Merge lp:~laurynas-biveinis/percona-server/tokudb-prepare-scans into lp:percona-server/5.6

[Laurynas Biveinis]

Handler extensions to notify SE of incoming index/range scan. Although ideally any required preparation actions should be added to existing methods (ha_index_init, read_range_first), or perhaps by exploiting ICP better, that's not trivial (e.g. cannot add TokuDB prelocking to ha_index_init if it is expensive, because ha_index_init is called to read few rows too). Investigating such changes is on Tokutek's TODO.

Add new handler methods for notifying storage engine of imminent index
scans, both for reading whole table and ranges. Implements
https://blueprints.launchpad.net/percona-server/+spec/prepare-index-range-scan.

The new handler methods added, no-ops in default implementations:
- prepare_index_scan for imminent index (range-less) scan, called in
rr_index_first, rr_index_last in records.cc, and
join_read_first/join_read_last.
- prepare_range_scan for imminent index range scan, called in
QUICK_SELECT_DESC::get_next.
- prepare_index_key_scan_map and prepare_index_key_scan for imminent
index (range-less) scan with used key part information, called in
join_read_always_key in sql_executor.cc.

http://jenkins.percona.com/job/percona-server-5.6-param/562/

[Alexey Kopytov (akopytov)]

review: Approve

=== modified file 'sql/handler.h'
--- sql/handler.h	2014-02-27 12:29:47 +0000
+++ sql/handler.h	2014-03-24 06:21:13 +0000
@@ -2311,6 +2311,44 @@
     return index_read_last(buf, key, key_len);
   }
 public:
+
+  /**
+    Notify storage engine about imminent index scan where a large number of
+    rows is expected to be returned. Does not replace nor call index_init.
+  */
+  virtual int prepare_index_scan(void) { return 0; }
+
+  /**
+    Notify storage engine about imminent index range scan.
+  */
+  virtual int prepare_range_scan(const key_range *start_key,
+                                 const key_range *end_key)
+  {
+    return 0;
+  }
+
+  /**
+    Notify storage engine about imminent index read with a bitmap of used key
+    parts.
+  */
+  int prepare_index_key_scan_map(const uchar *key, key_part_map keypart_map)
+  {
+    uint key_len= calculate_key_len(table, active_index, key, keypart_map);
+    return prepare_index_key_scan(key, key_len);
+  }
+
+protected:
+
+  /**
+    Notify storage engine about imminent index read with key length.
+  */
+  virtual int prepare_index_key_scan(const uchar *key, uint key_len)
+  {
+    return 0;
+  }
+
+public:
+
   virtual int read_range_first(const key_range *start_key,
                                const key_range *end_key,
                                bool eq_range, bool sorted);

=== modified file 'sql/opt_range.cc'
--- sql/opt_range.cc	2014-03-03 17:51:33 +0000
+++ sql/opt_range.cc	2014-03-24 06:21:13 +0000
@@ -10867,6 +10867,18 @@
       }
     }
 
+    key_range prepare_range_start;
+    key_range prepare_range_end;
+
+    last_range->make_min_endpoint(&prepare_range_start);
+    last_range->make_max_endpoint(&prepare_range_end);
+    result = file->prepare_range_scan((last_range->flag & NO_MIN_RANGE)
+                                       ? NULL : &prepare_range_start,
+                                      (last_range->flag & NO_MAX_RANGE)
+                                       ? NULL : &prepare_range_end);
+    if (result)
+      DBUG_RETURN(result);
+
     if (last_range->flag & NO_MAX_RANGE)        // Read last record
     {
       int local_error;

=== modified file 'sql/records.cc'
--- sql/records.cc	2013-08-14 03:57:21 +0000
+++ sql/records.cc	2014-03-24 06:21:13 +0000
@@ -391,8 +391,13 @@
 
 static int rr_index_first(READ_RECORD *info)
 {
-  int tmp= info->table->file->ha_index_first(info->record);
+  int tmp= info->table->file->prepare_index_scan();
   info->read_record= rr_index;
+  if (tmp) {
+    tmp= rr_handle_error(info, tmp);
+    return tmp;
+  }
+  tmp= info->table->file->ha_index_first(info->record);
   if (tmp)
     tmp= rr_handle_error(info, tmp);
   return tmp;
@@ -414,8 +419,13 @@
 
 static int rr_index_last(READ_RECORD *info)
 {
-  int tmp= info->table->file->ha_index_last(info->record);
+  int tmp= info->table->file->prepare_index_scan();
   info->read_record= rr_index_desc;
+  if (tmp) {
+    tmp= rr_handle_error(info, tmp);
+    return tmp;
+  }
+  tmp= info->table->file->ha_index_last(info->record);
   if (tmp)
     tmp= rr_handle_error(info, tmp);
   return tmp;

=== modified file 'sql/sql_executor.cc'
--- sql/sql_executor.cc	2014-02-17 11:12:40 +0000
+++ sql/sql_executor.cc	2014-03-24 06:21:13 +0000
@@ -2203,6 +2203,9 @@
 
   if (cp_buffer_from_ref(tab->join->thd, table, ref))
     return -1;
+  if ((error= table->file->prepare_index_key_scan_map(tab->ref.key_buff,
+                             make_prev_keypart_map(tab->ref.key_parts))))
+    return report_handler_error(table, error);
   if ((error= table->file->ha_index_read_map(table->record[0],
                                              tab->ref.key_buff,
                                              make_prev_keypart_map(tab->ref.key_parts),
@@ -2529,6 +2532,8 @@
     (void) report_handler_error(table, error);
     return 1;
   }
+  if ((error= tab->table->file->prepare_index_scan()))
+    return 1;
   if ((error= tab->table->file->ha_index_first(tab->table->record[0])))
   {
     if (error != HA_ERR_KEY_NOT_FOUND && error != HA_ERR_END_OF_FILE)
@@ -2567,6 +2572,8 @@
     (void) report_handler_error(table, error);
     return 1;
   }
+  if ((error= table->file->prepare_index_scan()))
+    return report_handler_error(table, error);
   if ((error= tab->table->file->ha_index_last(tab->table->record[0])))
     return report_handler_error(table, error);
   return 0;