[9fans] Plan 9 file server w/ indexed dirs and logging

[9fans] Plan 9 file server w/ indexed dirs and logging

[G.David]

Hey all!

I STILL haven't heard from anyone about licensing....

But I want to know how the work I've done thus far works,
so if you have a spare PC to use as a Plan9 file server
and want to have a go at it...  This fileserver has indexed
directories and logging.  [It also supports mirrored disks,
sort of.  It works but needs more attention.]  The logging comes
in two versions (compile flag) of full discrete transactions
and fileserver protection.  The one below only provides
fileserver protection.  The other one guarantees the results
of each successful 9P message.  Both are fully compatible with
the current Plan9 9P protocol.

Get ftp://ftp.dbSystems.com/pub/PLAN9/9pcfs.xor

Using the xor program from the updates on plan9.bell-labs.com
do "?.xor 9pcfs.xor /68020/9gnot > 9pcfs".  This is the same way
Bell Labs distributed dns.tar, so if it is good enough for them,
it's good enough for me.  (I had to do this since the file
server is on the CD.  In other words, you have to own the CD
to try this out, sorry.)

Follow the normal directions to get a Plan9 fileserver up, except:

The system will not boot on a config of a previous system
 (To protect *us*.  You will have to tell the system to destroy
 the drive contents, not I.)

The ipauth command is gone. (That is how I got the above.)

The "users default" only creates adm and none, so you must use
 user none on the initial load.

To enable logging on all filesystems, create a filsys called "log".
 (The system will not honor attaches to "log".)

For a fast temporary filesystem create a filsys called "nolog".
 If logging is enabled, "nolog" will be "reammed" on each boot.

The permissions are different on this filesystem as discussed on
 this list.

I think that is all you need to know to get this running.  You
may notice there is an "index" option to check, but it is not
enabled yet.  The idea is to "fix" the indexes.  Since I think
the whole idea of check is silly, (use the log luke), there are
only consistency checks for now.  If you do find yourself wishing
the index rebuild is there, you will have to clri the *parent*
directory and run check free to get to sanity again.

One other thing, I would recommend a small log so the search for
the end doesn't take too long.  Something like pw0.0.0.0.1 (Yeah
I fixed the partition syntax like the man page...)

I would really like to know if you have any problems with this.
Please email any panics and the like.

Have fun?!

David Butler
gdb@dbSystems.com

[9fans] Plan 9 file server w/ indexed dirs and logging

[G.David]

I wrote last time.

>
>so if you have a spare PC to use as a Plan9 file server
>and want to have a go at it...  This fileserver has indexed
>directories and logging.  [It also supports mirrored disks,
>sort of.  It works but needs more attention.]  The logging comes
>in two versions (compile flag) of full discrete transactions
>and fileserver protection.  The one below only provides
>fileserver protection.  The other one guarantees the results
>of each successful 9P message.  Both are fully compatible with
>the current Plan9 9P protocol.
>
>Get ftp://ftp.dbSystems.com/pub/PLAN9/9pcfs.xor
>
[snip]

Ok, this new one is even better!  I now use 16 ccbs on the 1542
and increased the number of log writting processes to 4.  Boy
does everything scream now...

David Butler
gdb@dbSystems.com

[9fans] Plan 9 file server w/ indexed dirs and logging

[G.David]

From: "G. David Butler" <gdb@dbSystems.com>
>
>so if you have a spare PC to use as a Plan9 file server
>and want to have a go at it...  This fileserver has indexed
>directories and logging.  [It also supports mirrored disks,
>sort of.  It works but needs more attention.]  The logging comes
>in two versions (compile flag) of full discrete transactions
>and fileserver protection.  The one below only provides
>fileserver protection.  The other one guarantees the results
>of each successful 9P message.  Both are fully compatible with
>the current Plan9 9P protocol.
>
>Get ftp://ftp.dbSystems.com/pub/PLAN9/9pcfs.xor
>
[snip]

I had a problem recognizing the end of the log during certain
crashs.  You see [take a deep breath], the log is circular and
so you have to find the last checkpoint and you have to make
sure you don't overwrite the last checkpoint and there are two
log writing processes using a log buffer and the changes in the
buffer cache can't be written till you know the changes in the
log buffer have been written...  You get the idea, it's hard.

There is now a new kernel that does a better job of using the log
after a crash.  Sorry.

>One other thing, I would recommend a small log so the search for
>the end doesn't take too long.  Something like pw0.0.0.0.1 (Yeah
>I fixed the partition syntax like the man page...)

Small is at least 1000 pages or about 5Mb.  If it is much smaller
you will start to see the message "sync: forced checkpoint" under
load and your I/O will slow down...

David Butler
gdb@dbSystems.com