maxpct option for small xfs filesystems

DISCUSSION Alexander Tsvetkov 3 participants 8 messages Jan 26, 2015 – Jan 28, 2015

Share & tools

Alexander Tsvetkov

author

2015-01-26 16:29 UTC #0 lore raw

Hello,

I'm trying to understand the expected behaviour of "maxpct" option in case of small xfs filesystem comparing the maximum percentage defined for this option with the percentage of actually allocated inodes in filesystem, but the result of prepared test case doesn't correspond to the expectations:

[root@fedora ~]#mkfs.xfs -f -d size=16m -i maxpct=1 /dev/sdb2 [root@fedora ~]# mount /dev/sdb2 /mnt/scratch/
[root@fedora ~]# mount | grep scratch
/dev/sdb2 on /mnt/scratch type xfs (rw,relatime,seclabel,attr2,inode64,noquota) [root@fedora ~]# xfs_info /mnt/scratch/
meta-data=/dev/sdb2 isize=256 agcount=1, agsize=4096 blks

                 =                       sectsz=512   attr=2, projid32bit=1
                 =                       crc=0        finobt=0

data = bsize=4096 blocks=4096, imaxpct=1
= sunit=0 swidth=0 blks
naming =version 2 bsize=4096 ascii-ci=0 ftype=0
log =internal bsize=4096 blocks=853, version=2
= sectsz=512 sunit=0 blks,
lazy-count=1
realtime =none extsz=4096 blocks=0, rtextents=0

then trying to fill filesystem with lot of empty files until enospc obtained and break the loop:

[root@fedora ~]# for i in {0..100000}; do str=$(mktemp --tmpdir=/mnt/scratch tmp.XXXXXXXXXX); echo $str; done

filesystem is full with created files:

[root@fedora ~]# df -Th | grep scratch
/dev/sdb2 xfs 13M 13M 148K 99% /mnt/scratch

and from the number of actually created inodes:

[root@fedora ~]# xfs_db -c "blockget -n" -c "ncheck" /dev/sdb2 | wc -l 40512

it's calculated the space allocated for file inodes 9.9Mb (40512*256b inode size) that is about 77% of filesystem space instead of defined maximum with maxpct=1% or even default 25% value.

Thanks,
Alexander Tsvetkov

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Dave Chinner

2015-01-26 22:37 UTC #1 lore raw

Quoted 10 lines · Alexander Tsvetkov

Hello,
I'm trying to understand the expected behaviour of "maxpct" option in case of small xfs filesystem
comparing the maximum percentage defined for this option with the percentage of actually allocated
inodes in filesystem, but the result of prepared test case doesn't correspond to the expectations:
[root@fedora ~]#mkfs.xfs -f -d size=16m -i maxpct=1 /dev/sdb2

On 3.19-rc5, immediately after mount:

# df -i /mnt/scratch
Filesystem     Inodes IUsed IFree IUse% Mounted on
/dev/ram1         640     3   637    1% /mnt/scratch

Which indicates that imaxpct=1 is being calculated correctly, before we even look at whether it is being enforced correctly or not.

So, what kernel version?

http://xfs.org/index.php/XFS_FAQ#Q:_What_information_should_I_include_when_reporting_a_problem.3F

Quoted 2 lines

[root@fedora ~]# for i in {0..100000}; do str=$(mktemp --tmpdir=/mnt/scratch tmp.XXXXXXXXXX); echo $str; done

Which is a complex (and very slow!) way of doing:

# for i in {0..100000}; do echo > /mnt/scratch/$i ; done 2> /dev/null

Quoted 4 lines

filesystem is full with created files:
[root@fedora ~]# df -Th | grep scratch
/dev/sdb2 xfs 13M 13M 148K 99% /mnt/scratch

# df -Th /mnt/scratch
Filesystem     Type  Size  Used Avail Use% Mounted on
/dev/ram1      xfs    13M  1.1M   12M   9% /mnt/scratch
# df -i /mnt/scratch
Filesystem     Inodes IUsed IFree IUse% Mounted on
/dev/ram1         640   640     0  100% /mnt/scratch

Quoted 4 lines

and from the number of actually created inodes:
[root@fedora ~]# xfs_db -c "blockget -n" -c "ncheck" /dev/sdb2 | wc -l 40512

That's a directory structure entry count, equivalent to 'find /mnt/scratch | wc -l', not an allocated inode count which is what 'df -i' reports. Even so, on 3.19-rc5:

# xfs_db -c "blockget -n" -c "ncheck" /dev/ram1 | wc -l
637

which matches what 'df -i' tells us about allocated inodes and hence imaxpct is working as expected.

