Elliswilson: Created page with "== Overview == Lustre quotas restrict how much disk space (blocks) and how many files (inodes) a user, group, or project can consume. The quota subsystem consists of: * '''QMT''' (Quota Master Target) — runs on the MDT, holds the global quota limits. * '''QSD''' (Quota Slave Device) — runs on each OST and MDT, enforces limits locally. * '''qunit''' — the allocation granularity the QMT grants to each QSD. The QMT dynamically adjusts qunit size based on remaining q..."

2026-08-31T15:19:21Z

Created page with "== Overview == Lustre quotas restrict how much disk space (blocks) and how many files (inodes) a user, group, or project can consume. The quota subsystem consists of: * '''QMT''' (Quota Master Target) — runs on the MDT, holds the global quota limits. * '''QSD''' (Quota Slave Device) — runs on each OST and MDT, enforces limits locally. * '''qunit''' — the allocation granularity the QMT grants to each QSD. The QMT dynamically adjusts qunit size based on remaining q..."

New page

== Overview ==

Lustre quotas restrict how much disk space (blocks) and how many files (inodes) a user, group, or project can consume. The quota subsystem consists of:

* '''QMT''' (Quota Master Target) — runs on the MDT, holds the global quota limits.
* '''QSD''' (Quota Slave Device) — runs on each OST and MDT, enforces limits locally.
* '''qunit''' — the allocation granularity the QMT grants to each QSD. The QMT dynamically adjusts qunit size based on remaining quota and the number of slaves.

Space accounting is enabled by default on ldiskfs. Quota ''enforcement'' must be explicitly enabled.

== Enabling Quota Enforcement ==

Enable enforcement for user, group, and project quotas on all targets:

<pre>
# Enable user, group, and project quota enforcement
lctl set_param -P osd-*.fsname*.quota_slave_dt.enabled=ugp
lctl set_param -P osd-*.fsname*.quota_slave_md.enabled=ugp
</pre>

To verify:

<pre>
lctl get_param osd-*.*.quota_slave_dt.enabled
lctl get_param osd-*.*.quota_slave_md.enabled
</pre>

Valid values: <code>u</code> (user), <code>g</code> (group), <code>p</code> (project), or any combination (e.g. <code>ugp</code>). Use <code>none</code> to disable.

== Setting User and Group Quotas ==

Set block and inode limits for a user:

<pre>
# -b soft block limit, -B hard block limit (in KiB by default)
# -i soft inode limit, -I hard inode limit
lfs setquota -u bob -b 307200 -B 309200 -i 10000 -I 11000 /mnt/testfs
</pre>

Set limits for a group:

<pre>
lfs setquota -g engineers -b 10G -B 12G -i 500000 -I 600000 /mnt/testfs
</pre>

Suffixes <code>K</code>, <code>M</code>, <code>G</code>, <code>T</code>, <code>P</code> are accepted for block limits.

== Viewing Quotas ==

<pre>
# Show quota for a specific user
lfs quota -u bob /mnt/testfs

# Show quota for a specific group
lfs quota -g engineers /mnt/testfs

# Show quota usage per OST for a user
lfs quota -u bob -v /mnt/testfs

# Show quotas for all users
lfs quota -u -a /mnt/testfs
</pre>

== Default Quotas ==

Default quotas apply to all users (or groups) that do not have explicit limits set:

<pre>
# Set default user quota
lfs setquota -U -b 1G -B 2G -i 100000 -I 200000 /mnt/testfs

# Set default group quota
lfs setquota -G -b 5G -B 10G -i 500000 -I 1000000 /mnt/testfs

# View default quotas
lfs quota -U /mnt/testfs
lfs quota -G /mnt/testfs
</pre>

Default quotas use <code>-U</code>, <code>-G</code>, or <code>-P</code> (uppercase) instead of <code>-u</code>, <code>-g</code>, <code>-p</code>.

== Grace Periods ==

When a user exceeds the soft limit but remains below the hard limit, a grace period timer starts. If usage is not reduced below the soft limit before the grace period expires, the soft limit is enforced as a hard limit.

The default grace period is '''1 week'''.

Set grace periods:

<pre>
# Set block and inode grace period for users (format: XXwXXdXXhXXmXXs)
lfs setquota -u -t -b 86400 -i 604800 /mnt/testfs

# Set group grace period to 3 days for blocks, 1 week for inodes
lfs setquota -g -t -b 3d -i 1w /mnt/testfs
</pre>

== Clearing Quotas ==

To remove quota limits for a user, set all limits to zero:

<pre>
lfs setquota -u bob -b 0 -B 0 -i 0 -I 0 /mnt/testfs
</pre>

== Architecture Notes ==

The QMT on the MDT is the single authority for quota limits. When a QSD on an OST needs to allocate space for a user, it requests a ''qunit'' allocation from the QMT. The qunit size is dynamic — it shrinks as quota usage approaches the limit (to reduce over-allocation) and grows when usage is low. This means:

* Quota enforcement is '''approximate''' until the user is near the limit. A user may slightly exceed a limit before enforcement kicks in.
* After changing a quota limit, it may take a short time for all QSDs to receive the updated allocation.

== See Also ==

* [[Lustre Quota Troubleshooting]]
* [[Lustre Project Quotas]]
* [[OST Pool Quotas]]

[[Category:Lustre Systems Administration]]
[[Category:Lustre Quotas]]

Lustre Quota Administration - Revision history