Bug#40595: Non-matching rows not released with READ-COMMITTED

on tables with partitions

Problem was that the handler function try_semi_consistent_read
was not propagated to the innodb handler.

Solution was to implement that function in the partitioning
handler.

mysql-test/r/partition_innodb.result:
  Bug#40595: Non-matching rows not released with READ-COMMITTED
  on tables with partitions
  
  Updated test result.
mysql-test/t/partition_innodb.test:
  Bug#40595: Non-matching rows not released with READ-COMMITTED
  on tables with partitions
  
  Added test case for bug#40595.
  
  Note: the replace_regex is taking long time. I have not found
  any better regex (it takes time using 'sed' too).
sql/ha_partition.cc:
  Bug#40595: Non-matching rows not released with READ-COMMITTED
  on tables with partitions
  
  Added function to the partitioning handler for handling
  semi consistent reads (unlock_row was already implemented,
  and this is needed for unlock_row to work properly in innodb).
  It uses pruning for optimizing the call.
sql/ha_partition.h:
  Bug#40595: Non-matching rows not released with READ-COMMITTED
  on tables with partitions
  
  Added function to the partitioning handler for handling
  semi consistent reads (unlock_row was already implemented,
  and this is needed for unlock_row to work properly in innodb).

2 files changed, 39 insertions, 1 deletions

diff --git a/sql/ha_partition.cc b/sql/ha_partition.cc
index 3324bbc2f45..a1131a99fa4 100644
--- a/sql/ha_partition.cc
+++ b/sql/ha_partition.cc
@@ -2813,8 +2813,42 @@ uint ha_partition::lock_count() const
 
 void ha_partition::unlock_row()
 {
+  DBUG_ENTER("ha_partition::unlock_row");
   m_file[m_last_part]->unlock_row();
-  return;
+  DBUG_VOID_RETURN;
+}
+
+
+/**
+  Use semi consistent read if possible
+
+  SYNOPSIS
+    try_semi_consistent_read()
+    yes   Turn on semi consistent read
+
+  RETURN VALUE
+    NONE
+
+  DESCRIPTION
+    See handler.h:
+    Tell the engine whether it should avoid unnecessary lock waits.
+    If yes, in an UPDATE or DELETE, if the row under the cursor was locked
+    by another transaction, the engine may try an optimistic read of
+    the last committed row value under the cursor.
+    Note: prune_partitions are already called before this call, so using
+    pruning is OK.
+*/
+void ha_partition::try_semi_consistent_read(bool yes)
+{
+  handler **file;
+  DBUG_ENTER("ha_partition::try_semi_consistent_read");
+  
+  for (file= m_file; *file; file++)
+  {
+    if (bitmap_is_set(&(m_part_info->used_partitions), (file - m_file)))
+      (*file)->try_semi_consistent_read(yes);
+  }
+  DBUG_VOID_RETURN;
 }
 
 
diff --git a/sql/ha_partition.h b/sql/ha_partition.h
index 78cf47dd1aa..dd06d8d647b 100644
--- a/sql/ha_partition.h
+++ b/sql/ha_partition.h
@@ -325,6 +325,10 @@ public:
     Call to unlock rows not to be updated in transaction
   */
   virtual void unlock_row();
+  /*
+    Call to hint about semi consistent read
+  */
+  virtual void try_semi_consistent_read(bool);
 
   /*
     -------------------------------------------------------------------------