Is the file corruption problem currently being discussed on LKML related to the RPM/Yum seg-faults I occasionally on my 300 MHz, PII, notebook (with only 256 MB of RAM)?
I usually see the seg-faults while glibc-common is being updated during a large update (within the first 2 or 3 files of 30+ file update). If I update glibc* first, that, and the following updated of the remainder always works. If glibc has a lot of dependency info, I can see the /var/lib/rpm/__db.00* databases taking a hit and requiring manual deletion to restart.
If this is true, this bug might have started before FC6 was released, but only gets hit when memory runs low.
-Paul
On Sun, 24 Dec 2006 09:16:53 -0700, Paul Dickson wrote:
Is the file corruption problem currently being discussed on LKML related to the RPM/Yum seg-faults I occasionally on my 300 MHz, PII, notebook (with only 256 MB of RAM)?
I usually see the seg-faults while glibc-common is being updated during a large update (within the first 2 or 3 files of 30+ file update). If I update glibc* first, that, and the following updated of the remainder always works. If glibc has a lot of dependency info, I can see the /var/lib/rpm/__db.00* databases taking a hit and requiring manual deletion to restart.
If this is true, this bug might have started before FC6 was released, but only gets hit when memory runs low.
I forgot to mention the notebook is running kernel-2.6.18-1.2798.fc6 which is 2.6.18.1.
-Paul
On Sun, 24 Dec 2006, Paul Dickson wrote:
On Sun, 24 Dec 2006 18:18:43 +0100, Arjan van de Ven wrote:
the bug wasn't seen prior to 2.6.19 so I doubt what you see is related to it...
In the note about 2.6.20-rc2 says its been there since at least 2.6.18.3.
Yup, and it certainly affects at least apt-rpm in current Fedora 5 & 6 kernels. I've seen a bunch of crash reports (and managed to reproduce it a few times myself) which all are caused by zeroes in apt's cachefile where data is expected...
In the cases I managed to reproduce it, no rpmdb corruption was present, just apt's own cachefile. But as rpm uses mmap it's at least *possible* some of the rpm issues people see in FC6 are caused by it.
- Panu -
I also suspect the apt-rpm segfaults which have started to appear with the 2.6.18 kernel update in FC5 and are still there in FC6 to be a kernel bug. (Search Bugzilla, there's already a couple reports here, and there's similar reports against Debian's apt, also with 2.6.18.)
Kevin Kofler