Cheers,

Dave.

Signature · 7 lines

Dave Chinner
[email protected]

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Alexander Tsvetkov

author

2015-01-27 16:22 UTC #2 lore raw

Quoted 17 lines · Dave Chinner

Quoted 10 lines · Alexander Tsvetkov
Hello,
I'm trying to understand the expected behaviour of "maxpct" option in case of small xfs filesystem
comparing the maximum percentage defined for this option with the percentage of actually allocated
inodes in filesystem, but the result of prepared test case doesn't correspond to the expectations:
[root@fedora ~]#mkfs.xfs -f -d size=16m -i maxpct=1 /dev/sdb2
On 3.19-rc5, immediately after mount:
# df -i /mnt/scratch
Filesystem     Inodes IUsed IFree IUse% Mounted on
/dev/ram1         640     3   637    1% /mnt/scratch

Quoted 6 lines

Which indicates that imaxpct=1 is being calculated correctly, before we even look at whether it is being enforced correctly or not.
So, what kernel version?
http://xfs.org/index.php/XFS_FAQ#Q:_What_information_should_I_include_when_reporting_a_problem.3F

I use Fedora20 on vbox virtual machine with latest kernel version available from fedora repos: 3.17.8-200.fc20.x86_64 and xfsprogs-3.2.1-1.fc20.x86_64.

/dev/sdb test storage is of VDI format, fixed size:
[root@fedora ~]# fdisk -l

Disk /dev/sda: 10.3 GiB, 11005845504 bytes, 21495792 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes I/O size (minimum/optimal): 512 bytes / 512 bytes Disk label type: dos
Disk identifier: 0x000011de

Device    Boot     Start       End  Blocks  Id System
/dev/sda1 *         2048   1026047  512000  83 Linux
/dev/sda2        1026048  20469759 9721856  83 Linux
/dev/sda3       20469760  21493759  512000  82 Linux swap / Solaris

Disk /dev/sdb: 8 GiB, 8589934592 bytes, 16777216 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes I/O size (minimum/optimal): 512 bytes / 512 bytes Disklabel type: dos
Disk identifier: 0x00006ee7

Device    Boot     Start       End  Blocks  Id System
/dev/sdb1           2048   8390655 4194304  83 Linux
/dev/sdb2        8390656  16777215 4193280  83 Linux

Quoted 17 lines

Quoted 2 lines
[root@fedora ~]# for i in {0..100000}; do str=$(mktemp --tmpdir=/mnt/scratch tmp.XXXXXXXXXX); echo $str; done
Which is a complex (and very slow!) way of doing:
# for i in {0..100000}; do echo > /mnt/scratch/$i ; done 2> /dev/null
Quoted 4 lines
filesystem is full with created files:
[root@fedora ~]# df -Th | grep scratch
/dev/sdb2 xfs 13M 13M 148K 99% /mnt/scratch
# df -Th /mnt/scratch
Filesystem     Type  Size  Used Avail Use% Mounted on
/dev/ram1      xfs    13M  1.1M   12M   9% /mnt/scratch
# df -i /mnt/scratch
Filesystem     Inodes IUsed IFree IUse% Mounted on
/dev/ram1         640   640     0  100% /mnt/scratch

Quoted 7 lines

Quoted 4 lines
and from the number of actually created inodes:
[root@fedora ~]# xfs_db -c "blockget -n" -c "ncheck" /dev/sdb2 | wc -l 40512
That's a directory structure entry count, equivalent to 'find /mnt/scratch | wc -l', not an allocated inode count which is what 'df -i' reports.

manual page for xfs_db ncheck says about inode numbers not a directory entry numbers:

"ncheck [-s] [-i ino] Print name-inode pairs"

Quoted 7 lines

Even so, on 3.19-rc5:
# xfs_db -c "blockget -n" -c "ncheck" /dev/ram1 | wc -l
637
which matches what 'df -i' tells us about allocated inodes and hence imaxpct is working as expected.

