weird crash

["Roy T. Fielding" <fi...@kiwi.ICS.UCI.EDU>]

I've been noticing something weird about running my test server.
If the server is running in the background when I re-make the
executable, the parent server just dies.  There is nothing in the
error_log, and no core file is present.  All the children die on the
next request to the server.

   % uname -a
   SunOS kiwi.ics.uci.edu 5.5 Generic sun4m sparc SUNW,SPARCstation-4

I have seen this happen about five times, though I was only able to
reproduce it just now.  I never saw it prior to 1.2-dev. Anyone have
any idea why the running server would be dependent on the executable file?

.....Roy

Re: weird crash

[Cliff Skolnick <cl...@steam.com>]

I assume the target of your compile is the same file that the executable
is running.  On UNIX systems that dynamically page executables from the
file system (all BSD and SVR4 based), when you copy new data over a
running executable and some gets paged in wierd things will happen.  You
can protect against this by either doing an "mv httpd httpd-" or even an
"rm httpd" before writing out the executable.

Why does this happen only sometimes?  Well if you are lucky enough not to 
have to dynamically page anything in before you restart the program, you 
will see no problem.

HTTPD is not alone, you can do the same thing to most any daemon that
pages by doing a "cp"  atop a running executable.  A "mv" will not do it
since under the hood a "mv" does a logical "rm" before messing with
anything. 

Cliff

On Sat, 30 Nov 1996, Roy T. Fielding wrote:

> I've been noticing something weird about running my test server.
> If the server is running in the background when I re-make the
> executable, the parent server just dies.  There is nothing in the
> error_log, and no core file is present.  All the children die on the
> next request to the server.
> 

--
Cliff Skolnick, Technical Consultant
Steam Tunnel Operations
cliff@steam.com, 415.297.5938
http://www.steam.com/