On 12/12/11 12:15, Daniel Troeder wrote:

Hello

I have an ext4-filesystem that contains /usr/src, the /usr/portage and
/var/cache/edb. It previously also contained /var/db/pkg, but I had to
move that some weeks ago, because the fs was "full". Now it's "full"
again, though it has free blocks. But no inodes are left:

$ fsck -vf /dev/sda5
[..]
655360 inodes used (100.00%)
[..]

$ find /gentoo -xdev | wc -l
655338

That's really disappointing. I was using reiser3fs and XFS before, and
they didn't have that kind of limitation... Uhm... not meant as a rant -
I like ext4 - that's why I'm moving (almost?) everything to it...

Is there any way to raise the number of inodes without using
$ mkfs.ext4 -N BIGNUM

Thank you,
Daniel


That is scary. I just install new HD with 2TB capacity and ext4 that is 2% full and:
$ find /home/joseph/ -xdev | wc -l
shows: 169977 that is 26% full.


So will run out of inodes before I run out of hard disk space :-/ that is not good.
If other filesystems don't have these kind of limitation I'll be switching.

--
Joseph

Quick googling around indicates that JFS, or XFS don't have these limitation.
-----quote--------
Many computer programs used by system administrators in UNIX operating systems often designate files with inode numbers. Examples include popular disk
integrity checking utilities such as the fsck or pfiles. Thus, the need naturally arises to translate inode numbers to file pathnames and vice versa. This can
be accomplished using the file finding utility find with the -inum option, or the ls command with the proper option (-i on POSIX compliant platforms).


It is possible to use up a device's set of inodes. When this happens, new files cannot be created on the device, even though there may be free space
available. For example, a mail server may have many small files that don't fill up the disk, but use many inodes to point to the numerous files.


Filesystems (such as JFS, or XFS) escape this limitation with extents and/or dynamic inode allocation, which can 'grow' the filesystem and/or increase the
number of inodes.

----end quote------

--
Joseph

On Dec 12, 2011 9:39 PM, "Joseph" <syscon780@gmail.com> wrote:

>

> Quick googling around indicates that JFS, or XFS don't have these limitation.

> -----quote--------

> Many computer programs used by system administrators in UNIX operating systems often designate files with inode numbers. Examples include popular disk integrity checking utilities such as the fsck or pfiles. Thus, the need naturally arises to translate inode numbers to file pathnames and vice versa. This can be accomplished using the file finding utility find with the -inum option, or the ls command with the proper option (-i on POSIX compliant platforms).


>

> It is possible to use up a device's set of inodes. When this happens, new files cannot be created on the device, even though there may be free space available. For example, a mail server may have many small files that don't fill up the disk, but use many inodes to point to the numerous files.


>

> Filesystems (such as JFS, or XFS) escape this limitation with extents and/or dynamic inode allocation, which can 'grow' the filesystem and/or increase the number of inodes.

> ----end quote------

>


ReiserFS also doesn't have problems with inodes because everything are kept in b*trees that can keep growing indefinitely.


In fact, I think I read somewhere that ReiserFS is perfect for /var/tmp and /usr/src due to the amount of small files in those directories.


Rgds,

Joseph writes:

> On 12/12/11 12:15, Daniel Troeder wrote:
> >Hello
> >
> >I have an ext4-filesystem that contains /usr/src, the /usr/portage and
> >/var/cache/edb. It previously also contained /var/db/pkg, but I had to
> >move that some weeks ago, because the fs was "full". Now it's "full"
> >again, though it has free blocks. But no inodes are left:
> >
> >$ fsck -vf /dev/sda5
> >[..]
> > 655360 inodes used (100.00%)
> >[..]
> >
> >$ find /gentoo -xdev | wc -l
> >655338
> >
> >That's really disappointing. I was using reiser3fs and XFS before, and
> >they didn't have that kind of limitation... Uhm... not meant as a rant
> >- I like ext4 - that's why I'm moving (almost?) everything to it...
> >
> >Is there any way to raise the number of inodes without using
> >$ mkfs.ext4 -N BIGNUM
> >
> >Thank you,
> >Daniel
>
> That is scary. I just install new HD with 2TB capacity and ext4 that is
> 2% full and:
> $ find /home/joseph/ -xdev | wc -l
> shows: 169977 that is 26% full.

No, that is 26% of the number of total inodes _Daniel_ has on his small
partition. Yours is bigger, so you have more inodes. My largest partition
has 724G, and 46 million inodes. Use df -i to see how many you have.

> So will run out of inodes before I run out of hard disk space :-/ that
> is not good. If other filesystems don't have these kind of limitation
> I'll be switching.

Having too few inodes has been a problem for me in the past. But that was
either a tiny partition for the portage tree, which has so many small
files. Or the partition where I back up my /var partition with
rdiff-backup, so it also has lots of files, and with each backup the same
amount gets added. If you intend to place unusually many files on a
partition, check how many inodes mkfs has created, and re-create the file
system using the -N option, giving it a somewhat larger number.

