[Gluster-devel] quota

Prashanth Pai ppai at redhat.com
Wed Oct 15 06:31:04 UTC 2014

I'm not sure if this is related but worth taking note of: write-behind when sees a shorter write, it chooses to ignore ENOSPC or EDQUOT that it received from the brick and will return a generic EIO sometimes.

 -Prashanth Pai

----- Original Message -----
From: "Raghavendra G" <raghavendra at gluster.com>
To: "Emmanuel Dreyfus" <manu at netbsd.org>
Cc: "Gluster Devel" <gluster-devel at gluster.org>
Sent: Wednesday, October 15, 2014 11:50:36 AM
Subject: Re: [Gluster-devel] quota

On Tue, Oct 14, 2014 at 9:18 AM, Emmanuel Dreyfus < manu at netbsd.org > wrote: 

Vijay Bellur < vbellur at redhat.com > wrote: 

> You would need to set features.soft-timeout and features.hard-timeout 
> values to 0 when testing with lower values of directory quota. 

It works more like expected this way, but there are still oddities: for 
instance once quota is reached, I can still append smal chunk to a file, 
and do it a lot of times. 

A few debug printf tell me this is because of write-behind: the writing 
process gets success, then glusterfs attemps to write - and fail. This 
means we silently discard data, which does not looks nice. Is it the 
expected behavior or is it a NetBSD bug? 

It happens on other environments too. Applications receive success for writes and writes wouldn't have happened on the brick. Write-behind propagates failures of cached-writes to application in the first file operation after write-failure to brick. In the worst case the application would see a non-zero return value for close of the fd. However not all applications check return value of close and hence write failures would go unnoticed by application. 

This behaviour is posix-conformant. From man 2 close, 

<man 2 close> 

Not checking the return value of close() is a common but nevertheless serious programming error. It is quite possible that errors on a 
previous write(2) operation are first reported at the final close(). Not checking the return value when closing the file may lead to 
silent loss of data. This can especially be observed with NFS and with disk quota. 

A successful close does not guarantee that the data has been successfully saved to disk, as the kernel defers writes. It is not common for 
a file system to flush the buffers when the stream is closed. If you need to be sure that the data is physically stored use fsync(2). (It 
will depend on the disk hardware at this point.) 

</man close> 

I assume it is expected but undesirable behavior: couldn't we check for 
quota space left, and avoid write behind if we aregoing to hit the 

Emmanuel Dreyfus 
manu at netbsd.org 
Gluster-devel mailing list 
Gluster-devel at gluster.org 

Raghavendra G 

Gluster-devel mailing list
Gluster-devel at gluster.org

More information about the Gluster-devel mailing list