From e7bfc1c4e63e0ac76a93cb334007c791732b200b Mon Sep 17 00:00:00 2001
From: Daniel Nichter <daniel@percona.com>
Date: Tue, 3 Nov 2015 15:32:55 -0800
Subject: [PATCH] Don't die on slave lag check, return undef and wait forever.

---
 bin/foo.pm                  | 21 +++++++++++++++++++++
 bin/pt-online-schema-change | 22 +++++++++++++++++-----
 2 files changed, 38 insertions(+), 5 deletions(-)
 create mode 100644 bin/foo.pm

diff --git a/bin/foo.pm b/bin/foo.pm
new file mode 100644
index 00000000..bc4c9a56
--- /dev/null
+++ b/bin/foo.pm
@@ -0,0 +1,21 @@
+package pt_online_schema_change_plugin;
+
+use strict;
+use warnings;
+
+sub new {
+   my ( $class, %args ) = @_;
+   my $self = {};
+   return bless $self, $class;
+}
+
+sub before_swap_tables {
+   my ($self, %args) = @_;
+   print `mysql -e "select * From mysql.innodb_index_stats where database_name='test'"`;
+   print `mysql -e "select * From mysql.innodb_table_stats where database_name='test'"`;
+   sleep 12;
+   print `mysql -e "select * From mysql.innodb_index_stats where database_name='test'"`;
+   print `mysql -e "select * From mysql.innodb_table_stats where database_name='test'"`;
+}
+
+1;
diff --git a/bin/pt-online-schema-change b/bin/pt-online-schema-change
index 435d9c6b..9fc55371 100755
--- a/bin/pt-online-schema-change
+++ b/bin/pt-online-schema-change
@@ -8462,13 +8462,25 @@ sub main {
             if ( !$dbh || !$dbh->ping() ) {
                eval { $dbh = $cxn->connect() };  # connect or die trying
                if ( $EVAL_ERROR ) {
-                  $oktorun = 0;  # flag for cleanup tasks
-                  chomp $EVAL_ERROR;
-                  die "Lost connection to replica " . $cxn->name()
-                     . " while attempting to get its lag ($EVAL_ERROR)\n";
+                  # As the docs say: "The tool waits forever for replicas
+                  # to stop lagging.  If any replica is stopped, the tool
+                  # waits forever until the replica is started."
+                  # https://bugs.launchpad.net/percona-toolkit/+bug/1402051
+                  PTDEBUG && _d('Cannot connect to', $cxn->name(), ':',
+                     $EVAL_ERROR);
+                  # Make ReplicaLagWaiter::wait() report slave is stopped.
+                  return undef; 
                }
             }
-            return $ms->get_slave_lag($dbh);
+            my $lag;
+            eval {
+               $lag = $ms->get_slave_lag($dbh);
+            };
+            if ( $EVAL_ERROR ) {
+               PTDEBUG && _d('Cannot get lag for', $cxn->name(), ':',
+                  $EVAL_ERROR);
+            }
+            return $lag; # undef if error
          };
       }