Bug#14521864: MYSQL 5.1 TO 5.5 BUGS PARTITIONING

Due to an internal change in the server code in between 5.1 and 5.5
(wl#2649) the hash function used in KEY partitioning changed
for numeric and date/time columns (from binary hash calculation
to character based hash calculation).

Also enum/set changed from latin1 ci based hash calculation to
binary hash between 5.1 and 5.5. (bug#11759782).

These changes makes KEY [sub]partitioned tables on any of
the affected column types incompatible with 5.5 and above,
since the calculation of partition id differs.

Also since InnoDB asserts that a deleted row was previously
read (positioned), the server asserts on delete of a row that
is in the wrong partition.

The solution for this situation is:

1) The partitioning engine will check that delete/update will go to the
partition the row was read from and give an error otherwise, consisting
of the rows partitioning fields. This will avoid asserts in InnoDB and
also alert the user that there is a misplaced row. A detailed error
message will be given, including an entry to the error log consisting
of both table name, partition and row content (PK if exists, otherwise
all partitioning columns).


2) A new optional syntax for KEY () partitioning in 5.5 is allowed:
[SUB]PARTITION BY KEY [ALGORITHM = N] (list_of_cols)
Where N = 1 uses the same hashing as 5.1 (Numeric/date/time fields uses
binary hashing, ENUM/SET uses charset hashing) N = 2 uses the same
hashing as 5.5 (Numeric/date/time fields uses charset hashing,
ENUM/SET uses binary hashing). If not set on CREATE/ALTER it will
default to 2.

This new syntax should probably be ignored by NDB.


3) Since there is a demand for avoiding scanning through the full
table, during upgrade the ALTER TABLE t PARTITION BY ... command is
considered a no-op (only .frm change) if everything except ALGORITHM
is the same and ALGORITHM was not set before, which allows manually
upgrading such table by something like:
ALTER TABLE t PARTITION BY KEY ALGORITHM = 1 () or
ALTER TABLE t PARTITION BY KEY ALGORITHM = 2 ()


4) Enhanced partitioning with CHECK/REPAIR to also check for/repair
misplaced rows. (Also works for ALTER TABLE t CHECK/REPAIR PARTITION)

CHECK FOR UPGRADE:
If the .frm version is < 5.5.3
and uses KEY [sub]partitioning
and an affected column type
then it will fail with an message:
KEY () partitioning changed, please run:
ALTER TABLE `test`.`t1`  PARTITION BY KEY ALGORITHM = 1 (a)
PARTITIONS 12
(i.e. current partitioning clause, with the addition of
ALGORITHM = 1)

CHECK without FOR UPGRADE:
if MEDIUM (default) or EXTENDED options are given:
Scan all rows and verify that it is in the correct partition.
Fail for the first misplaced row.

REPAIR:
if default or EXTENDED (i.e. not QUICK/USE_FRM):
Scan all rows and every misplaced row is moved into its correct
partitions.


5) Updated mysqlcheck (called by mysql_upgrade) to handle the
new output from CHECK FOR UPGRADE, to run the ALTER statement
instead of running REPAIR.

This will allow mysql_upgrade (or CHECK TABLE t FOR UPGRADE) to upgrade
a KEY [sub]partitioned table that has any affected field type
and a .frm version < 5.5.3 to ALGORITHM = 1 without rebuild.


Also notice that if the .frm has a version of >= 5.5.3 and ALGORITHM
is not set, it is not possible to know if it consists of rows from
5.1 or 5.5! In these cases I suggest that the user does:
(optional)
LOCK TABLE t WRITE;
SHOW CREATE TABLE t;
(verify that it has no ALGORITHM = N, and to be safe, I would suggest
backing up the .frm file, to be used if one need to change to another
ALGORITHM = N, without needing to rebuild/repair)
ALTER TABLE t <old partitioning clause, but with ALGORITHM = N>;
which should set the ALGORITHM to N (if the table has rows from
5.1 I would suggest N = 1, otherwise N = 2)
CHECK TABLE t;
(here one could use the backed up .frm instead and change to a new N
and run CHECK again and see if it passes)
and if there are misplaced rows:
REPAIR TABLE t;
(optional)
UNLOCK TABLES;

13 files changed, 1012 insertions, 96 deletions

diff --git a/sql/ha_partition.cc b/sql/ha_partition.cc
index 6e1b3fc37b8..4afae052b76 100644
--- a/sql/ha_partition.cc
+++ b/sql/ha_partition.cc
@@ -1,5 +1,5 @@
 /*
-  Copyright (c) 2005, 2012, Oracle and/or its affiliates. All rights reserved.
+  Copyright (c) 2005, 2013, Oracle and/or its affiliates. All rights reserved.
 
