Hi,
the problem is not only with a power outage. You should think that if
kannel crashes and you commented fflush out you will probably loss
backtrace in log file. For a access log it's even more critical because
you will loss messages that you sent.
It's up to you whether you will loss log or access log entries...
I would vote -1 for disabling fflush in general and +0 for disabling it
at compile time (patch welcome).
Thanks,
Alex
Mindaugas wrote:
Just to note about gwlib/log.c:output() and gwlib/accesslog.c:alog()
functions which IMHO is a base for the whole logging and both of them
uses platform's standard version of vfprintf implementation which
sometimes is buggy, slow and so on.
Shall Kannel implement it's own implementation or maybe one defined at
compile time?
I doubt that it will matter. Maybe Apache uses *printf functions for
serving
web pages so it's critical for them.
But I did very simple test. I wrote tiny C program which just writes into
file
and in one case it fflush() file handle after each fprintf() and does not do
that
in another case. Difference is huge:
$ time ./fflush ( without fflush() )
real 0m5.649s
user 0m1.635s
sys 0m3.836s
$ time ./fflush ( with fflush() )
real 1m32.389s
user 0m5.569s
sys 1m23.935s
So Stipe, Alexander - can I safely comment out fflush() after vfprintf? Or
have
configure or configuration file setting for that? :)
As I've already said - I do not care if I'll loose few lines of log in a
power outage.
Mindaugas
