As Daniel said, the server cristal.inria.fr, which centralizes the
results of the ECDL7 challenge, experienced two related problems
over the week-end, making it unable to receive distinguished points
sent in "http" mode, as well as batches of points sent via the Web
interface http://cristal.inria.fr/bin/ecdlresult, during the following
time intervals:
Saturday 20th, 7:40 am - 7:45 am French time (6:40 - 6:45 UTC)
Sunday 21st, 5:00 am - 10:30 am French time (4:00 - 9:30 UTC)
Clients running in "mail" or "alt" mode are not affected. If this is
the case for your clients, no points were lost and you can stop
reading this message at this point.
Clients running in "batch" mode are not affected either. However, you
may have been unable to send your batch of points if you tried to use
the Web interface at http://cristal.inria.fr/bin/ecdlresult at the
times listed above. (You probably got an "Internal server error" or
something similar.) If so, please try again to send your points.
Batches of points sent manually via e-mail were received all right.
Clients running in "http" mode are affected: all distinguished points
found during the time intervals above were not recorded. This represents
about 1800 points according to the server logs.
To see if your ECDL client running in "http" mode failed to send
points, look at the log file ecdl.log in the directory containing the
ECDL client, and search for lines that say
Warning: an error occurred while sending point via HTTP, send by hand!
For instance, here is what I found in the logs of one of my machines:
Distinguished point found at: Sun Feb 20 09:06:36 2000
ECC-L0|Unix32|1.1.1 MMX|
ECC-L1|8C59982DEDD00017C0A4FE675A7|CC8B624D8FC4A59D96FDC237181|
ECC-L2|Xavier.Leroy@inria.fr|
ECC-L3|00005FEF7AEA|
ECC-L4|estephe|
Iterations used = 1.60953e+09.
Total iterations = 2.4068e+10.
Rate = 144320 per second.
Warning: Web server or proxy cristal.inria.fr returned an error
(status code = 500)
Warning: an error occurred while sending point via HTTP, send by hand!
If you don't see any "send by hand!" lines in the log, then you were
lucky and you can stop reading this message now.
If you do see "send by hand" lines like the above, then please launch
your Web browser, point it to
http://cristal.inria.fr/bin/ecdlresult
and cut-and-paste the points that couldn't be sent in the
"Distinguished points" text area. The lines you need to cut-and-paste
are those starting with "ECC-L". Then, click the "Send" button.
That's it! Your lost points are now safely recorded.
For those of you who are curious about what happened exactly,
here is an executive summary:
- On Saturday, at 7:40, the "cristal" server did an unattended reboot
because of a "System uncorrectable machine check" (that's Digital's
newspeak for "a hardware glitch").
- The reboot exposed an NFS problem on another server.
- The NFS problem caused a file synchronization operation between
cristal and the other server, performed at 5:00 on Sunday, to fail.
- The failure of the synchronization affected the Perl interpreter on
cristal.
- The CGI scripts that handle points sent via HTTP are written in
Perl, and thus stopped working.
Quite an unlikely sequence of events, none of which was really
critical in itself... At any rate, the problem is now fixed, and I
will take further steps to harden the "cristal" server against that
kind of failures.
Please accept my personal apologies for the inconvenience of resending
points by hand.
- Xavier Leroy
This archive was generated by hypermail 2b29 : Sun Feb 20 2000 - 17:16:27 MET