mtr_timer.pl, mysql-test-run.pl, mtr_report.pl, mtr_process.pl:

  Added suite and test case timeout
mtr_timer.pl:
  new file

mysql-test/lib/mtr_process.pl

@@ -210,6 +210,7 @@ sub spawn_parent_impl {
       # normally from running a test case?
 
       my $exit_value= -1;
+      my $saved_exit_value;
       my $ret_pid;                      # What waitpid() returns
 
       while ( ($ret_pid= waitpid(-1,0)) != -1 )
@@ -219,6 +220,24 @@ sub spawn_parent_impl {
         # but not $exit_value, this is flagged from
         # 
 
+        my $timer_name= mtr_timer_timeout($::glob_timers, $ret_pid);
+        if ( $timer_name )
+        {
+          if ( $timer_name eq "suite" )
+          {
+            # We give up here
+            # FIXME we should only give up the suite, not all of the run?
+            print STDERR "\n";
+            mtr_error("Test suite timeout");
+          }
+          elsif ( $timer_name eq "testcase" )
+          {
+            $saved_exit_value=  63;       # Mark as timeout
+            kill(9, $pid);                # Kill mysqltest
+            next;                         # Go on and catch the termination
+          }
+        }
+
         if ( $ret_pid == $pid )
         {
           # We got termination of mysqltest, we are done
@@ -270,7 +289,7 @@ sub spawn_parent_impl {
         }
       }
 
-      return $exit_value;
+      return $saved_exit_value || $exit_value;
     }
   }
   else

mysql-test/lib/mtr_report.pl

@@ -109,7 +109,14 @@ sub mtr_report_test_failed ($) {
   my $tinfo= shift;
 
   $tinfo->{'result'}= 'MTR_RES_FAILED';
+  if ( $tinfo->{'timeout'} )
+  {
+    print "[ fail ]  timeout\n";
+  }
+  else
+  {
     print "[ fail ]\n";
+  }
 
   # FIXME Instead of this test, and meaningless error message in 'else'
   # we should write out into $::path_timefile when the error occurs.

mysql-test/lib/mtr_timer.pl

+# -*- cperl -*-
+
+# This is a library file used by the Perl version of mysql-test-run,
+# and is part of the translation of the Bourne shell script with the
+# same name.
+
+use Carp qw(cluck);
+use Socket;
+use Errno;
+use strict;
+
+#use POSIX ":sys_wait_h";
+use POSIX 'WNOHANG';
+
+sub mtr_init_timers ();
+sub mtr_timer_start($$$);
+sub mtr_timer_stop($$);
+sub mtr_timer_waitpid($$$);
+
+##############################################################################
+#
+#  Initiate a structure shared by all timers
+#
+##############################################################################
+
+sub mtr_init_timers () {
+  my $timers = { timers => {}, pids => {}};
+  return $timers;
+}
+
+
+##############################################################################
+#
+#  Start, stop and poll a timer
+#
+#  As alarm() isn't portable to Windows, we use separate processes to
+#  implement timers. That is why there is a mtr_timer_waitpid(), as this
+#  is where we catch a timeout.
+#
+##############################################################################
+
+sub mtr_timer_start($$$) {
+  my ($timers,$name,$duration)= @_;
+
+  if ( exists $timers->{'timers'}->{$name} )
+  {
+    # We have an old running timer, kill it
+    mtr_timer_stop($timers,$name);
+  }
+
+ FORK:
+  {
+    my $tpid= fork();
+
+    if ( ! defined $tpid )
+    {
+      if ( $! == $!{EAGAIN} )           # See "perldoc Errno"
+      {
+        mtr_debug("Got EAGAIN from fork(), sleep 1 second and redo");
+        sleep(1);
+        redo FORK;
+      }
+      else
+      {
+        mtr_error("can't fork");
+      }
+    }
+
+    if ( $tpid )
+    {
+      # Parent, record the information
+      $timers->{'timers'}->{$name}->{'pid'}= $tpid;
+      $timers->{'timers'}->{$name}->{'duration'}= $duration;
+      $timers->{'pids'}->{$tpid}= $name;
+    }
+    else
+    {
+      # Child, redirect output and exec
+      # FIXME do we need to redirect streams?
+      $0= "mtr_timer(timers,$name,$duration)";
+      sleep($duration);
+      exit(0);
+    }
+  }
+}
+
+
+sub mtr_timer_stop ($$) {
+  my ($timers,$name)= @_;
+
+  if ( exists $timers->{'timers'}->{$name} )
+  {
+    my $tpid= $timers->{'timers'}->{$name}->{'pid'};
+
+    # FIXME as Cygwin reuses pids fast, maybe check that is
+    # the expected process somehow?!
+    kill(9, $tpid);
+
+    # As the timers are so simple programs, we trust them to terminate,
+    # and use blocking wait for it. We wait just to avoid a zombie.
+    waitpid($tpid,0);
+
+    delete $timers->{'timers'}->{$name}; # Remove the timer information
+    delete $timers->{'pids'}->{$tpid};   # and PID reference
+
+    return 1;
+  }
+  else
+  {
+    mtr_debug("Asked to stop timer \"$name\" not started");
+    return 0;
+  }
+}
+
+
+sub mtr_timer_timeout ($$) {
+  my ($timers,$pid)= @_;
+
+  return "" unless exists $timers->{'pids'}->{$pid};
+
+  # We got a timeout
+  my $name= $timers->{'pids'}->{$pid};
+  mtr_timer_stop($timers, $timers->{'timers'}->{$name});
+  return $name;
+}
+
+1;

