Re: [PATCH 0/1] Enable inter server to server copies on a export

[Steve Dickson <steved@redhat.com>]

On 11/1/21 15:25, J. Bruce Fields wrote:
> On Mon, Nov 01, 2021 at 12:22:15PM -0700, dai.ngo@oracle.com wrote:
>>
>> On 11/1/21 12:02 PM, Steve Dickson wrote:
>>>
>>>
>>> On 11/1/21 11:40, J. Bruce Fields wrote:
>>>> On Mon, Nov 01, 2021 at 11:30:48AM -0400, Steve Dickson wrote:
>>>>> Now that cp will use copy_file_range() when available,
>>>>> what are the steps needed to enable these fast copies?
>>>>
>>>> 1) Make sure client and both servers support NFSv4.2 and
>>>> server-to-server copy.
>>> Something is already figuring this out... The only time
>>> the client sends a COPY_NOTIFY and COPY is when both
>>> mounts are 4.2. I have not looked into but that is what
>>> the network traces are showing.
> 
> Right.  I was thinking what I'd tell an admin who wanted to set up
> server-to-server copy.  The first thing they'd need to do was check that
> their clients and servers were new enough.
Well the code went in over two years ago
ce0887ac96d35 (Olga Kornievskaia 2019-10-09 11:50:48 -0400 1663)
so most modern kernel will have the feature.

The real question is does the cp command have the
copy_file_range() support which didn't go in until
Mar of this year
* Wed Mar 24 2021 Kamil Dudka <kdudka@redhat.com> - 8.32-20
- cp: use copy_file_range if available

> 
>>>> 2) Make sure destination server can access (at least for read) any
>>>> exports on the source that you want to be able to copy from.
>>> How can one server know what the other server has exported
>>> or access to??
> 
> And the second is to make sure that the destination server is able to
> read from the source.
> 
>>>> 3) echo 1 >/sys/module/nfsd/parameters/inter_copy_offload_enable on the
>>>> destination server.
>>> Who would be doing this? Plus this would not survive over a reboot.
>>> An export would as well a /etc/modprobe.d/ file.
>>
>> You can add a line in /etc/modprobe.d/nfsd.conf:
>>
>> options nfsd inter_copy_offload_enable=Y
>>
>> to enable the option.
Yes... This is what I meant by "a /etc/modprobe.d/ file"

> 
> Yep, it would be better to document it that way, thanks.
But the question is who would create this file?
nfs-utils or the admin?

If it is nfs-utils, it is a global switch unbeknown by
the admin (but documented). If by the admin, the person
would know about but it still will not documented.

With a export option, it would be per export switch
and documented. Just saying... :-)

steved.