Re: [GNUnet-developers] bugs. this and that.

[Igor Wronsky]

On Thu, 20 Jun 2002, Christian Grothoff wrote:

> CTRL-C is as safe as it currently gets since there is no propper shutdown 
> implemented (I've started to draft some code for individual modules, but not 
> much yet). That's also why you will never see HANGUP messages in the current 
> GNUnet. On the positive side, the entire code has been written to tolerate 
> very abrupt and inconvenient (kill -9) ways to die :-). 

Good, better to be on the safe side.

> gdbm failed on lseek!? How can that happen? 4GB database? :-). 

Dunno. Thats, nevertheless, the last thing gnunetd (or libgdbm,
actually) croaked before sweet oblivion. Might very well be
due to the database being somehow corrupted. If this problem
reoccurs, I'll try to make a stub to iterate through the 
keys to see if the lib crashes on one of them. 


A more baffling and less-easy-to-track problem I just had is
that gnunet-download hanged gnunetd, symptons were over
95% cpu load and almost total stop in network activity (udp flow went
down to dripping, none of the gnunet stats grew anymore, though 
they still could be queried for - so gnunetd wasn't totally
jammed, only just enough to stop everything useful). gnunet-download
did get >1MB of 4MB before gnunetd went postal. I'll compile 
gnunetd with gprof and try to reproduce the effect and 
find the load source.


Igor

ps. if the data files change (like the collisions recently?!)
and old ones need deletion for correct operation, perhaps 
that could be mentioned in the ChangeLog.