#6 Inode-tracing patch now included in kernel: auto-triggering?

[Anonymous]

Starting from kernel 2.6.9 (or possibly earlier) the
inode-tracing patch has been included in the standard
kernel tree.

It is compiled in by default and is accessible by
sysctl vm.block_dump (instead of fs.block_dump as in
the former patch.) I think it should be mentioned in
the documentation.

I also think it would be a great feature to have
noflushd switch it on/off automatically, ergo this
feature request.

When noflushd successfully spins a disk down it would
switch the tracing on (via sysctl); when it spins a
disk up it would switch the tracing off. This way we
would have a very useful log of what made the disk spin
up and when, while avoiding to clutter the logs with
thousands of useless entries while the disk is up.

Best regards
Tobia Conforto <tobia.conforto@linux.it>

PS: Thanks for making/mantaining noflushd; it's a very
useful piece of software, not only for notebooks!

[Daniel Kobras]

Logged In: YES
user_id=7832

[It seems I never got the usual e-mail when this feature
request was added, so I just noticed it. Sorry for the long
delay.]

From a quick look, block_dump can only be switched on
globally, but not per device. Therefore, I'm not very
enthusiastic about turning it on automatically from noflushd
on each spindown--if there's more than one block device on
the system, logs will go crazy. Or am I misunderstanding how
things work? That said, I'm all for adding it to the
documentation, of course.

[Nobody/Anonymous]

Logged In: NO

> I'm not very enthusiastic about turning it on automatically
> from noflushd on each spindown--if there's more than one
> block device on the system, logs will go crazy.

Yes, but:
1. This feature would need to be explicitly activated by
means of a commandline switch or such. So it would only be
used by people who are trying to track down whatever program
is causing unneeded spinups. Having noflushd automatically
turn the feature off when the given hard disk goes on is
better than doing it by hand, independently of any logging
problem.

2. Most people who want to track down things like these are
on a notebook or embedded/compact system and most of those
only have one disk :-)

Proposed interface (to be amended with a generic description
of what block_dump is):

-m | -mm | -m <disk>
Monitor hard disk access. If -m is specified, noflushd will
turn block_dump on when ALL hard disk being handled spin
down and it wil turn it off when ANY of them spin up. If -mm
is specified, block_dump is turned on when ANY hard disk
spin down and it's turned off when ALL disks spin up.
WARNING: -mm is not logfile-size friendly! The last format
ties this behaviour to one specific disk.