I have not the same results, just installed 3.19-rc6 and repeated the test, df -i reports 640 inodes for filesystem, but actually created 40512 files:

[root@fedora ~]# mkfs.xfs -f -d size=16m -i maxpct=1 /dev/sdb2

meta-data=/dev/sdb2              isize=256    agcount=1, agsize=4096 blks
         =                       sectsz=512   attr=2, projid32bit=1
         =                       crc=0        finobt=0
data     =                       bsize=4096   blocks=4096, imaxpct=1
         =                       sunit=0      swidth=0 blks
naming   =version 2              bsize=4096   ascii-ci=0 ftype=0
log      =internal log           bsize=4096   blocks=853, version=2
         =                       sectsz=512   sunit=0 blks, lazy-count=1
realtime =none                   extsz=4096   blocks=0, rtextents=0

[root@fedora ~]# mount /dev/sdb2 /mnt/scratch/
fill with files until enospc...
[root@fedora ~]# df -i /mnt/scratch/
Filesystem Inodes IUsed IFree IUse% Mounted on
/dev/sdb2 640 640 0 100% /mnt/scratch
[root@fedora ~]# df -Th /mnt/scratch/
Filesystem Type Size Used Avail Use% Mounted on
/dev/sdb2 xfs 13M 13M 156K 99% /mnt/scratch
[root@fedora ~]# umount /mnt/scratch
[root@fedora ~]# xfs_db -c "blockget -n" -c "ncheck" /dev/sdb2 | wc -l 40512

Looking into ncheck output there are 40512 pairs reported in the output each with own unique inode number. ncheck doesn't report inodes count by definition, but what does these 40512 reported inode numbers mean if only actually 640 inodes were allocated? From another hand each new file should have associated meta-data in the corresponding allocated inode structure, so for 40512 newly created files I expect the same count of allocated inodes, is it correct?

Quoted 3 lines

Cheers,
Dave.

Thanks,
Alexander Tsvetkov

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Eric Sandeen

2015-01-27 16:31 UTC #3 lore raw

On 1/27/15 10:23 AM, Alexander Tsvetkov wrote:

...

Quoted 13 lines

I have not the same results, just installed 3.19-rc6 and repeated the test., df -i reports 640 inodes for filesystem, but actually created 40512 files:

[root@fedora ~]# mkfs.xfs -f -d size=16m -i maxpct=1 /dev/sdb2

meta-data=/dev/sdb2              isize=256    agcount=1, agsize=4096 blks
         =                       sectsz=512   attr=2, projid32bit=1
         =                       crc=0        finobt=0
data     =                       bsize=4096   blocks=4096, imaxpct=1
         =                       sunit=0      swidth=0 blks
naming   =version 2              bsize=4096   ascii-ci=0 ftype=0
log      =internal log           bsize=4096   blocks=853, version=2
         =                       sectsz=512   sunit=0 blks, lazy-count=1
realtime =none                   extsz=4096   blocks=0, rtextents=0

Quoted 11 lines

[root@fedora ~]# mount /dev/sdb2 /mnt/scratch/
fill with files until enospc...
[root@fedora ~]# df -i /mnt/scratch/
Filesystem Inodes IUsed IFree IUse% Mounted on
/dev/sdb2 640 640 0 100% /mnt/scratch
[root@fedora ~]# df -Th /mnt/scratch/
Filesystem Type Size Used Avail Use% Mounted on
/dev/sdb2 xfs 13M 13M 156K 99% /mnt/scratch
[root@fedora ~]# umount /mnt/scratch
[root@fedora ~]# xfs_db -c "blockget -n" -c "ncheck" /dev/sdb2 | wc -l 40512

and what does df -i say after remount?

This is actually a problem with the lazy superblock counters I've run into before, but haven't yet fixed. This kind of workload is such that it never trips the runtime rebalancing.

Quoted 5 lines

Looking into ncheck output there are 40512 pairs reported in the output each with own unique inode number. ncheck doesn't report inodes count by definition, but what does these 40512 reported inode numbers mean if only actually 640 inodes were allocated? From another hand each new file should have associated meta-data in the corresponding allocated inode structure, so for 40512 newly created files I expect the same count of allocated inodes, is it correct?

