Re: Red alert but apparently harmless setscene errors

Jose Quaresma

Richard Purdie <richard.purdie@...> escreveu no dia quinta, 13/01/2022 à(s) 21:53:
On Thu, 2022-01-13 at 19:37 +0100, Michael Opdenacker wrote:
> Hi Jose,
> Thanks for your advice!
> On 1/13/22 6:37 PM, Jose Quaresma wrote:
> > Hi,
> >
> > I've seen the same for some time now when building core-image-minimal
> > with yocto state mirror and hashequiv servers.
> >
> > After some investigation I think the root cause is some
> > bandwidth/connection limitation on the http yocto mirror.
> >
> > Reducing the BB_NUMBER_THREADS="1" to a single core solves many of the
> > fetch issues for me,
> > my config is the default one with:
> >
> >   SSTATE_MIRRORS = "file://.*
> >;downloadfilename=PATH
> > <;downloadfilename=PATH>"
> > <>"
> Oops, it seems you're right. I have the same SSTATE_MIRRORS and
> BB_HASHSERVE_UPSTREAM settings as you do, and adding
> BB_NUMBER_THREADS="1" makes the errors disappear.
> It seems I should file a bug about that. The error messages just mention
> local paths, not remote ones, so it sounds like a local issue while it's
> apparently not.

That is useful to debug it but suggests there is an issue with concurrent
connections to the YP server.

I think we should have an open bug and get Michael Halstead's input on this.



During the last yocto project summit 2021.11 I have submitted a patch
to reduce the time needed for handling the sstate mirrors on startup,
that without my realizing it actually launches a DoS attack to the mirror.
This patch was rejected as expected.

After that patch, maybe by coincidence I got the idea that some measures
had been taken to mitigate this problem but if it wasn't the case then there is
a bug in concurrent connections to the YP server.

When the patch was sent there was no problem with the concurrent connections to the YP server.


Join to automatically receive all group messages.