FINALLY find and fix the elusive parallel test deadlock

[dbsrgits/DBIx-Class.git] / t / lib / DBICTest / Util.pm

diff --git a/t/lib/DBICTest/Util.pm b/t/lib/DBICTest/Util.pm
index 4a8f2c2..74ba068 100644 (file)
--- a/t/lib/DBICTest/Util.pm
+++ b/t/lib/DBICTest/Util.pm
@@ -17,12 +17,78 @@ BEGIN {
   }
 }
 
+use constant DEBUG_TEST_CONCURRENCY_LOCKS =>
+  ( ($ENV{DBICTEST_DEBUG_CONCURRENCY_LOCKS}||'') =~ /^(\d+)$/ )[0]
+    ||
+  0
+;
+
 use Config;
 use Carp 'confess';
+use Fcntl ':flock';
 use Scalar::Util qw(blessed refaddr);
+use DBIx::Class::_Util;
 
 use base 'Exporter';
-our @EXPORT_OK = qw(local_umask stacktrace check_customcond_args);
+our @EXPORT_OK = qw(
+  dbg stacktrace
+  local_umask
+  visit_namespaces
+  check_customcond_args
+  await_flock DEBUG_TEST_CONCURRENCY_LOCKS
+);
+
+if (DEBUG_TEST_CONCURRENCY_LOCKS) {
+  require DBI;
+  my $oc = DBI->can('connect');
+  no warnings 'redefine';
+  *DBI::connect = sub {
+    DBICTest::Util::dbg("Connecting to $_[1]");
+    goto $oc;
+  }
+}
+
+sub dbg ($) {
+  require Time::HiRes;
+  printf STDERR "\n%.06f  %5s %-78s %s\n",
+    scalar Time::HiRes::time(),
+    $$,
+    $_[0],
+    $0,
+  ;
+}
+
+# File locking is hard. Really hard. By far the best lock implementation
+# I've seen is part of the guts of File::Temp. However it is sadly not
+# reusable. Since I am not aware of folks doing NFS parallel testing,
+# nor are we known to work on VMS, I am just going to punt this and
+# use the portable-ish flock() provided by perl itself. If this does
+# not work for you - patches more than welcome.
+#
+# This figure esentially means "how long can a single test hold a
+# resource before everyone else gives up waiting and aborts" or
+# in other words "how long does the longest test-group legitimally run?"
+my $lock_timeout_minutes = 15;  # yes, that's long, I know
+my $wait_step_seconds = 0.25;
+
+sub await_flock ($$) {
+  my ($fh, $locktype) = @_;
+
+  my ($res, $tries);
+  while(
+    ! ( $res = flock( $fh, $locktype | LOCK_NB ) )
+      and
+    ++$tries <= $lock_timeout_minutes * 60 / $wait_step_seconds
+  ) {
+    select( undef, undef, undef, $wait_step_seconds );
+
+    # "say something" every 10 cycles to work around RT#108390
+    # jesus christ our tooling is such a crock of shit :(
+    print "#\n" if not $tries % 10;
+  }
+
+  return $res;
+}
 
 sub local_umask {
   return unless defined $Config{d_umask};
@@ -38,6 +104,8 @@ sub local_umask {
 {
   package DBICTest::Util::UmaskGuard;
   sub DESTROY {
+    &DBIx::Class::_Util::detected_reinvoked_destructor;
+
     local ($@, $!);
     eval { defined (umask ${$_[0]}) or die };
     warn ( "Unable to reset old umask ${$_[0]}: " . ($!||'Unknown error') )
@@ -50,7 +118,7 @@ sub stacktrace {
   $frame++;
   my (@stack, @frame);
 
-  while (@frame = caller($frame++)) {
+  while (@frame = CORE::caller($frame++)) {
     push @stack, [@frame[3,1,2]];
   }
 
@@ -106,4 +174,34 @@ sub check_customcond_args ($) {
   $args;
 }
 
+sub visit_namespaces {
+  my $args = { (ref $_[0]) ? %{$_[0]} : @_ };
+
+  my $visited_count = 1;
+
+  # A package and a namespace are subtly different things
+  $args->{package} ||= 'main';
+  $args->{package} = 'main' if $args->{package} =~ /^ :: (?: main )? $/x;
+  $args->{package} =~ s/^:://;
+
+  if ( $args->{action}->($args->{package}) ) {
+    my $ns =
+      ( ($args->{package} eq 'main') ? '' :  $args->{package} )
+        .
+      '::'
+    ;
+
+    $visited_count += visit_namespaces( %$args, package => $_ ) for
+      grep
+        # this happens sometimes on %:: traversal
+        { $_ ne '::main' }
+        map
+          { $_ =~ /^(.+?)::$/ ? "$ns$1" : () }
+          do { no strict 'refs'; keys %$ns }
+    ;
+  }
+
+  return $visited_count;
+}
+
 1;