httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Brian Behlendorf <>
Subject Re: [STATUS] 1.3b1: Tue Sep 30 10:00:33 PDT 1997
Date Tue, 30 Sep 1997 23:13:20 GMT
At 03:47 PM 9/30/97 -0700, Brian Behlendorf wrote:
>At 10:02 AM 9/30/97 -0700, Dean Gaudet wrote:
>>    * send_fb doesn't do aborts right
>Presuming this is the patch which you suggested as a fix to my
>graceful-restart-not-timing-out-nph-scripts bug, that appears to not have
>been the right fix.  I'm running the latest CVS tree on now, and
>I've got one child still hung in "G" a full 2500 seconds after the restart
>(I'm using the default TimeOut, 1200 seconds). 
>Blasted nph scripts.  I'd force them to be regular cgi now that cgi's
>unbuffered, but this seems like a useful bug to fix.  By the way all it
>takes is hitting "stop" in the browser to cause the nph script to live

let me rephrase the problem for those of you just tuning in;

nph- scripts without end (i.e. all those "server-push" scripts,
multipart/x-mixed-replace, etc) will seriously not end; a connection close
by the client won't be noticed, and the nph program will spin indefinitely.
 They don't burn much CPU or memory; they just sit there, bound to a child,
and when a graceful restart is attempted in order to rotate logs, the child
doesn't time out and stays in "G" mode, which also means it never lets go
of its file descriptor to the logs.  This is the current code base, and
it's a recent (last couple of weeks) problem, but I haven't narrowed it
further.  It'd be wise to test it with 1.2.4 and 1.3 as of right before the
nph- changes.

In my opinion, this isn't a showstopper, but should be highlighted as a
known bug, since any site which makes any use of server-push scripts will
have problems.


"it's a big world, with lots of records to play." - sig

View raw message