Recheck df -i after remount, I think you will see many more than 640.

-Eric

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Eric Sandeen

2015-01-27 19:15 UTC #4 lore raw

Quoted 3 lines · Eric Sandeen

This is actually a problem with the lazy superblock counters I've run into before, but haven't yet fixed. This kind of workload is such that it never trips the runtime rebalancing.

working on a fix now ;)

-Eric

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Alexander Tsvetkov

author

2015-01-28 10:41 UTC #5 lore raw

In reply to Eric Sandeen

Quoted 28 lines · Eric Sandeen

On 1/27/15 10:23 AM, Alexander Tsvetkov wrote:
...
Quoted 11 lines
I have not the same results, just installed 3.19-rc6 and repeated the test., df -i reports 640 inodes for filesystem, but actually created 40512 files:
[root@fedora ~]# mkfs.xfs -f -d size=16m -i maxpct=1 /dev/sdb2
meta-data=/dev/sdb2 isize=256 agcount=1, agsize=4096 blks
          =                       sectsz=512   attr=2, projid32bit=1
          =                       crc=0        finobt=0
data     =                       bsize=4096   blocks=4096, imaxpct=1
          =                       sunit=0      swidth=0 blks
naming   =version 2              bsize=4096   ascii-ci=0 ftype=0
log      =internal log           bsize=4096   blocks=853, version=2
Quoted 13 lines
= sectsz=512 sunit=0 blks, lazy-count=1
realtime =none extsz=4096 blocks=0, rtextents=0
[root@fedora ~]# mount /dev/sdb2 /mnt/scratch/
fill with files until enospc...
[root@fedora ~]# df -i /mnt/scratch/
Filesystem Inodes IUsed IFree IUse% Mounted on
/dev/sdb2 640 640 0 100% /mnt/scratch
[root@fedora ~]# df -Th /mnt/scratch/
Filesystem Type Size Used Avail Use% Mounted on
/dev/sdb2 xfs 13M 13M 156K 99% /mnt/scratch
[root@fedora ~]# umount /mnt/scratch
[root@fedora ~]# xfs_db -c "blockget -n" -c "ncheck" /dev/sdb2 | wc -l 40512

Quoted 1 line

and what does df -i say after remount?

Nothing changed after remount:

[root@fedora ~]# xfs_db -c "blockget -n" -c "ncheck" /dev/sdb2 | wc -l 40512
[root@fedora ~]# mount /dev/sdb2 /mnt/scratch/
[root@fedora ~]# xfs_info /mnt/scratch/
meta-data=/dev/sdb2 isize=256 agcount=1, agsize=4096 blks
= sectsz=512 attr=2, projid32bit=1
= crc=0 finobt=0
data = bsize=4096 blocks=4096, imaxpct=1
= sunit=0 swidth=0 blks
naming =version 2 bsize=4096 ascii-ci=0 ftype=0
log =internal bsize=4096 blocks=853, version=2
= sectsz=512 sunit=0 blks, lazy-count=1
realtime =none extsz=4096 blocks=0, rtextents=0
[root@fedora ~]# df -i /mnt/scratch/
Filesystem Inodes IUsed IFree IUse% Mounted on
/dev/sdb2 640 640 0 100% /mnt/scratch
[root@fedora ~]# umount /mnt/scratch
[root@fedora ~]# xfs_db -c "blockget -n" -c "ncheck" /dev/sdb2 | wc -l 40512
[root@fedora ~]# mount /dev/sdb2 /mnt/scratch/
[root@fedora ~]# df -i /mnt/scratch/
Filesystem Inodes IUsed IFree IUse% Mounted on
/dev/sdb2 640 640 0 100% /mnt/scratch
[root@fedora ~]# mount -o remount /dev/sdb2 /mnt/scratch/ [root@fedora ~]# df -i /mnt/scratch/
Filesystem Inodes IUsed IFree IUse% Mounted on
/dev/sdb2 640 640 0 100% /mnt/scratch
[root@fedora ~]# umount /mnt/scratch
[root@fedora ~]# xfs_db -c "blockget -n" -c "ncheck" /dev/sdb2 | wc -l 40512

Quoted 11 lines

This is actually a problem with the lazy superblock counters I've run into before, but haven't yet fixed. This kind of workload is such that it never trips the runtime rebalancing.
Quoted 5 lines
Looking into ncheck output there are 40512 pairs reported in the output each with own unique inode number. ncheck doesn't report inodes count by definition, but what does these 40512 reported inode numbers mean if only actually 640 inodes were allocated? From another hand each new file should have associated meta-data in the corresponding allocated inode structure, so for 40512 newly created files I expect the same count of allocated inodes, is it correct?
Recheck df -i after remount, I think you will see many more than 640.

