From e9e8285b5446407208d5ee42d6b43d0576e8a3d2 Mon Sep 17 00:00:00 2001
From: Baron Schwartz <baron@xaprb.com>
Date: Mon, 4 Jun 2012 11:10:41 -0400
Subject: [PATCH] pt-slave-delay doesn't reconnect reliably, this is a start

---
 bin/pt-slave-delay              |  9 +++++++++
 t/pt-slave-delay/auto_restart.t | 33 ++++++++++++++++++++++++++++++---
 2 files changed, 39 insertions(+), 3 deletions(-)

diff --git a/bin/pt-slave-delay b/bin/pt-slave-delay
index ddbb946d..99947767 100755
--- a/bin/pt-slave-delay
+++ b/bin/pt-slave-delay
@@ -2117,6 +2117,7 @@ sub main {
 
       $now = time();
 
+      # TODO: this is a race condition. See 0xdeadbeef below.
       if ( !$slave_dbh || !$slave_dbh->ping() ) {
          # Try 10 times, for about 2 minutes, to reconnect to the slave,
          # increasing wait time from 3 to 15 seconds.
@@ -2151,6 +2152,8 @@ sub main {
          );
          last unless $oktorun;  # might have gotten interrupt while waiting
       }
+      # 0xdeadbeef (see above): just because we reconnected in the above Retry
+      # does not mean we have a connection here!
       $status = $slave_dbh->selectrow_hashref("SHOW SLAVE STATUS");
       if ( !$status || ! %$status ) {
          die "No SLAVE STATUS found";
@@ -2456,6 +2459,12 @@ C<pt-slave-delay> with Control-C.
 Fork to the background and detach from the shell.  POSIX
 operating systems only.
 
+=item --database
+
+short form: -D; type: string
+
+The database to use for the connection.
+
 =item --defaults-file
 
 short form: -F; type: string
diff --git a/t/pt-slave-delay/auto_restart.t b/t/pt-slave-delay/auto_restart.t
index 7dc81d63..3e26b22f 100644
--- a/t/pt-slave-delay/auto_restart.t
+++ b/t/pt-slave-delay/auto_restart.t
@@ -12,6 +12,7 @@ use strict;
 use warnings FATAL => 'all';
 use English qw( -no_match_vars );
 use Test::More;
+use Data::Dumper;
 
 use PerconaTest; 
 use Sandbox;
@@ -28,7 +29,7 @@ elsif ( !@{$dbh->selectcol_arrayref('SHOW DATABASES LIKE "sakila"')} ) {
    plan skip_all => 'sakila db not loaded';
 }
 else {
-   plan tests => 1;
+   plan tests => 2;
 }
 
 my $cnf = '/tmp/12346/my.sandbox.cnf';
@@ -47,7 +48,7 @@ my $output;
 my $pid = fork();
 if ( $pid ) {
    # parent
-   $output = `$cmd --interval 1 --run-time 4 2>&1`;
+   $output = `$cmd --interval 1 --run-time 8 2>&1`;
    like(
       $output,
       qr/Lost connection.+?Reconnected to slave.+Setting slave to run/ms,
@@ -58,13 +59,39 @@ else {
    # child
    sleep 1;
    diag(`/tmp/12346/stop >/dev/null`);
-
    sleep 1;
    diag(`/tmp/12346/start >/dev/null`);
    diag(`/tmp/12346/use -e "set global read_only=1"`);
    exit;
 }
+# Reap the child.
+waitpid ($pid, 0);
 
+# Do it all over again, but this time KILL instead of restart.
+$pid = fork();
+if ( $pid ) {
+   # parent. Note the --database mysql
+   $output = `$cmd --database mysql --interval 1 --run-time 8 2>&1`;
+   like(
+      $output,
+      qr/Lost connection.+?Reconnected to slave.+Setting slave to run/ms,
+      "Reconnect to slave"
+   );
+}
+else {
+   # child. Note that we'll kill the parent's 'mysql' connection
+   sleep 1;
+   my $c_dbh = $sb->get_dbh_for('slave1');
+   my @cxn = @{$c_dbh->selectall_arrayref('show processlist', {Slice => {}})};
+   foreach my $c ( @cxn ) {
+      # The parent's connection:
+      # {command => 'Sleep',db => 'mysql',host => 'localhost',id => '5',info => undef,state => '',time => '1',user => 'msandbox'}
+      if ( ($c->{db} || '') eq 'mysql' && ($c->{user} || '') eq 'msandbox' ) {
+         $c_dbh->do("KILL $c->{id}");
+      }
+   }
+   exit;
+}
 # Reap the child.
 waitpid ($pid, 0);