   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -62,6 +62,8 @@
 #include "key.h"
 #include "sql_plugin.h"
 #include "table.h"                           /* HA_DATA_PARTITION */
+#include "sql_show.h"                        // append_identifier
+#include "sql_admin.h"                       // SQL_ADMIN_MSG_TEXT_SIZE
 
 #include "debug_sync.h"
 
@@ -271,6 +273,7 @@ void ha_partition::init_handler_variables()
   m_rec_length= 0;
   m_last_part= 0;
   m_rec0= 0;
+  m_err_rec= NULL;
   m_curr_key_info[0]= NULL;
   m_curr_key_info[1]= NULL;
   m_part_func_monotonicity_info= NON_MONOTONIC;
@@ -1035,10 +1038,11 @@ int ha_partition::preload_keys(THD *thd, HA_CHECK_OPT *check_opt)
     0                         Success
 */
 
-static int handle_opt_part(THD *thd, HA_CHECK_OPT *check_opt,
-                           handler *file, uint flag)
+int ha_partition::handle_opt_part(THD *thd, HA_CHECK_OPT *check_opt,
+                                  uint part_id, uint flag)
 {
   int error;
+  handler *file= m_file[part_id];
   DBUG_ENTER("handle_opt_part");
   DBUG_PRINT("enter", ("flag = %u", flag));
 
@@ -1047,9 +1051,27 @@ static int handle_opt_part(THD *thd, HA_CHECK_OPT *check_opt,
   else if (flag == ANALYZE_PARTS)
     error= file->ha_analyze(thd, check_opt);
   else if (flag == CHECK_PARTS)
+  {
     error= file->ha_check(thd, check_opt);
+    if (!error ||
+        error == HA_ADMIN_ALREADY_DONE ||
+        error == HA_ADMIN_NOT_IMPLEMENTED)
+    {
+      if (check_opt->flags & (T_MEDIUM | T_EXTEND))
+        error= check_misplaced_rows(part_id, false);
+    }
+  }
   else if (flag == REPAIR_PARTS)
+  {
     error= file->ha_repair(thd, check_opt);
+    if (!error ||
+        error == HA_ADMIN_ALREADY_DONE ||
+        error == HA_ADMIN_NOT_IMPLEMENTED)
+    {
+      if (check_opt->flags & (T_MEDIUM | T_EXTEND))
+        error= check_misplaced_rows(part_id, true);
+    }
+  }
   else if (flag == ASSIGN_KEYCACHE_PARTS)
     error= file->assign_to_keycache(thd, check_opt);
   else if (flag == PRELOAD_KEYS_PARTS)
@@ -1168,7 +1190,7 @@ int ha_partition::handle_opt_partitions(THD *thd, HA_CHECK_OPT *check_opt,
           part= i * num_subparts + j;
           DBUG_PRINT("info", ("Optimize subpartition %u (%s)",
                      part, sub_elem->partition_name));
-          if ((error= handle_opt_part(thd, check_opt, m_file[part], flag)))
+          if ((error= handle_opt_part(thd, check_opt, part, flag)))
           {
             /* print a line which partition the error belongs to */
             if (error != HA_ADMIN_NOT_IMPLEMENTED &&
@@ -1194,7 +1216,7 @@ int ha_partition::handle_opt_partitions(THD *thd, HA_CHECK_OPT *check_opt,
       {
         DBUG_PRINT("info", ("Optimize partition %u (%s)", i,
                             part_elem->partition_name));
-        if ((error= handle_opt_part(thd, check_opt, m_file[i], flag)))
+        if ((error= handle_opt_part(thd, check_opt, i, flag)))
         {
           /* print a line which partition the error belongs to */
           if (error != HA_ADMIN_NOT_IMPLEMENTED &&
@@ -3368,7 +3390,7 @@ exit:
 
