[Gluster-users] Quota problem : overquota late

[Thierry Lamoureux]

Hello,

I have a problem using GlusterFS Quota on a distributed volume.

Here is my conf :
- GlusterFS 3.5 (LATEST)
- 3 servers
- 1 brick of 10TB per server

Once my server have been added to the pool, I've performed the following commands :
- gluster volume create vol0 server1:/glusterfs/brick1/vol0 server2:/glusterfs/brick2/vol0 server3:/glusterfs/brick3/vol0
- gluster volume start vol0
- gluster volume set vol0 features.quota-timeout 5
- gluster volume quota vol0 enable
- gluster volume quota vol0 limit-usage /quota-test 50MB

First of all, the last command doesn't work if the "quota-test" folder isn't created, contrary to what it is mentioned in the GlusterFS documentation : "You can set the disk limit on the directory even if it is not created"

Here is the result of gluster volume info :

Volume Name: vol0
Type: Distribute
Volume ID: ee585e7a-1b89-40b5-bd37-1c977920e87f
Status: Started
Number of Bricks: 3
Transport-type: tcp
Bricks:
Brick1: server1:/glusterfs/projets-brick1/vol0
Brick2: server2:/glusterfs/projets-brick2/vol0
Brick3: server3:/glusterfs/projets-brick3/vol0
Options Reconfigured:
performance.cache-size: 256MB
features.quota-timeout: 5
features.quota: on

Then, I mount the volume on a client, go the the /mnt/quota-test folder and make a diskdump :
dd if=/dev/zero of=test bs=1M

After approx. 1min I've got the message :
dd: write of « test »: Quota exceeded
dd: closing output file « test »: Quota exceeded

Then, I lauch "gluster voluime quota vol0 list" :
                  Path                   Hard-limit Soft-limit   Used  Available
--------------------------------------------------------------------------------
/quota-test                               50.0MB       80%       1.3GB  0Bytes

All info are fine, but we are for 1GB overquota.

This behavior is the same with multiples copies : it take to long time to the gluster system to detect the overquota.

Have I miss something in the configuration steps?

Thank you for your help.

Kind regards,

Thierry !