List:Commits« Previous MessageNext Message »
From:Andrei Elkin Date:March 29 2011 2:56pm
Subject:bzr commit into mysql-trunk branch (andrei.elkin:3344)
View as plain text  
#At file:///home/andrei/MySQL/BZR/2a-23May/MERGE/mysql-trunk/ based on revid:andrei.elkin@stripped

 3344 Andrei Elkin	2011-03-29 [merge]
      merge from local bugfixing branch to local mysql-trunk prior to push

    modified:
      mysql-test/extra/rpl_tests/rpl_deadlock.test
      mysql-test/suite/rpl/r/rpl_deadlock_innodb.result
      sql/rpl_reporting.cc
      sql/rpl_reporting.h
      sql/rpl_slave.cc
=== modified file 'mysql-test/extra/rpl_tests/rpl_deadlock.test'
--- a/mysql-test/extra/rpl_tests/rpl_deadlock.test	2010-12-19 17:07:28 +0000
+++ b/mysql-test/extra/rpl_tests/rpl_deadlock.test	2011-03-27 18:16:32 +0000
@@ -123,6 +123,57 @@ SELECT * FROM t3;
 source include/check_slave_is_running.inc;
 --echo
 
+#
+# bug#11748510/36524 incident of deadlock on slave is overdramatized
+#
+# Observe that the slave stopped when the number of transation retries 
+# exceeds @@global.slave_transaction_retries
+#
+connection master;
+
+--echo *** Test the deadlock warning to be escalated into the error ***
+
+delete from t1;
+delete from t2;
+delete from t3;
+
+sync_slave_with_master;
+
+# make sure slave's unilateral row gone as well
+delete from t1;
+delete from t2;
+delete from t3;
+
+# the first attempt to run a deadlock scenario of p 1) leads to the error
+set @save.slave_transaction_retries= @@global.slave_transaction_retries;
+set @@global.slave_transaction_retries= 0;
+source include/stop_slave.inc;
+
+connection master;
+
+BEGIN;
+INSERT INTO t1 VALUES (1);
+# We make a long transaction here
+INSERT INTO t2 VALUES (2), (2), (2), (2), (2), (2), (2), (2), (2), (2);
+INSERT INTO t3 VALUES (3);
+COMMIT;
+
+connection slave;
+BEGIN;
+SELECT count(*) as zero FROM t1 FOR UPDATE;
+
+start slave;
+
+--echo *** Now the slave must be stopped due to timeout ***
+
+let $slave_sql_errno= 1205; # ER_LOCK_TIMEOUT
+let $show_slave_sql_error= 0;
+source include/wait_for_slave_sql_error.inc;
+
+rollback;
+
+set @@global.slave_transaction_retries= @save.slave_transaction_retries;
+source include/start_slave.inc;
 # Clean up
 --echo *** Clean up ***
 connection master;

=== modified file 'mysql-test/suite/rpl/r/rpl_deadlock_innodb.result'
--- a/mysql-test/suite/rpl/r/rpl_deadlock_innodb.result	2010-12-19 17:22:30 +0000
+++ b/mysql-test/suite/rpl/r/rpl_deadlock_innodb.result	2011-03-27 18:16:32 +0000
@@ -103,6 +103,31 @@ a
 3
 include/check_slave_is_running.inc
 
+*** Test the deadlock warning to be escalated into the error ***
+delete from t1;
+delete from t2;
+delete from t3;
+delete from t1;
+delete from t2;
+delete from t3;
+set @save.slave_transaction_retries= @@global.slave_transaction_retries;
+set @@global.slave_transaction_retries= 0;
+include/stop_slave.inc
+BEGIN;
+INSERT INTO t1 VALUES (1);
+INSERT INTO t2 VALUES (2), (2), (2), (2), (2), (2), (2), (2), (2), (2);
+INSERT INTO t3 VALUES (3);
+COMMIT;
+BEGIN;
+SELECT count(*) as zero FROM t1 FOR UPDATE;
+zero
+0
+start slave;
+*** Now the slave must be stopped due to timeout ***
+include/wait_for_slave_sql_error.inc [errno=1205]
+rollback;
+set @@global.slave_transaction_retries= @save.slave_transaction_retries;
+include/start_slave.inc
 *** Clean up ***
 DROP TABLE t1,t2,t3;
 SET global max_relay_log_size= @my_max_relay_log_size;

=== modified file 'sql/rpl_reporting.cc'
--- a/sql/rpl_reporting.cc	2010-08-05 17:45:25 +0000
+++ b/sql/rpl_reporting.cc	2011-03-28 13:19:08 +0000
@@ -17,6 +17,7 @@
 #include "rpl_reporting.h"
 #include "log.h" // sql_print_error, sql_print_warning,
                  // sql_print_information
+#include "rpl_slave.h"
 