Wonko

On 12.12.2011 15:54, Alex Schuster wrote:
> Joseph writes:
>> That is scary. I just install new HD with 2TB capacity and ext4 that is
>> 2% full and:
>> $ find /home/joseph/ -xdev | wc -l
>> shows: 169977 that is 26% full.
> No, that is 26% of the number of total inodes _Daniel_ has on his small
> partition. Yours is bigger, so you have more inodes. My largest partition
> has 724G, and 46 million inodes. Use df -i to see how many you have.
Ah yes... My partition is only 10GB, and mkfs.ext4 sais in its man page
(at "-N" option) it uses a "calculation [..] based on the number of
blocks and the bytes-per-inode ratio". So a small partition will have
fewer inodes than a big partition.

mkfs.extX uses settings for the inode-block-ratio from /etc/mke2fs.conf.
The "-T" option configures which one to use, my partition falls into
category 512MB < "default" < 4TB, which makes it use inode_ratio=16384.
I think I should use the "news" type. It has inode_ratio=4096, which
should give me 4 times the inodes.... testing... yes: 2621440 inodes
instead of 655360.

Previously I used reiser3fs for this kind of filesystem usage, but it's
not faster than extX anymore. Moreover it's running on a SSD now, and
afaik reiser3fs doesn't support TRIM

OK - thank you all. It seems I'll have to reformat.

Bye,
Daniel

--
PGP key @ http://pgpkeys.pca.dfn.de/pks/lookup?search=0xBB9D4887&op=get
# gpg --recv-keys --keyserver hkp://subkeys.pgp.net 0xBB9D4887

On Mon, Dec 12, 2011 at 12:15:52PM +0100, Daniel Troeder wrote:

> I have an ext4-filesystem that contains /usr/src, the /usr/portage and
> /var/cache/edb. It previously also contained /var/db/pkg, but I had to
> move that some weeks ago, because the fs was "full". Now it's "full"
> again, though it has free blocks. But no inodes are left:
>
> $ fsck -vf /dev/sda5
> [..]
> 655360 inodes used (100.00%)
> [..]
>
> $ find /gentoo -xdev | wc -l
> 655338
>
> Is there any way to raise the number of inodes without using
> $ mkfs.ext4 -N BIGNUM

You can’t increase the number in an existing FS, you’ll have to reformat it.
If you have a backup, as you should anyway ;-) , you could boot a live system
and reformat.

On my system, /usr/portage currently contains 127000 files. But for reason of
increased performance I put it into a squashfs file. (There was a nice howto
on this ML some months ago). You could try that, which will free those inodes
up and ideally leaves you with one used inode for the squashfs image. Plus, if
you have enough RAM, you could put /var/tmp/portage into tmpfs. I have 3GB,
and this is fairly enough. For other hogs like firefox, LO and java, I use
binary packages though.
For comparision, I too have one (seldom two) kernel source trees and everything
else on / except /home. And while of the 17GB capacity barely 1GB is left free,
I still have 480k inodes free of the 1M in total. (I figured that I may have
more space for content if I reserved less for inodes).
--
Gruß | Greetings | Qapla'
I forbid any use of my email addresses with Facebook services.

A hammer is a wonderful tool,
but it is plain unsuitable for cleaning windows. (SelfHTML forum)

On 12/12/2011 06:15 AM, Daniel Troeder wrote:


$ fsck -vf /dev/sda5
[..]
655360 inodes used (100.00%)
[..]

$ find /gentoo -xdev | wc -l
655338

That's really disappointing. I was using reiser3fs and XFS before, and
they didn't have that kind of limitation... Uhm... not meant as a rant -
I like ext4 - that's why I'm moving (almost?) everything to it...

Is there any way to raise the number of inodes without using
$ mkfs.ext4 -N BIGNUM


640k should be enough for anybody.

On 13.12.2011 01:44, Frank Steinmetzger wrote:
> On my system, /usr/portage currently contains 127000 files. But for reason of
> increased performance I put it into a squashfs file. (There was a nice howto
> on this ML some months ago). You could try that, which will free those inodes
> up and ideally leaves you with one used inode for the squashfs image. Plus, if
> you have enough RAM, you could put /var/tmp/portage into tmpfs. I have 3GB,
> and this is fairly enough. For other hogs like firefox, LO and java, I use
> binary packages though.
> For comparision, I too have one (seldom two) kernel source trees and everything
> else on / except /home. And while of the 17GB capacity barely 1GB is left free,
> I still have 480k inodes free of the 1M in total. (I figured that I may have
> more space for content if I reserved less for inodes).
I had portage in a squashfs before too - that was nice
That's also the reason I had /usr/portage /var/cache/edb and /var/db/pkg
on one filesystem - all together in the squashfs
Because one day my / became full I moved /usr/src onto that partition
too (it was now on a reiser3fs). All fine, and other partitions less
fragmented... until I moved to ext4.

Now I have a SSD, and it's simpler than squashfs'ing and still fast.
distfiles is on a HDD (Thinkpad notebook with ultrabay - love it) and
compilation on tmpfs (8GB RAM, so no problems). I always have like 8
kernel trees lying around, so there are already like 400k files... For
various reasons I like to keep my stuff on separate partitions -> my
system is distributed over 6 partitions and my personal data over 3
partitions

That inode-trouble was actually quiet interesting

Daniel

--
PGP key @ http://pgpkeys.pca.dfn.de/pks/lookup?search=0xBB9D4887&op=get
# gpg --recv-keys --keyserver hkp://subkeys.pgp.net 0xBB9D4887