Latets yum thoroughly hosed?!

[Jim Cornette]

George R Goffe wrote:
> Howdy,
> 
> My two cents.
> 
> I'm having the same yum problem that everyone is/was having. I have
> run these commands:
> 
> rpm -e sqlite-devel-3.3.3-1.i386 \ rpm-devel-4.4.2-13.i386
> net-snmp-devel-5.3-3.i386
> 
> and then this one:
> 
> rpm -Uvh --oldpackage sqlite-3.3.2-1.i386.rpm
> 
> All appears to be cool now except that libneon is causing(?) a
> missing dependency:
> 
> Error: Missing Dependency: libneon.so.24 is needed by package kdesvn
> Error: Missing Dependency: libneon.so.24 is needed by package
> openoffice.org-core
> 
> Not provided by any rpm?
> 
> 
> The question about what's killing the yum process might be answered
> by the presence of a "new" (new to the later 2.4 kernels anyway)
> kernel option (I forget the exact terminology) "oom killer?". I
> think it's either this "feature" (that made it into the 2.6 kernels
> as a permanent fixture) or the system just running out of memory and
> killing the process as part of a recovery. I let my system run for a
> while (completely disabled apparently) when the "killed" message
> appeared. On subsequent attempts to use yum I watched xosview and it
> showed a steady increas in paging and swap and memory usage which
> you might expect given the previous reports. I have a "strace -xvf
> yum update" output file that shows what's happening. Of course you
> have to have strace installed, apparently it does NOT come with a
> "default" system install, whatever that is. Yum(?) or someone is
> looping on a brk system call. I haven't analyzed the trace yet.

Thanks for the view regarding what is going wrong with the "feature" 
which seems to becoming more of a vulnerability with the systems instead 
of allowing tasks that own memory to continue using their allocated 
memory. The idea of a system killing processes in order to make more 
memory available for later run applications does not sound like rational 
reasoning. If this is the case, the latest application started should 
pause instead of killing previous processes.
This problem with sqlite sounds similar to discussions related to 
bombing systems, where the system was put into a bad state by repeatdly 
starting processes in a loop. If this "feature is added to the kernel, 
the kernel needs to also limit saturation of swap or memory.

Just my thoughts regarding having even the possibility to down a system 
with OOM possibilities. Those that think that downing systems will 
surely enjoy the thrill of having Linux with this sort of vulnerability.

Jim

> 
> Regards,
> 
> George...
> 
> p.s., This group is GREAT!
> 
-- 
Where do you want Bill Gates to go today?

    -- From a Slashdot.org post