     Called from sql_select.cc, sql_acl.cc, sql_update.cc, and sql_insert.cc.
     new_data is always record[0]
-    old_data is normally record[1] but may be anything
+    old_data is always record[1]
 */
 
 int ha_partition::update_row(const uchar *old_data, uchar *new_data)
@@ -3379,6 +3401,7 @@ int ha_partition::update_row(const uchar *old_data, uchar *new_data)
   longlong func_value;
   timestamp_auto_set_type orig_timestamp_type= table->timestamp_field_type;
   DBUG_ENTER("ha_partition::update_row");
+  m_err_rec= NULL;
 
   /*
     We need to set timestamp field once before we calculate
@@ -3396,6 +3419,25 @@ int ha_partition::update_row(const uchar *old_data, uchar *new_data)
     m_part_info->err_value= func_value;
     goto exit;
   }
+  /*
+    The protocol for updating a row is:
+    1) position the handler (cursor) on the row to be updated,
+       either through the last read row (rnd or index) or by rnd_pos.
+    2) call update_row with both old and new full records as arguments.
+
+    This means that m_last_part should already be set to actual partition
+    where the row was read from. And if that is not the same as the
+    calculated part_id we found a misplaced row, we return an error to
+    notify the user that something is broken in the row distribution
+    between partitions! Since we don't check all rows on read, we return an
+    error instead of correcting m_last_part, to make the user aware of the
+    problem!
+  */
+  if (old_part_id != m_last_part)
+  {
+    m_err_rec= old_data;
+    DBUG_RETURN(HA_ERR_NO_PARTITION_FOUND);
+  }
 
   m_last_part= new_part_id;
   start_part_bulk_insert(thd, new_part_id);
@@ -3501,12 +3543,34 @@ int ha_partition::delete_row(const uchar *buf)
   int error;
   THD *thd= ha_thd();
   DBUG_ENTER("ha_partition::delete_row");
+  m_err_rec= NULL;
 
   if ((error= get_part_for_delete(buf, m_rec0, m_part_info, &part_id)))
   {
     DBUG_RETURN(error);
   }
-  m_last_part= part_id;
+  /*
+    The protocol for deleting a row is:
+    1) position the handler (cursor) on the row to be deleted,
+       either through the last read row (rnd or index) or by rnd_pos.
+    2) call delete_row with the full record as argument.
+
+    This means that m_last_part should already be set to actual partition
+    where the row was read from. And if that is not the same as the
+    calculated part_id we found a misplaced row, we return an error to
+    notify the user that something is broken in the row distribution
+    between partitions! Since we don't check all rows on read, we return an
+    error instead of forwarding the delete to the correct (m_last_part)
+    partition!
+    TODO: change the assert in InnoDB into an error instead and make this one
+    an assert instead and remove the get_part_for_delete()!
+  */
+  if (part_id != m_last_part)
+  {
+    m_err_rec= buf;
+    DBUG_RETURN(HA_ERR_NO_PARTITION_FOUND);
+  }
+
   tmp_disable_binlog(thd);
   error= m_file[part_id]->ha_delete_row(buf);
   reenable_binlog(thd);
@@ -4347,7 +4411,6 @@ int ha_partition::index_init(uint inx, bool sorted)
   file= m_file;
   do
   {
-    /* TODO RONM: Change to index_init() when code is stable */
     if (bitmap_is_set(&(m_part_info->used_partitions), (file - m_file)))
       if ((error= (*file)->ha_index_init(inx, sorted)))
       {
@@ -6827,6 +6890,57 @@ enum row_type ha_partition::get_row_type() const
 }
 
 
+void ha_partition::append_row_to_str(String &str)
+{
+  Field **field_ptr;
+  const uchar *rec;
+  bool is_rec0= !m_err_rec || m_err_rec == table->record[0];
+  if (is_rec0)
+    rec= table->record[0];
+  else
+    rec= m_err_rec;
+  // If PK, use full PK instead of full part field array!
+  if (table->s->primary_key != MAX_KEY)
+  {
+    KEY *key= table->key_info + table->s->primary_key;
+    KEY_PART_INFO *key_part=     key->key_part;
+    KEY_PART_INFO *key_part_end= key_part + key->key_parts;
+    if (!is_rec0)
+      set_key_field_ptr(key, rec, table->record[0]);
+    for (; key_part != key_part_end; key_part++)
+    {
+      Field *field= key_part->field;
+      str.append(" ");
+      str.append(field->field_name);
+      str.append(":");
+      field_unpack(&str, field, rec, 0, false);
+    }
+    if (!is_rec0)
+      set_key_field_ptr(key, table->record[0], rec);
+  }
+  else
+  {
+    if (!is_rec0)
+      set_field_ptr(m_part_info->full_part_field_array, rec,
+                    table->record[0]);
+    /* No primary key, use full partition field array. */
+    for (field_ptr= m_part_info->full_part_field_array;
+         *field_ptr;
+         field_ptr++)
+    {
+      Field *field= *field_ptr;
+      str.append(" ");
+      str.append(field->field_name);
+      str.append(":");
+      field_unpack(&str, field, rec, 0, false);
+    }
+    if (!is_rec0)
+      set_field_ptr(m_part_info->full_part_field_array, table->record[0],
+                    rec);
+  }
+}
+
+
 void ha_partition::print_error(int error, myf errflag)
 {
   THD *thd= ha_thd();
@@ -6835,24 +6949,72 @@ void ha_partition::print_error(int error, myf errflag)
   /* Should probably look for my own errors first */
   DBUG_PRINT("enter", ("error: %d", error));
 
