It appears that .siginfo files are not being accessed from local or NFS-mounted SSTATE_MIRROR (file://), so their timestamp never gets updated. If you periodically clean SSTATE_MIRROR based on access time, they get deleted, even though they are still being used, leading to errors like: WARNING: quilt-native-0.64-r0 do_populate_sysroot_setscene: Failed to fetch URL file://Ubuntu-16.04/52/sstate:quilt-native:x86_64-linux:0.64:r0:x86_64:3:5227556845cbc224f05726d1633de4e5_populate_sysroot.tgz.siginfo, attempting MIRRORS if available ERROR: quilt-native-0.64-r0 do_populate_sysroot_setscene: Fetcher failure: Unable to find file file://Ubuntu-16.04/52/sstate:quilt-native:x86_64-linux:0.64:r0:x86_64:3:5227556845cbc224f05726d1633de4e5_populate_sysroot.tgz.siginfo anywhere. The paths that were searched were: In this case, corresponding file://Ubuntu-16.04/52/sstate:quilt-native:x86_64-linux:0.64:r0:x86_64:3:5227556845cbc224f05726d1633de4e5_populate_sysroot.tgz is still there, but not its .siginfo Everything works fine for HTTP-based SSTATE_MIRROR, because .tgz and their corresponding .siginfo files are being downloaded, updating both files timestamps on the server. So, regardless of SSTATE_MIRROR access protocol, when .tgz is being used, corresponding .siginfo file should probably be "touch"-ed to update it's timestamp to prevent this issue.
Any updates or comments on this?
I'm sorry, I haven't looked at this yet. I'm planning to do it after I fix all 2.3 M3 bugs assigned to me with higher priority.
Can you provide more info please? How your sstate NFS share is mounted? Did you use noatime or relatime option to mount it?
It uses "relatime" by default. Should I try something else?
No, you shouldn't unless you want atime to be updated immediately when file is accessed. relatime is delaying atime updates as far as I know. I reproduced this bug with strictatime mount option. Hopefully will fix it soon.
Please review: http://lists.openembedded.org/pipermail/openembedded-core/2017-March/133500.html
sent v2: http://lists.openembedded.org/pipermail/openembedded-core/2017-March/133701.html
The fix has been merged to oe-core and poky master branches.
Thanks!