OpenSimulator · EN

FSAssets: inodes, storage and performance

Monitor inodes, filesystem usage, small files, mountpoints and I/O risks.

Guide grounded in the OpenSimulator 0.9.3.1 / .NET 8 fork. Verify your version and back up before changing a live installation.

Why inodes are critical

On ext4, each file and directory consumes an inode; the total inode count is set when the filesystem is created. FSAssets creates many .gz files in nested directories. A volume can show 50% free disk space and 100% inode usage: new files fail despite apparent capacity.

Measure disk and inode usage

df -h reports byte capacity; df -i reports inode capacity. Run both on the actual directory, not only /. Resolve symlinks and verify the real mount.

Adapt this example: never run blindly in production.

readlink -f /srv/opensim/fsassets/data
findmnt -T /srv/opensim/fsassets/data
df -h /srv/opensim/fsassets/data
df -i /srv/opensim/fsassets/data
du -sh /srv/opensim/fsassets/data /srv/opensim/fsassets/tmp

Count files without guessing

A full tree walk over millions of files can be expensive and add metadata I/O load. Run off-peak and only on the correct tree. The number of .gz files is not identical to metadata rows because of content deduplication.

Adapt this example: never run blindly in production.

# Potentiellement très coûteux sur un grand volume :
find /srv/opensim/fsassets/data -type f -name '*.gz' -printf '.' | wc -c
# Échantillonner les sous-répertoires de premier niveau :
du -h --max-depth=1 /srv/opensim/fsassets/data | sort -h | tail

Ext4 and XFS differ

Ext4 normally establishes inode density during mkfs with options such as -i. XFS allocates inodes dynamically, subject to filesystem limits and available space. df -i remains useful on both. Never reformat an existing FSAssets disk to 'add inodes': this destroys data.

When inodes run out

Symptoms include create-file/directory failures, 'No space left on device', growing spool and new asset failures. Preserve logs, then inspect both df -h and df -i; similar messages can also mean full disk, quotas or read-only remounts.

Capacity alerts and forecasting

Monitor byte usage, inode usage, file count and spool size independently. An example local policy is early warning at 70%, action at 85% and urgent response at 95%, adjusted to your growth. These are operational examples, not OpenSimulator constants.

Adapt this example: never run blindly in production.

df -P /srv/opensim/fsassets/data
df -Pi /srv/opensim/fsassets/data
stat -f -c 'Type=%T Inodes libres=%f? (voir df -i)' /srv/opensim/fsassets/data
# Évolution : enregistrer ces métriques chaque jour via supervision

Spool size is a health indicator

FSAssets uses SpoolDirectory/spool and SpoolDirectory before writing the final store. If the writer stops processing due to permissions, disk, inode or I/O errors, spool may accumulate. Never clear it as a cache: it can contain assets not yet committed to BaseDirectory.

Adapt this example: never run blindly in production.

du -sh /srv/opensim/fsassets/tmp
find /srv/opensim/fsassets/tmp -maxdepth 1 -type f | wc -l
ls -ld /srv/opensim/fsassets/tmp /srv/opensim/fsassets/tmp/spool

HDD, SSD and NAS performance

Many small files stress metadata and IOPS; latency can matter more than sequential throughput. Measure iostat, iowait, latency and disk errors before migration. On NAS, also validate locking, rename semantics, durability and network outages; not every network filesystem is suitable for spool.

Adapt this example: never run blindly in production.

iostat -xz 1 5
vmstat 1 10
journalctl -k --since '2 hours ago' | grep -Ei 'I/O error|ext4|xfs|nvme|ata'

Other constraints beyond inodes

Check user quotas, filesystem reserves, permissions, snapshots, backup destination capacity and temporary file allocation. SQL metadata and log files may live on different disks: monitor each separately.

Further reading

Verifiable sources