www-apache-bugdb mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chris Bradley <furr...@iglou.com>
Subject general/3174: TransferLog with pipe hangs server on SIGHUP restart
Date Sat, 10 Oct 1998 03:56:52 GMT

>Number:         3174
>Category:       general
>Synopsis:       TransferLog with pipe hangs server on SIGHUP restart
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    apache
>State:          open
>Class:          sw-bug
>Submitter-Id:   apache
>Arrival-Date:   Fri Oct  9 21:00:02 PDT 1998
>Originator:     furrier@iglou.com
>Release:        1.3.2 and 1.3.3
Sun Sparc / Solaris 2.5.1 at latest recommended patch level as of October 9th
GCC 2.8.1

Uname:  SunOS mayserv 5.5.1 Generic_103640-20 sun4u sparc SUNW,Ultra-1
Apache 1.3.3 (as well as 1.3.2) experiences a hang during a SIGHUP invoked
restart if the directory "LogTransfer" is set to a piped program under
Solaris 2.5.1.

It doesn't matter what the piped program does.  You could have a sh script
of simply "cat > /tmp/foobar" and the problem is still reproducable.

Basically, when doing a SIGHUP on the server, it will kill off all children
processes *except* the TransferLog script.  For some reason, it wont die
with a regular SIGTERM that Apache tries to send it.  Eventually, however,
Apache appears to get more forceful and eventually kills it off with a -9
signal (I suppose), but it takes almost 20 minutes it seems.  After that,
it works fine.

Most curiously, a graceful restart (with a SIGUSR1) works perfectly every
time with no delays at all.  The piped process dies immediately and restarts
like it should.

It almost appears like a signal masking problem with SIGHUP restarts with
regards to TransferLog piped processes.
Take any Solaris system running 2.5.1 (or maybe even 2.6) and set the TransferLog to a piped
program.  Make the piped program do something simple
like "cat > /tmp/foobar".

Then, start apache.  After it's running, SIGHUP the server.  You will see
all the children httpd's die off immediately, but the server will then
hang for up to 15 minutes.

If you then start all over and use a SIGUSR1 instead, it will work fine.

Not familiar enough with Apache's or Solaris' signaling code to recommend a fix.
[In order for any reply to be added to the PR database, ]
[you need to include <apbugs@Apache.Org> in the Cc line ]
[and leave the subject line UNCHANGED.  This is not done]
[automatically because of the potential for mail loops. ]
[If you do not include this Cc, your reply may be ig-   ]
[nored unless you are responding to an explicit request ]
[from a developer.                                      ]
[Reply only with text; DO NOT SEND ATTACHMENTS!         ]

View raw message