Freeze break: cachefilesd

[Kevin Fenzi]

Greetings.

I'd like to take download05 out of dns and test out using cachefilesd
on it. We tried this a number of years ago with rhel6 and ran into
problems, but perhaps rhel7 might be more stable.

Basically this would be setting up a 500GB cache on the server and then
caching 500GB of stuff from nfs. In theory this would reduce load on
the backend NFS server. There is a pending netapp upgrade and storage
folks would need to throttle our access while the upgrade was
happening, so the more load we can take off the filer, the better it
will be when we have to do the upgrade.

I'd take download05 out of dns (so no one should hit it), setup
cachefilesd and test. If all goes well on testing we could re-add it
and see if handles normal load. If not or if it cannot handle regular
load we can just repave it and put it back the way it is now.

4 download machines should be fine for the normal load we have, and
this testing should long be done before the end of the week when we
might need to stage Alpha.

Thoughts? +1s?

kevin

[Stephen John Smoogen]

This sounds reasonable.. so +1

On 21 March 2016 at 16:08, Kevin Fenzi  wrote:
> Greetings.
>
> I'd like to take download05 out of dns and test out using cachefilesd
> on it. We tried this a number of years ago with rhel6 and ran into
> problems, but perhaps rhel7 might be more stable.
>
> Basically this would be setting up a 500GB cache on the server and then
> caching 500GB of stuff from nfs. In theory this would reduce load on
> the backend NFS server. There is a pending netapp upgrade and storage
> folks would need to throttle our access while the upgrade was
> happening, so the more load we can take off the filer, the better it
> will be when we have to do the upgrade.
>
> I'd take download05 out of dns (so no one should hit it), setup
> cachefilesd and test. If all goes well on testing we could re-add it
> and see if handles normal load. If not or if it cannot handle regular
> load we can just repave it and put it back the way it is now.
>
> 4 download machines should be fine for the normal load we have, and
> this testing should long be done before the end of the week when we
> might need to stage Alpha.
>
> Thoughts? +1s?
>
> kevin
>
> infrastructure mailing list
> [e-mail]
> http://lists.fedoraproject.org/admin/lists/[e-mail]...

[Tim Flink]

On Mon, 21 Mar 2016 16:08:41 -0600
Kevin Fenzi  wrote:

+1 sounds reasonable and something that will be very useful to have
sooner than later.

Tim
> Greetings.
>
> I'd like to take download05 out of dns and test out using cachefilesd
> on it. We tried this a number of years ago with rhel6 and ran into
> problems, but perhaps rhel7 might be more stable.
>
> Basically this would be setting up a 500GB cache on the server and
> then caching 500GB of stuff from nfs. In theory this would reduce
> load on the backend NFS server. There is a pending netapp upgrade and
> storage folks would need to throttle our access while the upgrade was
> happening, so the more load we can take off the filer, the better it
> will be when we have to do the upgrade.
>
> I'd take download05 out of dns (so no one should hit it), setup
> cachefilesd and test. If all goes well on testing we could re-add it
> and see if handles normal load. If not or if it cannot handle regular
> load we can just repave it and put it back the way it is now.
>
> 4 download machines should be fine for the normal load we have, and
> this testing should long be done before the end of the week when we
> might need to stage Alpha.
>
> Thoughts? +1s?
>
> kevin

[Matthew Galgoci]

> Date: Mon, 21 Mar 2016 16:08:41 -0600
> From: Kevin Fenzi 
> Reply-To: Fedora Infrastructure 
> To: [e-mail]
> Subject: Freeze break: cachefilesd
>
> Greetings.
>
> I'd like to take download05 out of dns and test out using cachefilesd
> on it. We tried this a number of years ago with rhel6 and ran into
> problems, but perhaps rhel7 might be more stable.
>
> Basically this would be setting up a 500GB cache on the server and then
> caching 500GB of stuff from nfs. In theory this would reduce load on
> the backend NFS server. There is a pending netapp upgrade and storage
> folks would need to throttle our access while the upgrade was
> happening, so the more load we can take off the filer, the better it
> will be when we have to do the upgrade.
>
> I'd take download05 out of dns (so no one should hit it), setup
> cachefilesd and test. If all goes well on testing we could re-add it
> and see if handles normal load. If not or if it cannot handle regular
> load we can just repave it and put it back the way it is now.
>
> 4 download machines should be fine for the normal load we have, and
> this testing should long be done before the end of the week when we
> might need to stage Alpha.
>
> Thoughts? +1s?

+1 IFF david howells is involved and available to help. I am sure he would be
curious at least to know you are trying it.