Package: disorderfs Version: 0.5.1-1 Severity: normal Hi,
so we had disorderfs disabled for a long time on our reproducible setup and yesterday I enabled it again for i386 only, which rather immediatly caused profitbricks-build2-i386.debian.net to hang/die… so I rebootet it and logged in to investigate… and while watched diskspace… and suddenly the host was "gone". I observed this as I typed "ps fax|grep wesnot" in another shell. Also Trying to ssh in resulted in this: $ ssh profitbricks-build2-i386.debian.net Last login: Wed Nov 16 10:14:27 2016 from … and hangs there. So "killing the host" means: the kernel and processes are still running but are stuck on io, eg I had "watch df" running in an ssh session, it was not updating anymore but I could ctrl-c it and get back to the shell prompt. as soon as i pressed enter on a command, it hung. Interesting, disorderfs was used on two hosts, but this only happened on pb2, but not pb6! pb2 is running 3.16.0-4-686-pae while pb6 is running 3.16.0-4-amd64 (and both run i386 userland). Thogh it's unclear whether it would not also "kill" pb6 if we were using it long enough. Also, using disorderfs for a few jobs seems to be fine, but there are 8 pbuilder jobs running simultaenously on each of these hosts and half of them are using disorderfs. Next easy tests: - enable disorderfs on some armhf hosts (running 32bit kernels too) - upgrade pb2 to a 4.7 bpo 686 kernel - eneable disorderfs on amd64 -- cheers, Holger
signature.asc
Description: Digital signature
_______________________________________________ Reproducible-builds mailing list Reproducible-builds@lists.alioth.debian.org http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/reproducible-builds