termination problem mustgather

IBM HTTP Server customers occasionally encounter problems stopping the web server. The expected result of apachectl stop or sending the SIGTERM signal to the IHS parent process is that all httpd processes for that web server instance will exit within around thirty seconds. When this does not occur, IBM HTTP Server support needs data to be collected on the remaining httpd processes to help diagnose why the processes have not terminated.

When this type of problem occurs, the remaining httpd processes normally show "1" for their parent process id, as shown in this example:

$ ps -ef | grep httpd 
  nobody 15876     1   0   Feb 08      -  0:00   /usr/HTTPServer/bin/httpd
  nobody 17112     1   0   Feb 08      -  0:00   /usr/HTTPServer/bin/httpd
  nobody 17524     1   0   Feb 08      -  0:00   /usr/HTTPServer/bin/httpd
  nobody 19252     1   0   Feb 17      -  0:00   /usr/HTTPServer/bin/httpd
  nobody 20048     1   0   Feb 08      -  0:00   /usr/HTTPServer/bin/httpd
  nobody 20944     1   0   Feb 08      -  0:00   /usr/HTTPServer/bin/httpd
  nobody 21734     1   0   Feb 17      -  0:00   /usr/HTTPServer/bin/httpd
  nobody 24166     1   0   Feb 17      -  0:00   /usr/HTTPServer/bin/httpd

Before killing the processes manually, gather the required documentation, as directed below.

Special note for known problems

IBM HTTP Server 2.0.42.x

Before submitting documentation to IBM for analysis, ensure that you are using cumulative e-fix PK01070 or later to correct a pervasive termination problem.

IBM HTTP Server 2.0.47.x

Failure to terminate IBM HTTP Server on Unix and Linux when used with IBM Tivoli Monitoring

The apachectl stop command sends the SIGTERM signal to the httpd parent process in order to shut down the web server. When IBM HTTP Server is started under IBM Tivoli Monitoring, the SIGTERM signal may be masked by IBM Tivoli Monitoring. This will cause the apachectl stop command to be ineffective.

A work-around for this problem is to start IBM HTTP Server using a special Perl script which restores the SIGTERM signal before starting IBM HTTP Server.

1. Save this Perl script in a file called /usr/local/bin/unblocksigterm.pl:

   #!/usr/bin/perl -w                    
   use strict;                           
   use POSIX;                            
   my $how = SIG_UNBLOCK;                
   my $set = POSIX::SigSet->new(SIGTERM);
   if (!sigprocmask($how, $set)) {       
     die "sigprocmask failed: $!";       
   }                                     
   my $cmd = join(' ', @ARGV);           
   print "$0: running $cmd\n";           
   my $rc = system($cmd);                
   exit($rc != 0 ? 1 : 0);
   

2. Mark the Perl script as executable with chmod:

   # chmod +x /usr/local/bin/unblocksigterm.pl
   

3. Start IBM HTTP Server using unblocksigterm.pl by adjusting the command specified in the IBM Tivoli Monitoring configuration:

   • Old:
     /usr/HTTPServer/bin/apachectl start

   • New:
     /usr/local/bin/unblocksigterm.pl /usr/HTTPServer/bin/apachectl start

Gathering documentation

The hang documentation tool is used to gather the required documentation. When using the tool, specify auto as the value of the parent process id parameter, and specify - as the non-SSL address parameter, as in the following example:

# java -jar ServerDoc.jar GatherHangDoc /usr/HTTPServer auto -

After trying to stop the web server, wait at least sixty seconds before gathering documentation.

The hang documentation instructions have further information about the use of the tool.