 Slave_reporting_capability::Slave_reporting_capability(char const *thread_name)
   : m_thread_name(thread_name)
@@ -25,15 +26,94 @@ Slave_reporting_capability::Slave_report
                    &err_lock, MY_MUTEX_INIT_FAST);
 }
 
+#if !defined(EMBEDDED_LIBRARY)
+/**
+  Check if the current error is of temporary nature or not.
+  Some errors are temporary in nature, such as
+  ER_LOCK_DEADLOCK and ER_LOCK_WAIT_TIMEOUT.  Ndb also signals
+  that the error is temporary by pushing a warning with the error code
+  ER_GET_TEMPORARY_ERRMSG, if the originating error is temporary.
+
+  @param      thd  a THD instance, typically of the slave SQL thread's.
+  @error_arg  the error code for assessment. 
+              defaults to zero which makes the function check the top
+              of the reported errors stack.
+
+  @return 1 as the positive and 0 as the negative verdict
+*/
+int Slave_reporting_capability::has_temporary_error(THD *thd, uint error_arg) const
+{
+  uint error;
+  DBUG_ENTER("has_temporary_error");
+
+  DBUG_EXECUTE_IF("all_errors_are_temporary_errors",
+                  if (thd->stmt_da->is_error())
+                  {
+                    thd->clear_error();
+                    my_error(ER_LOCK_DEADLOCK, MYF(0));
+                  });
+
+  /*
+    The state of the slave thread can't be regarded as
+    experiencing a temporary failure in cases of @c is_slave_error was set TRUE,
+    or if there is no message in THD, we can't say if it's a temporary
+    error or not. This is currently the case for Incident_log_event,
+    which sets no message.
+  */
+  if (thd->is_fatal_error || !thd->is_error())
+    DBUG_RETURN(0);
+
+  error= (error_arg == 0)? thd->stmt_da->sql_errno() : error_arg;
+
+  /*
+    Temporary error codes:
+    currently, InnoDB deadlock detected by InnoDB or lock
+    wait timeout (innodb_lock_wait_timeout exceeded).
+    Notice, the temporary error requires slave_trans_retries != 0)
+  */
+  if (slave_trans_retries &&
+      (error == ER_LOCK_DEADLOCK || error == ER_LOCK_WAIT_TIMEOUT))
+    DBUG_RETURN(1);
+
+#ifdef HAVE_NDB_BINLOG
+  /*
+    currently temporary error set in ndbcluster
+  */
+  List_iterator_fast<MYSQL_ERROR> it(thd->warning_info->warn_list());
+  MYSQL_ERROR *err;
+  while ((err= it++))
+  {
+    DBUG_PRINT("info", ("has condition %d %s", err->get_sql_errno(),
+                        err->get_message_text()));
+    switch (err->get_sql_errno())
+    {
+    case ER_GET_TEMPORARY_ERRMSG:
+      DBUG_RETURN(1);
+    default:
+      break;
+    }
+  }
+#endif
+  DBUG_RETURN(0);
+}
+#endif // EMBEDDED_LIBRARY
+
+
 void
 Slave_reporting_capability::report(loglevel level, int err_code,
                                    const char *msg, ...) const
 {
+#if !defined(EMBEDDED_LIBRARY)
+  THD *thd= current_thd;
   void (*report_function)(const char *, ...);
   char buff[MAX_SLAVE_ERRMSG];
   char *pbuff= buff;
   uint pbuffsize= sizeof(buff);
   va_list args;
+
+  if (level == ERROR_LEVEL && has_temporary_error(thd, err_code))
+    level= WARNING_LEVEL;
+
   va_start(args, msg);
 
   mysql_mutex_lock(&err_lock);
@@ -51,10 +131,12 @@ Slave_reporting_capability::report(logle
     report_function= sql_print_error;
     break;
   case WARNING_LEVEL:
-    report_function= sql_print_warning;
+    report_function= global_system_variables.log_warnings?
+      sql_print_warning : NULL;
     break;
   case INFORMATION_LEVEL:
-    report_function= sql_print_information;
+    report_function= global_system_variables.log_warnings?
+      sql_print_information : NULL;
     break;
   default:
     DBUG_ASSERT(0);                            // should not come here
@@ -71,6 +153,7 @@ Slave_reporting_capability::report(logle
                   m_thread_name, pbuff,
                   (pbuff[0] && *(strend(pbuff)-1) == '.') ? "" : ",",
                   err_code);
+#endif  
 }
 
 Slave_reporting_capability::~Slave_reporting_capability()

=== modified file 'sql/rpl_reporting.h'
--- a/sql/rpl_reporting.h	2010-08-05 17:45:25 +0000
+++ b/sql/rpl_reporting.h	2011-03-28 13:19:08 +0000
@@ -23,6 +23,11 @@
  */
 #define MAX_SLAVE_ERRMSG      1024
 
+// todo: consider to remove rpl_reporting.cc,h from building embedded
+#if !defined(EMBEDDED_LIBRARY)
+class THD;
+#endif
+
 /**
    Mix-in to handle the message logging and reporting for relay log
    info and master log info structures.
@@ -65,6 +70,13 @@ public:
     mysql_mutex_unlock(&err_lock);
   }
 
+#if !defined(EMBEDDED_LIBRARY)
+  /**
+     Check if the current error is of temporary nature or not.
+  */
+  int has_temporary_error(THD *thd, uint error_arg= 0) const;
+#endif // EMBEDDED_LIBRARY
+  
   /**
      Error information structure.
    */

=== modified file 'sql/rpl_slave.cc'
--- a/sql/rpl_slave.cc	2011-03-29 13:44:23 +0000
+++ b/sql/rpl_slave.cc	2011-03-29 14:56:01 +0000
@@ -2518,63 +2518,6 @@ static ulong read_event(MYSQL* mysql, Ma
   DBUG_RETURN(len - 1);
 }
 
-/*
-  Check if the current error is of temporary nature of not.
-  Some errors are temporary in nature, such as
-  ER_LOCK_DEADLOCK and ER_LOCK_WAIT_TIMEOUT.  Ndb also signals
-  that the error is temporary by pushing a warning with the error code
-  ER_GET_TEMPORARY_ERRMSG, if the originating error is temporary.
-*/
-static int has_temporary_error(THD *thd)
-{
-  DBUG_ENTER("has_temporary_error");
-
-  DBUG_EXECUTE_IF("all_errors_are_temporary_errors",
-                  if (thd->stmt_da->is_error())
-                  {
-                    thd->clear_error();
-                    my_error(ER_LOCK_DEADLOCK, MYF(0));
-                  });
-
-  /*
-    If there is no message in THD, we can't say if it's a temporary
-    error or not. This is currently the case for Incident_log_event,
-    which sets no message. Return FALSE.
-  */
-  if (!thd->is_error())
-    DBUG_RETURN(0);
-
-  /*
-    Temporary error codes:
-    currently, InnoDB deadlock detected by InnoDB or lock
-    wait timeout (innodb_lock_wait_timeout exceeded
-  */
-  if (thd->stmt_da->sql_errno() == ER_LOCK_DEADLOCK ||
-      thd->stmt_da->sql_errno() == ER_LOCK_WAIT_TIMEOUT)
-    DBUG_RETURN(1);
-
-#ifdef HAVE_NDB_BINLOG
-  /*
-    currently temporary error set in ndbcluster
-  */
-  List_iterator_fast<MYSQL_ERROR> it(thd->warning_info->warn_list());
-  MYSQL_ERROR *err;
-  while ((err= it++))
-  {
-    DBUG_PRINT("info", ("has condition %d %s", err->get_sql_errno(),
-                        err->get_message_text()));
-    switch (err->get_sql_errno())
-    {
-    case ER_GET_TEMPORARY_ERRMSG:
-      DBUG_RETURN(1);
-    default:
-      break;
-    }
-  }
-#endif
-  DBUG_RETURN(0);
-}
-
 
 /**
   If this is a lagging slave (specified with CHANGE MASTER TO MASTER_DELAY = X), delays accordingly. Also unlocks rli->data_lock.
@@ -2965,7 +2908,7 @@ static int exec_relay_log_event(THD* thd
     if (slave_trans_retries)
     {
       int UNINIT_VAR(temp_err);
-      if (exec_res && (temp_err= has_temporary_error(thd)))
+      if (exec_res && (temp_err= rli->has_temporary_error(thd)))
       {
         const char *errmsg;
         /*
@@ -3011,10 +2954,13 @@ static int exec_relay_log_event(THD* thd
           }
         }
         else
-          sql_print_error("Slave SQL thread retried transaction %lu time(s) "
-                          "in vain, giving up. Consider raising the value of "
-                          "the slave_transaction_retries variable.",
-                          slave_trans_retries);
+        {
+          thd->is_fatal_error= 1;
+          rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(),
+                      "Slave SQL thread retried transaction %lu time(s) "
+                      "in vain, giving up. Consider raising the value of "
+                      "the slave_transaction_retries variable.", rli->trans_retries);
+        }
       }
       else if ((exec_res && !temp_err) ||
                (opt_using_transactions &&


Attachment: [text/bzr-bundle] bzr/andrei.elkin@oracle.com-20110329145601-asqy5hj1a1gzrlra.bundle
Thread
bzr commit into mysql-trunk branch (andrei.elkin:3344) Andrei Elkin29 Mar