[lustre-discuss] Interpretation of tot_granted, tot_pending, tot_dirty stats

Rasool Almasikoupaei rasool.almasikoupaei at uni-tuebingen.de
Tue Apr 8 03:12:56 PDT 2025


Thanks for the answer!

Best,
Rasool

On 07.04.25 20:07, Andreas Dilger wrote:
> On Apr 6, 2025, at 05:18, Rasool Almasikoupaei via lustre-discuss 
> <lustre-discuss at lists.lustre.org> wrote:
>>
>> Hi everyone,
>>
>>
>> Just wondering if anyone knows what these metric values actually mean 
>> and how I can make sense of them:
>>
>> obdfilter.testfs-OST0000.tot_granted
>> obdfilter.testfs-OST0000.tot_pending
>> obdfilter.testfs-OST0000.tot_dirty
>>
>> Appreciate any help!
>
> The "space grant" mechanism controls how much dirty data can be cached 
> on the clients (per OST) before the writes are throttled.  This avoids 
> clients dirtying GB (or TB these days) of data in RAM but the OST is 
> actually out of space and cannot write the data to persistent storage.
>
> tot_granted = total amount of space granted to all clients for this OST
> tot_pending = total amount of dirty data in process of being written 
> (i.e. in RPCs)
> tot_dirty = total amount of dirty data reported by clients
>
> The clients send these values to the server with every write RPC, so 
> the server knows the client totals even before the write RPCs are 
> sent.  This allows the server to pre-emptively send more write grants 
> to a client so that it doesn't need to throttle writes, if there is 
> still lots of free space on the server.
>
> Cheers, Andreas
>> Andreas Dilger
> Lustre Principal Architect
> Whamcloud/DDN
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20250408/07468174/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4858 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20250408/07468174/attachment.p7s>


More information about the lustre-discuss mailing list