OCS Providers File - High Numbers of Requests

[Aleix Pol]

On Thu, Sep 23, 2021 at 1:54 PM Aleix Pol <aleixpol at kde.org> wrote:
>
> On Thu, Sep 23, 2021 at 11:52 AM Ben Cooksley <bcooksley at kde.org> wrote:
> >
> > Hi all,
> >
> > It has recently come to our attention that the number of queries being handled for the endpoint https://autoconfig.kde.org/ocs/providers.xml on a day to day basis has gotten to the point where it is causing issues with server responsiveness to other traffic. This is perhaps best summarised by the following:
> >
> > root at nicoda /var/log/apache2 # ls -lah ...
> > -rw-r----- 1 root adm 458M Sep 23 06:25 autoconfig.kde.org.log.1
> > -rw-r----- 1 root adm 381M Sep 23 06:25 networkcheck.kde.org.log.1
> > -rw-r----- 1 root adm 143M Sep 23 06:25 www.kde.org.log.1
> >
> > root at nicoda /var/log/apache2 # cat autoconfig.kde.org.log.1 | wc -l
> > 4,222,343
> >
> > Based on those numbers we're looking at 48-49 requests per second (on average - peaks are much higher by many magnitudes), which seems extremely excessive given that this file is only supposed to be retrieved by KDE software when GHNS functionality is triggered. That is supported by the substantial size difference it has over networkcheck.kde.org - which is used by plasma-nm and NetworkManager (on Neon) to check for whether they have a working internet connection - which i'd expect to be the site receiving the most traffic.
> >
> > As such, I therefore suspect we have bug(s) in software that makes use of GHNS functionality.
> >
> > It would therefore be appreciated if we could please review the software in question to determine whether it is operating correctly. Given that it usually runs in the background on user systems, i'd especially appreciate it if a detailed review could be conducted on Discover and other software that conducts package management operations or assists in managing updates.
> >
> > Unfortunately all these applications submit a fairly useless user agent (Mozilla/5.0) so it is impossible for Sysadmin to ascertain any further information. If we could get information on the software that is originating the request added to the user agent to assist in investigating these issues in the future that would be extremely helpful.
> >
> > Thanks,
> > Ben
>
> That's correct. Discover fetches them at startup. It's necessary to be
> able to check if there are updates on KNS-provided resources.
>
> Incidentally,  I was looking into this yesterday incidentally. We
> could see if caching is broken somehow. A request will still be needed
> though to check if the cache is out of date.
>
> Or we can prefer the cached version of the file for a few days and
> assume they don't change that much.
>
> I've seen some of them point at
> "https://download.kde.org/ocs/providers.xml", I don't know if this is
> any better.
>
> Aleix

https://invent.kde.org/frameworks/attica/-/merge_requests/15/
https://invent.kde.org/frameworks/knewstuff/-/merge_requests/141
https://invent.kde.org/plasma/discover/-/merge_requests/165

This should lessen the problem from Discover's side. Still that one
providers.xml request at startup remains.

Aleix