Created attachment 3047 [details] nightly-world log file The master build last night failed in nightly-world and nightly-world-lsb at the PublishingArtifacts stage as it refuses to copy over existing artefacts. It's possible this is because a previous build was aborted? (the Fedora22 machine hung, causing the worker to be lost and the nightly build to fail). For nightly-world the failure was: cp: cannot create regular file ‘/srv/www/vhosts/autobuilder.yoctoproject.org/pub/nightly/20160320-2/deb/x86/lib32-gawk-locale-ca_4.1.3-r0_i386.deb’: File exists For nightly-world-lsb there are many more of a similar pattern. Logs are attached.
Created attachment 3048 [details] nightly-world-lsb log file
Halstead, This looks like something with the NFS server. We've seen this before, any thoughts?
I do not see anything problematic in the logs on the NFS server. A cursory search makes me think this is a race condition between 2 cp processes. I suspect it is something like: cp #1 checks if a file exists: doesn't exist cp #2 checks if a file exists: doesn't exist cp #1 creates the new file cp #2 tries to create the file but it already exists, it prints this error Is there a case where multiple copy processes would be launched copying into 20160320-2/deb/x86/ ?
(In reply to comment #3) > I do not see anything problematic in the logs on the NFS server. A cursory > search makes me think this is a race condition between 2 cp processes. > > I suspect it is something like: > > cp #1 checks if a file exists: doesn't exist > cp #2 checks if a file exists: doesn't exist > cp #1 creates the new file > cp #2 tries to create the file but it already exists, it prints this error I can't think of how else this warning could appear ("cp -R" should overwrite existing files AFAIK) but doesn't that sound like an unlikely race condition to hit over 8000 times in one world build? I check random packages on the list and every time I found all three (ipk,rpm,deb) versions listed with a warning... as if this happened to every package. Incidentally I wonder if we should actually us "cp --no-clobber": should "publishing artifacts" ever overwrite files?
Changed the buildset-config for nightly-world-lsb so it does not publish artifacts. This eliminates the potential name conflict. Commit id: 69843cdf1eda6a1b52fa52eb4824ed8a52b9ae20