Yum Repo Broken for F15?

[Ralf Corsepius]

On 12/14/2011 09:25 AM, Chris Johns wrote:
> On 14/12/11 1:44 PM, Ralf Corsepius wrote:
>>
>> BTW: Was this from inside OAR's network?
>> Several of my gdb-build jobs have died with bizarre build errors, I have
>> no real explanation for. Some of them look like effects of a networking
>> issues other like dying HW. An explanation bor both would be a
>> man-in-the-middle attack or a misconfigured router/machine on OAR's
>> network (e.g. duplicate IP addresses)
>>
>
> Is this over nfs ?
Indirectly. The $(pwd) the scripts were launched from, the directory the 
log files are being stored on and the directories the results are 
supposed to be stored on are on nfs.


Building itself doesn't use nfs, however retrieves files world-wide via 
ftp and http.

The symptoms I saw in my logs were unlike everything, I've ever seen 
before.

1) some build jobs seem to have died midst of ftp and http file 
retrievals _without error_ (log files simply ended abruptly)
Rerunning the identical jobs this morning, they finished without complaint.

2) some build jobs died with ordinary time outs during ftp/http file 
retrieval.
Rerunning the identical jobs this morning, they finished without complaint.

3) some build log messages were duplicated, similar to
...
makemake enteringentering directorydirectory foobarfoobar
...

3) could be a bug related to parallel building (The builds jobs are 
parallel)

But wrt. 1) and 2), I can only guess:
- System overload (The load is very high)? However, such cases being 
able to bring a system down should be a subject of the past and in fact, 
they usually don't bring the build servers down.
- Network overload? The traffic these build jobs produce is temporarily 
high, but beyond unusal.
- Router/nat-failure? Dunno - Currently my favorite, because there are 
moments, it's imaginable 10-20 processes will be poking around the 
_internet_ through various protocols simultaneously at very high rate.

Ralf