www-apache-bugdb mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ed Korthof ...@cloudfactory.org>
Subject Re: mod_jserv/5755: mod_jserv and httpd -HUP restarting problem
Date Sun, 09 Apr 2000 04:40:00 GMT
The following reply was made to PR mod_jserv/5755; it has been noted by GNATS.

From: Ed Korthof <ed@cloudfactory.org>
To: Vladislav Malyshkin <mal@mail1.nai.net>
Cc: jon * <jon@clearink.com>, apbugs@apache.org
Subject: Re: mod_jserv/5755: mod_jserv and httpd -HUP restarting problem
Date: Sat, 8 Apr 2000 21:32:54 -0700 (PDT)

   This message is in MIME format.  The first part should be readable text,
   while the remaining parts are likely unreadable without MIME-aware tools.
   Send mail to mime@docserver.cac.washington.edu for more info.
 Content-Type: TEXT/PLAIN; CHARSET=x-user-defined
 Content-ID: <Pine.LNX.3.96.1000408212754.20243H@stratus.cloudfactory.org>
 Hi Vladislav --
 I'm still of the opinion that in normal operating conditions, this isn't a
 bug; but after thinking about it, I also can't see any reason why it'd
 hurt to make these changes.  Defensive programming, and all that.
 However, a simpler solution to the problem you were having would be to
 move the error log at the same time you move the access log: that way only
 one HUP is required.  Is there a standard script which can only do this
 with two separate HUPs?   'Cause that's a bug, as I see it: it means extra
 downtime, and the possibility of confusion since the normal (orderly)
 restart process will generally be interrupted.  That's not how log
 rotation should happen.
 In any case, thanks for the patch and for persuing this; if I'd been less
 busy, I'd have gotten to it earlier.
 On Sat, 8 Apr 2000, Vladislav Malyshkin wrote:
 > Something strange. It may be netscape screw up some spaces in
 > my letter. Now I am sending the patch in encoded form.
 > I am sending 3 files:
 > original file: 19385 Oct 31 06:27 jserv_wrapper_unix.c.original
 > patches file: 20038 Mar  7 13:54 jserv_wrapper_unix.c
 > the diff -u file:  5804 Apr  8 22:59 patch_kill_JVM_ON_HUP
 > It applies cleanly (at least for me).
 > I can not do diff aganist CVS version right now,
 > I will be able to do this some time next week.
 > Also, I think some things may be improved in my patch,
 > I wrote it in a great rush because it was a problem
 > with production server, which was necessary to fix ASAP.
 > I can look at my patch again later next week.
 > The ideas behind my patch are:
 > 1. Add -HUP handler to watchdog process. This process is created by a separate fork()
 >     and has no standard signal haldlers set by apache. Another option may be to set
signal handler
 >     by apache initialization functions. I did not do this. I added it directly.
 > 2. Add signal handler inside signal handler (because on SYSV siognals
 >     it will be reset after first signal was delivered. Because of this we may have a
 >     when several signals are delivered during very short period of time).
 >     Without this logrotate causes a problem because it delivers two signals
 >     during very short period of time. First one - after access_log rotation, second
one - after error_log rotation.
 > 3. Always ping JVM before starting a new one (even if we were not able to start JVM
process before).
 >     If ping succedded - send EXIT command via servlet protocol.
 >     This allows us to stop a runaway JVM if we have any. We stop a runaway JVM process
 >     and then start a new one in accordance with standard jserv_wrapper_unix.c logic.
 > Items 1 and 3 are done well in my patch. About item 2 - I am not sure. I think it may
be done
 > in a more clean way. I can look at it again some time next week.
 > Vladislav
 > jon * wrote:
 > > can you send me your patch as an enclosure directly?
 > >
 > > it isn't applying very well to the source code.
 > >
 > > also, it would be best if you could send me the diff -U against the current
 > > CVS source code.
 > >
 > > thanks,
 > >
 > > -jon

View raw message