[LU-483] Lustre quota not usabe

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.0.0
• Type: Bug
• Priority: Minor
• Reporter: Patrick Valentin (Inactive)
• Assignee: Johann Lombardi (Inactive)
• Resolution: Fixed
• Severity: 3

Description

The CEA Bull customer is complaining about quota that are not working properly with Luster 2.0 Bull.
Until last week, quota were disabled because they were experiencing frequent system crashes on some IO servers.
We delivered an Efix containing the patch proposed in LU-369, and they re-enabled quota and made some tests.
There is no longer system crashes, but a user is able to write more than allowed by quotas.

Here is the description of there tests:
1. starting of the fs
2. quotaoff/quotacheck/quotaon/quotacheck
3. set the quota to a user who has already exceeded it's quota
4. this user is able to continue to write
5. set the user quota to a value higher than the actual used capacity
6. this user is able to write and to exceed is quota

I attached the syslog of the IO servers they provided (MDS:curie113, OSS:curie200-207).

Comments

[Johann Lombardi (Inactive)]

> Here is the description of there tests:
> 1. starting of the fs
> 2. quotaoff/quotacheck/quotaon/quotacheck
> 3. set the quota to a user who has already exceeded it's quota
> 4. this user is able to continue to write

Please note that there is no integration between the data writeback cache and quota. This means that a user can overrun his quota limit, up to #clients x #OSC x max_dirty_mb in the worst case scenario.
Could you please tell us how much data you tried to write? Do you eventually get"quota exceeded" if you write more than 32MB from one client? Could you please give us the output of lfs quota -u $username?

[Patrick Valentin (Inactive)]

Hi Johan,

I got in touch with our on-site support and it seems it was a copy of about 1 Tb.
They will transmit your questions to the customer.

Patrick

[Sebastien Buisson (Inactive)]

No news from the customer since last Friday.

[Sebastien Buisson (Inactive)]

Hi,

Is there any way to check per-OST quota accounting? An internal file stored in the MDT or in each OST for instance?
That would help this problem's diagnostic.

TIA,
Sebastien.

[Johann Lombardi (Inactive)]

Sure, lfs quota -uv $username /path should return those info.

[Johann Lombardi (Inactive)]

Any news or shall we just close this bug? Please advise.

[Patrick Valentin (Inactive)]

After one of the latest CEA EFIX installation, in the course of september, they made some quota tests and it seems to work correctly.
This ticket can be closed as it was not reproduced (probably a system which was not at the latest EFIX level).
Cheers

[Peter Jones]

Good news! Thanks Patrick!