[9fans] farewell to /sys/src/fs and IL

[9fans] farewell to /sys/src/fs and IL

[geoff]

In the interest of minimising maintenance, we're going to stop
maintaining Ken's fs and IL entirely.  We thank them (and Ken) for
many years of faithful service.  We're now running fossil with nventi
on our main file servers.  cwfs(4) now exists, so old juke boxes are
still accessible.

This does not mean that you need to stop running Ken's fs, just that
we're going to remove /sys/src/fs and /sys/src/9/ip/il.c from the
distribution, and that a replica/pull in the near future may remove
them from your system, so you may want to rename them or make copies
beforehand, or copy them out of the dump afterwards.

If anybody wants to maintain Ken's fs and keep copies in
sources/contrib (for example), that's fine.  I've put a snapshot of
the current /sys/src/fs in /n/sources/extra/kenfs.tgz.

Re: [9fans] farewell to /sys/src/fs and IL

[erik quanstrom]

On Thu Aug 30 18:27:44 EDT 2007, geoff@plan9.bell-labs.com wrote:
> In the interest of minimising maintenance, we're going to stop
> maintaining Ken's fs and IL entirely.  We thank them (and Ken) for
> many years of faithful service.  We're now running fossil with nventi
> on our main file servers.  cwfs(4) now exists, so old juke boxes are
> still accessible.

i hope the decision to remove il from the distribution is revisited.
without il, anyone still using ken's fileserver will not be able to run
a kernel from sources.

- erik

Re: [9fans] farewell to /sys/src/fs and IL

[Skip Tavakkolian]

thanks for pointing this out.  it would make things more
difficult here.

anyone have any notes on migrating from kenfs to fossil/venti?

> On Thu Aug 30 18:27:44 EDT 2007, geoff@plan9.bell-labs.com wrote:
>> In the interest of minimising maintenance, we're going to stop
>> maintaining Ken's fs and IL entirely.  We thank them (and Ken) for
>> many years of faithful service.  We're now running fossil with nventi
>> on our main file servers.  cwfs(4) now exists, so old juke boxes are
>> still accessible.
> 
> i hope the decision to remove il from the distribution is revisited.
> without il, anyone still using ken's fileserver will not be able to run
> a kernel from sources.
> 
> - erik

Re: [9fans] farewell to /sys/src/fs and IL

[Francisco J Ballesteros]

Time ago we did it the hard way, just go day from day and archive on venti.
We use vacfs to bind old worm dumps below the actual fossil dump on venti.
But there's probably a better way.

On 8/31/07, Skip Tavakkolian <9nut@9netics.com> wrote:
> thanks for pointing this out.  it would make things more
> difficult here.
>
> anyone have any notes on migrating from kenfs to fossil/venti?
>
> > On Thu Aug 30 18:27:44 EDT 2007, geoff@plan9.bell-labs.com wrote:
> >> In the interest of minimising maintenance, we're going to stop
> >> maintaining Ken's fs and IL entirely.  We thank them (and Ken) for
> >> many years of faithful service.  We're now running fossil with nventi
> >> on our main file servers.  cwfs(4) now exists, so old juke boxes are
> >> still accessible.
> >
> > i hope the decision to remove il from the distribution is revisited.
> > without il, anyone still using ken's fileserver will not be able to run
> > a kernel from sources.
> >
> > - erik
>
>

Re: [9fans] farewell to /sys/src/fs and IL

[Francisco J Ballesteros]

FTS, did not have cwfs those days. Today we'd probably just copy the storage
to new disks and then mount it with cwfs.

On 8/31/07, Francisco J Ballesteros <nemo@lsub.org> wrote:
> Time ago we did it the hard way, just go day from day and archive on venti.
> We use vacfs to bind old worm dumps below the actual fossil dump on venti.
> But there's probably a better way.
>
> On 8/31/07, Skip Tavakkolian <9nut@9netics.com> wrote:
> > thanks for pointing this out.  it would make things more
> > difficult here.
> >
> > anyone have any notes on migrating from kenfs to fossil/venti?
> >
> > > On Thu Aug 30 18:27:44 EDT 2007, geoff@plan9.bell-labs.com wrote:
> > >> In the interest of minimising maintenance, we're going to stop
> > >> maintaining Ken's fs and IL entirely.  We thank them (and Ken) for
> > >> many years of faithful service.  We're now running fossil with nventi
> > >> on our main file servers.  cwfs(4) now exists, so old juke boxes are
> > >> still accessible.
> > >
> > > i hope the decision to remove il from the distribution is revisited.
> > > without il, anyone still using ken's fileserver will not be able to run
> > > a kernel from sources.
> > >
> > > - erik
> >
> >
>

Re: [9fans] farewell to /sys/src/fs and IL

[Eric Van Hensbergen]

On 8/31/07, Skip Tavakkolian <9nut@9netics.com> wrote:
> thanks for pointing this out.  it would make things more
> difficult here.
>
> anyone have any notes on migrating from kenfs to fossil/venti?
>

in a semi-related question, is their a published procedure for
migrating to nventi, or is it painless?

