ACK: [m-f:linux-generic] & [jf:linux-gke]: [SRU][j/gke][f/gke][PATCH 0/1] Make new NFS cache behaviour opt-in
Dimitri John Ledkov
dimitri.ledkov at canonical.com
Thu Jun 1 19:24:13 UTC 2023
Hi,
On Thu, 1 Jun 2023 at 20:11, Khalid Elmously
<khalid.elmously at canonical.com> wrote:
>
> BugLink: https://bugs.launchpad.net/bugs/2022098
>
> The fix to LP #2003053 has caused massively increased NFS server access in some use-cases, which caused severe performance degradation to the point of being unusable.
>
> The solution to this issue, at least temporarily and at least for linux-gke, is to make the new behaviour optional using the "nfs_fasc=1" module parameter. Without this parameter specified, (or specified as =0) will keep the old behaviour.
>
I like this approach as it maintains status-quo for most people, yet
has a runtime option for those that are affected by 2003053.
We should mention this incoming change to users that are affected by 2003053.
Nothing in this issue is unique to linux-gke kernel - and likely
happens to be a canary as an early adopter. I believe we are / will
experience similar issue on other kernels. Separately GKE product can
be deployed with other kernels too.
My preference is to apply this patch series to: j/f linux-gke for a
quicker respin. But also apply it to all generic kernels, as a sauce
patch mantic..focal:linux-generic (and any other top levels that have
this code).
>
> Thanks Thadeu for the feedback and the original inspiration for this fix.
>
>
> Testing:
> - Used a slightly modified version of this patch (which simply logs whether the new or the old behaviour is taking effect) on an NFS client to confirm that 'modprobe nfs' and 'modprobe nfs nfs_fasc=0' maintain the old behaviour, and that 'modprobe nfs nfs_fasc=1' implements the new cache behaviour.
>
Ack.
>
> Regresion potential:
> - Regression potential is considered low considering the scope of the change and is limited to NFS only.
>
With prejudice
Acked-by: Dimitri John Ledkov <dimitri.ledkov at canonical.com>
for
* [jf:linux-gke] fast track
* [m-f:linux-generic] next cycle
Regards,
Dimitri.
More information about the kernel-team
mailing list