Do you mean that ncheck reports right number of allocated inodes instead of df -i? If counters are incorrect then the maxpct limit is not working as well, because there are much more of filesystem space allocated for inodes then defined limit 1%.

What is expected behaviour in this case when it's required for fs to allocate new inodes starting to exceed defined maxpct percentage? Which error is expected to be returned to user, enospc or probably just some warning?

Quoted 2 lines

-Eric

Thanks,
Alexander Tsvetkov

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Eric Sandeen

2015-01-28 15:44 UTC #6 lore raw

On 1/28/15 4:41 AM, Alexander Tsvetkov wrote:

...

Quoted 2 lines

Quoted 1 line
and what does df -i say after remount?
Nothing changed after remount:

Sorry, I misremembered the problem. :(

...

Quoted 10 lines

Quoted 10 lines
This is actually a problem with the lazy superblock counters I've run into before, but haven't yet fixed. This kind of workload is such that it never trips the runtime rebalancing.
Quoted 5 lines
Looking into ncheck output there are 40512 pairs reported in the output each with own unique inode number. ncheck doesn't report inodes count by definition, but what does these 40512 reported inode numbers mean if only actually 640 inodes were allocated? From another hand each new file should have associated meta-data in the corresponding allocated inode structure, so for 40512 newly created files I expect the same count of allocated inodes, is it correct?

Quoted 7 lines

Quoted 1 line
Recheck df -i after remount, I think you will see many more than 640.
Do you mean that ncheck reports right number of allocated inodes instead of df -i? If counters are incorrect then the maxpct limit is not working as well, because there are much more of filesystem space allocated for inodes then defined limit 1%.
What is expected behaviour in this case when it's required for fs to allocate new inodes starting to exceed defined maxpct percentage? Which error is expected to be returned to user, enospc or probably just some warning?

There are a couple of things going on here.

XFS superblock counters are kept in per-cpu variables for scalability; at certain times these per-cpu counts are coalesced into a global counter based on various thresholds. The thing about the inode counter is that it generally counts up, and it's the high thresholds that we care about (in order to determine if we've hit maxpct), but it's written as one of the low-threshold counters (as with free inodes / free space).

So in your test, you're rapidly running up the counter on a single cpu, and it's not getting coalesced into the main counter, which is where the imaxpct test happens. So that's why you're able to exceed imaxpct.

The df -i (statfs) output is wrong because of a bug in xfs_fs_statfs. For a filesystem like XFS which dynamically allocates and deallocates inodes, it's a little tricky to answer the question "how many inodes are available?" It's limited by free space as well as by imaxpct. There's code in there which does this:

        if (mp->m_maxicount)
                statp->f_files = min_t(typeof(statp->f_files),
                                        statp->f_files,
                                        mp->m_maxicount);

so it will never report more than mp->m_maxicount, which is essentially the imaxpct limit, even if we've actually allocated more than that.

A simple fix for this is to recognize that imaxpct is something of a soft limit (it's not critical if we overrun by a few hundred or even thousand inodes), and if the superblock counter of allocated inodes (sbp->sb_icount) exceeds mp->m_maxicount due to such an overrun, we should report that (true) value instead.

I'm working on a couple of patches which should fix both issues.

-Eric

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Eric Sandeen

2015-01-28 18:05 UTC #7 lore raw

In reply to Alexander Tsvetkov

Quoted 2 lines · Alexander Tsvetkov

What is expected behaviour in this case when it's required for fs to allocate new inodes starting to exceed defined maxpct percentage? Which error is expected to be returned to user, enospc or probably just some warning?

Sorry, I didn't answer this. ENOSPC is expected.

There's not a lot of value in enforcing this strictly to the last inode, because a % is fairly coarse anyway, but we should not blow right past it as we do today.

-Eric

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs