From: Andrei Elkin
Date: July 18 2011 6:49pm
Subject: bzr push into mysql-next-mr-wl5569 branch (andrei.elkin:3351 to 3352) WL#5569
List-Archive: http://lists.mysql.com/commits/140346
Message-Id: <201107181849.p6IInkGc001997@mysql1000.dsl.inet.fi>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit

 3352 Andrei Elkin	2011-07-18
      wl#5569 MTS
      
      Cosmetic changes are done to address readability and clearness of source code of the MTS patch.
     @ sql/binlog.cc
        Comments improved.
     @ sql/log_event.cc
        Warning text is improved.
     @ sql/log_event.h
        More comments are added.
     @ sql/rpl_rli.h
        More comments are added.
     @ sql/rpl_slave.cc
        Error constant was changed.
     @ sql/share/errmsg-utf8.txt
        Error constant is changed.

    modified:
      sql/binlog.cc
      sql/log_event.cc
      sql/log_event.h
      sql/rpl_rli.h
      sql/rpl_slave.cc
      sql/share/errmsg-utf8.txt
 3351 Andrei Elkin	2011-07-18
      wl#5569 MTS
      
      Inadvertently introduced hunk two rev:s back is reverted to please rpl_*_mts_crash_safe.

    modified:
      sql/rpl_rli_pdb.cc
=== modified file 'sql/binlog.cc'
--- a/sql/binlog.cc	2011-07-16 17:08:31 +0000
+++ b/sql/binlog.cc	2011-07-18 18:46:45 +0000
@@ -4951,8 +4951,8 @@ THD::binlog_set_pending_rows_event(Rows_
    @param db    db name c-string to be inserted into alphabetically sorted
                 THD::binlog_accessed_db_names list.
                 
-                Note, as the list node data so the node
-                struct itself are allocated in THD::main_mem_root.
+                Note, that space for both the data and the node
+                struct are allocated in THD::main_mem_root.
                 The list lasts for the top-level query time and is reset
                 in @c THD::cleanup_after_query().
 */

=== modified file 'sql/log_event.cc'
--- a/sql/log_event.cc	2011-07-16 17:08:31 +0000
+++ b/sql/log_event.cc	2011-07-18 18:46:45 +0000
@@ -2735,8 +2735,9 @@ int Log_event::apply_event(Relay_log_inf
         {
           // handle synchronization error
           rli->report(WARNING_LEVEL, 0,
-                      "Coordinator thread of multi-threaded slave found "
-                      "a failed to apply an event Worker.");
+                      "Slave worker thread has failed to apply an event. As a "
+                      "consequence, the coordinator thread is stopping "
+                      "execution.");
           DBUG_RETURN(-1);
         }
         /*

=== modified file 'sql/log_event.h'
--- a/sql/log_event.h	2011-07-16 21:51:45 +0000
+++ b/sql/log_event.h	2011-07-18 18:46:45 +0000
@@ -542,9 +542,11 @@ struct sql_ex_info
 /**
    MTS: group of events can be marked to force its execution
    in isolation from any other Workers.
+   So it's a marker for Coordinator to memorize and perform necessary
+   operations in order to guarantee no interference from other Workers.
+   The flag can be set ON only for an event that terminates its group.
    Typically that is done for a transaction that contains 
    a query accessing more than OVER_MAX_DBS_IN_EVENT_MTS databases.
-   The flag can be set ON only for an event that terminates its group.
 */
 #define LOG_EVENT_MTS_ISOLATE_F 0x200
 
@@ -1338,23 +1340,22 @@ private:
 
      Besides the parallelizable case, some events have to be applied by
      Coordinator concurrently with Workers and some to require synchronization
-     with Workers before to apply them.
+     with Workers (@c see wait_for_workers_to_finish) before to apply them.
 
      @retval EVENT_EXEC_PARALLEL  if event is executed by a Worker
      @retval EVENT_EXEC_ASYNC     if event is executed by Coordinator
-     @retval EVENT_EXEC_ASYNC     if event is executed by Coordinator
+     @retval EVENT_EXEC_SYNC      if event is executed by Coordinator
                                   with synchronization against the Workers
   */
   enum enum_mts_event_exec_mode get_mts_execution_mode(ulong slave_server_id,
                                                    bool mts_in_group)
   {
     if ((get_type_code() == FORMAT_DESCRIPTION_EVENT &&
-         ((server_id == (uint32) ::server_id) || (log_pos == 0)))
-        ||
+         ((server_id == (uint32) ::server_id) || (log_pos == 0))) ||
         (get_type_code() == ROTATE_EVENT &&
          ((server_id == (uint32) ::server_id) ||
-          (log_pos == 0    /* very first fake Rotate */
-           && mts_in_group /* ignored events, R_f at slave restart */))))
+          (log_pos == 0    /* very first fake Rotate (R_f) */
+           && mts_in_group /* ignored event turned into R_f at slave stop */))))
       return EVENT_EXEC_ASYNC;
     else if (is_mts_sequential_exec())
       return EVENT_EXEC_SYNC;
@@ -1385,6 +1386,8 @@ private:
     in isolation from any other Workers.
     Typically that is done for a transaction that contains 
     a query accessing more than OVER_MAX_DBS_IN_EVENT_MTS databases.
+    Factually that's a sequential mode where a Worker remains to
+    be the applier.
   */
   virtual void set_mts_isolate_group()
   { 
@@ -1429,7 +1432,7 @@ public:
   virtual bool starts_group() { return FALSE; }
 
   /**
-     @return TRUE  if the event starts a group (transaction)
+     @return TRUE  if the event ends a group (transaction)
              FASE  otherwise
   */
   virtual bool ends_group()   { return FALSE; }

=== modified file 'sql/rpl_rli.h'
--- a/sql/rpl_rli.h	2011-07-16 17:08:31 +0000
+++ b/sql/rpl_rli.h	2011-07-18 18:46:45 +0000
@@ -453,7 +453,12 @@ public:
   /* 
      Ongoing excessive overrun counter to correspond to number of events that
      are being scheduled while a WQ is close to be filled up.
-     The counter describes level of saturation that Workers are experiencing.
+     `Close' is defined as (100 - mts_worker_underrun_level) %.
+     The counter is incremented each time a WQ get filled over that level
+     and decremented when the level drops below.
+     The counter therefore describes level of saturation that Workers 
+     are experiencing and is used as a parameter to compute a nap time for
+     Coordinator in order to avoid reaching WQ limits.
   */
   volatile long mts_wq_excess_cnt;
   long  mts_worker_underrun_level; // % of WQ size at which W is considered hungry

=== modified file 'sql/rpl_slave.cc'
--- a/sql/rpl_slave.cc	2011-07-16 17:08:31 +0000
+++ b/sql/rpl_slave.cc	2011-07-18 18:46:45 +0000
@@ -1179,8 +1179,8 @@ static bool sql_slave_killed(THD* thd, R
         if (is_parallel_warn)
           rli->report(!rli->is_error() ? ERROR_LEVEL :
                       WARNING_LEVEL,    // an error was reported by Worker
-                      ER_MTS_PARALLEL_INCONSISTENT_DATA,
-                      ER(ER_MTS_PARALLEL_INCONSISTENT_DATA),
+                      ER_MTS_INCONSISTENT_DATA,
+                      ER(ER_MTS_INCONSISTENT_DATA),
                       msg_stopped_mts);
         else
           rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,

=== modified file 'sql/share/errmsg-utf8.txt'
--- a/sql/share/errmsg-utf8.txt	2011-07-16 17:08:31 +0000
+++ b/sql/share/errmsg-utf8.txt	2011-07-18 18:46:45 +0000
@@ -6518,5 +6518,5 @@ ER_MTS_UPDATED_DBS_GREATER_MAX
   eng "The number of modified databases exceeds the maximum %d; the database names will not be included in the replication event metadata."
 ER_MTS_CANT_PARALLEL
   eng "Cannot execute the current event group in the parallel mode. Encountered event %s, relay-log name %s, position %s which prevents execution of this event group in parallel mode. Reason: %s."
-ER_MTS_PARALLEL_INCONSISTENT_DATA
+ER_MTS_INCONSISTENT_DATA
   eng "%s"

No bundle (reason: useless for push emails).