        -eric

Re: [9fans] farewell to /sys/src/fs and IL

[erik quanstrom]

i wrote a paper on how i ported history from ken's fs to ken's fs.
i think the technique would work for porting history from any fs
to any fs.  /n/sources/contrib/quanstro/history.ms

however, we have a working kenfs.  kenfs has been stable for a
very long time.  venti+fossil has been problematic for us.  i
would perfer not to switch.

- erik

Re: [9fans] farewell to /sys/src/fs and IL

[erik quanstrom]

by the way, i am actively working on ken's fs.  i have added
an aoe initiator (big suprise), an intel ahci driver and a few
other things.  if you're interested, drop me a note.

- erik

Re: [9fans] farewell to /sys/src/fs and IL

[geoff]

If you're running ken's fileserver, I trust that you're compiling your
own kernels.

All you have to add to cpu kernels from sources to reinstate IL are
the current /sys/src/9/ip/il.c and add "il" to the kernel
configuration files (or keep your current ones).  replica/pull will
remove il.c once, but thereafter pull should leave it alone (if it
doesn't, I'll manually remove il.c from plan9.db on sources, which
will force pull to leave it alone).  Likewise for /sys/src/fs (if
that's the `kernel from sources' you're referring to).

I'll have to do something similar at home, at least for a little
while.  It's a minor one-time nuisance.

Re: [9fans] farewell to /sys/src/fs and IL

[erik quanstrom]

> If you're running ken's fileserver, I trust that you're compiling your
> own kernels.
> 
> All you have to add to cpu kernels from sources to reinstate IL are
> the current /sys/src/9/ip/il.c and add "il" to the kernel
> configuration files (or keep your current ones).  replica/pull will
> remove il.c once, but thereafter pull should leave it alone (if it
> doesn't, I'll manually remove il.c from plan9.db on sources, which
> will force pull to leave it alone).  Likewise for /sys/src/fs (if
> that's the `kernel from sources' you're referring to).

we run kernels directly from sources on quite a number of machines.

one advantage of this is that if there's a problem with the kernel,
it's clear that local modifications are not to blame.

i can't see the harm in leaving il.c in the kernel.  it shouldn't require
much maintence at all.  i volunteer to do any work required to keep
it up-to-date.

- erik

Re: [9fans] farewell to /sys/src/fs and IL

[geoff]

I'll see if there's anything of general use in my notes on migration.

As nemo says, if you're running kenfs now, it's probably easiest to
use cwfs(4) to keep using your existing file systems, with a small
fossil for booting.

Migrating from venti to nventi is almost painless and maybe we can
make it completely so.  You'll likely want to at least add a 512MB
bloom filter to improve performance, but nventi copes with venti
arenas just fine.  The only incompatibility I can think of is that
nventi/venti doesn't accept all the options that venti/venti does, and
refuses to start if you give it unknown options.  It might be better
to complain on stderr and in syslog and ignore venti options that
nventi doesn't currently recognise.

Re: [9fans] farewell to /sys/src/fs and IL

[Steve Simon]

> anyone have any notes on migrating from kenfs to fossil/venti?

I copied some old tar and zip archives from tapes
and cds a while back thusly:

	cpu% @{cd /n/cdrom/fred/1-Apr-2001 && tar cf /fd/1 .} | @{cd /usr/steve/tmp/2001/0104/usr/fred && tar xTf /fd/0}
	cpu% con -l /srv/fscons
	prompt: fsys main snap -a -s /active/usr/steve/tmp/2002/0104 -d /archive/2002/0104

This is from memory but I think its right.

Though it would probably be better to do this using mkfs|extfs rather
than tar|tar to preserve as much plan9 metadata as possible. Note you can
only transfer as much data as you have space in your fossil store.

<wistful>

I have always planed to add an option to fossil to allow you to install
a given venti score in a named directory, this would allow you to just
vac(1) up a directory hierarchy and then install its score using somthing like

	
	cpu% vac /n/cdrom/fred/1-Apr-2001 
	vac:64daefaecc4df4b5cb48a368b361ef56012a4f46
	cpu% con -l /srv/fscons
	prompt: fsys main snap -a -v vac:64daefaecc4df4b5cb48a368b361ef56012a4f46 -d /archive/2002/0104

This would have the advantage of collecting all metadata correctly and allowing
archives bigger than your fossil disk to be installed.

Unfortunately though I had a look at this I never understood the
fossil code well enough to work out how to do it.

</wistful>

-Steve