mysql-test/mysql-test-run.pl

@@ -90,6 +90,7 @@ use strict;
 
 require "lib/mtr_cases.pl";
 require "lib/mtr_process.pl";
+require "lib/mtr_timer.pl";
 require "lib/mtr_io.pl";
 require "lib/mtr_gcov.pl";
 require "lib/mtr_gprof.pl";
@@ -137,6 +138,7 @@ our $glob_mysql_test_dir=         undef;
 our $glob_mysql_bench_dir=        undef;
 our $glob_hostname=               undef;
 our $glob_scriptname=             undef;
+our $glob_timers=                 undef;
 our $glob_use_running_server=     0;
 our $glob_use_running_ndbcluster= 0;
 our $glob_use_embedded_server=    0;
@@ -234,6 +236,8 @@ our $opt_ps_protocol;
 
 our $opt_sleep_time_after_restart=  1;
 our $opt_sleep_time_for_delete=    10;
+our $opt_testcase_timeout=          5; # 5 min max
+our $opt_suite_timeout=           120; # 2 hours max
 
 our $opt_socket;
 
@@ -435,6 +439,8 @@ sub initial_setup () {
 
   $path_my_basedir=
     $opt_source_dist ? $glob_mysql_test_dir : $glob_basedir;
+
+  $glob_timers= mtr_init_timers();
 }
 
 
@@ -530,6 +536,8 @@ sub command_line_setup () {
              'vardir=s'                 => \$opt_vardir,
              'verbose'                  => \$opt_verbose,
              'wait-timeout=i'           => \$opt_wait_timeout,
+             'testcase-timeout=i'       => \$opt_testcase_timeout,
+             'suite-timeout=i'          => \$opt_suite_timeout,
              'warnings|log-warnings'    => \$opt_warnings,
              'with-openssl'             => \$opt_with_openssl,
 
@@ -1197,6 +1205,8 @@ sub run_suite () {
 
   mtr_report("Finding  Tests in the '$suite' suite");
 
+  mtr_timer_start($glob_timers,"suite", 60 * $opt_suite_timeout);
+
   my $tests= collect_test_cases($suite);
 
   mtr_report("Starting Tests in the '$suite' suite");
@@ -1205,7 +1215,9 @@ sub run_suite () {
 
   foreach my $tinfo ( @$tests )
   {
+    mtr_timer_start($glob_timers,"testcase", 60 * $opt_testcase_timeout);
     run_testcase($tinfo);
+    mtr_timer_stop($glob_timers,"testcase");
   }
 
   mtr_print_line();
@@ -1226,6 +1238,8 @@ sub run_suite () {
   }
 
   mtr_report_stats($tests);
+
+  mtr_timer_stop($glob_timers,"suite");
 }
 
 
@@ -1523,6 +1537,11 @@ sub run_testcase ($) {
       # Testcase itself tell us to skip this one
       mtr_report_test_skipped($tinfo);
     }
+    elsif ( $res == 63 )
+    {
+      $tinfo->{'timeout'}= 1;           # Mark as timeout
+      report_failure_and_restart($tinfo);
+    }
     else
     {
       # Test case failed, if in control mysqltest returns 1
@@ -1657,8 +1676,6 @@ sub mysqld_arguments ($$$$$) {
   my $extra_opt=         shift;
   my $slave_master_info= shift;
 
-#  print STDERR Dumper($extra_opt);
-
   my $sidx= "";                 # Index as string, 0 is empty string
   if ( $idx > 0 )
   {
@@ -2262,6 +2279,10 @@ Misc options
   help                  Get this help text
   unified-diff | udiff  When presenting differences, use unified diff
 
+  testcase-timeout=MINUTES Max test case run time (default 5)
+  suite-timeout=MINUTES    Max test suite run time (default 120)
+
+
 Options not yet described, or that I want to look into more
 
   big-test              
@@ -2281,4 +2302,5 @@ Options not yet described, or that I want to look into more
 
 HERE
   mtr_exit(1);
+
 }