37:21PM +0100, Steve Kirk wrote:
> Oops, bit slow replying here...

Never mind as I am even slower in implementing! Finding difficult to make
time..

> If you're still looking at the NFS-based solution, the first thing I'd
> suggest for improving perfomance is mounting it it with the noatime
> option on the client side. That can improve things when you're dealing
> with large numbers of files and directories.

Yes will do that.

What I am thinking is to use NFS/sshfs for

- pkgsrc (shared with the host system and readonly for both)

- distfiles (again shared with the host and its cleanup/housekeeping is
  common for both host and guest)

- packages (keeping on host makes them easy to distribute)

I am slightly inclined to sshfs as

- The access to above distfiles/packages is infrequent and access to
  pkgsrc is not as high volume as pkg or work. So speed should not be a
  worry.

- NFS is faster but I'll have to worry about additional open ports and
  their security to make them accessible only to the localhost. I think
  the services involved themselves do not have the setting for this. So
  have to get into configuring npf etc. Also it's not 1 daemon. There are
  multiple of them. ssh is already open and has some protection with
  blacklistd.  But if someone could give some simple tricks for this I'd
  go with NFS.

For pkg and work I'll use 2 separate raw images. I'll be recycling work
image often.

Why switching to raw from qcow2:

- The way qcow2 fills up is unintuitive to me. I hope raw doesn't have
  that problem, but not sure.
 
- I read raw would perform slightly better than qcow2. For workobj area
  even slight gain means a lot.

- I hope I can loop mount raw image if needed occasionally

- qcow2's advantage of growing-as-needed is no more attractive as I'll be
  recycling work image almost after every build session. Besides it was
  any way filling up fast.

Still looking for systematic ways to keep pkg size in check. I can keep
deleting packages after build, though I have to protect some of them etc.
But the main problem is, if a package is needed again as a dependency
pkgsrc would start building again even if it is present in packages. If
there is a way to make pkgsrc prefer to install it from package (under
some criteria such as version hasn't changed etc.) that would be a great
way to conserve space on build servers.

Mayuresh