-  if ((error == HA_ERR_NO_PARTITION_FOUND) &&
-      ! (thd->lex->alter_info.flags & ALTER_TRUNCATE_PARTITION))
-    m_part_info->print_no_partition_found(table);
-  else
+  if (error == HA_ERR_NO_PARTITION_FOUND)
   {
-    /* In case m_file has not been initialized, like in bug#42438 */
-    if (m_file)
+    switch(thd_sql_command(thd))
     {
-      if (m_last_part >= m_tot_parts)
+    case SQLCOM_DELETE:
+    case SQLCOM_DELETE_MULTI:
+    case SQLCOM_UPDATE:
+    case SQLCOM_UPDATE_MULTI:
+      if (m_err_rec)
+      {
+        uint max_length;
+        char buf[MAX_KEY_LENGTH];
+        const char *msg= "Found a row in wrong partition (";
+        String str(buf,sizeof(buf),system_charset_info);
+        uint32 part_id;
+        /* Should only happen on DELETE or UPDATE! */
+        str.length(0);
+        str.append_ulonglong(m_last_part);
+        str.append(" != ");
+        if (!get_part_for_delete(m_err_rec, m_rec0, m_part_info, &part_id))
+        {
+          str.append_ulonglong(part_id);
+        }
+        str.append(")");
+        append_row_to_str(str);
+        /* Log this error, so the DBA can notice it and fix it! */
+        sql_print_error("Table '%-192s' corrupted: %s%s\n"
+                        "Please CHECK and REPAIR the table!",
+                        table->s->table_name.str, msg, str.c_ptr_safe());
+
+        max_length= (MYSQL_ERRMSG_SIZE-
+                     (uint) strlen(msg));
+        if (str.length() >= max_length)
+        {
+          str.length(max_length-4);
+          str.append(STRING_WITH_LEN("..."));
+        }
+        my_printf_error(ER_NO_PARTITION_FOR_GIVEN_VALUE, "%s%s", MYF(0),
+                        msg, str.c_ptr_safe());
+        m_err_rec= NULL;
+        DBUG_VOID_RETURN;
+      }
+    default:
       {
-        DBUG_ASSERT(0);
-        m_last_part= 0;
+        if (!(thd->lex->alter_info.flags & ALTER_TRUNCATE_PARTITION))
+        {
+          m_part_info->print_no_partition_found(table);
+          DBUG_VOID_RETURN;
+        }
       }
-      m_file[m_last_part]->print_error(error, errflag);
+      /* fall through to generic error handling. */
     }
-    else
-      handler::print_error(error, errflag);
   }
+
+  /* In case m_file has not been initialized, like in bug#42438 */
+  if (m_file)
+  {
+    if (m_last_part >= m_tot_parts)
+    {
+      DBUG_ASSERT(0);
+      m_last_part= 0;
+    }
+    m_file[m_last_part]->print_error(error, errflag);
+  }
+  else
+    handler::print_error(error, errflag);
   DBUG_VOID_RETURN;
 }
 
@@ -7561,6 +7723,287 @@ int ha_partition::indexes_are_disabled(void)
 }
 
 
+/**
+  Check/fix misplaced rows.
+
+  @param read_part_id  Partition to check/fix.
+  @param repair        If true, move misplaced rows to correct partition.
+
+  @return Operation status.
+    @retval 0     Success
+    @retval != 0  Error
+*/
+
+int ha_partition::check_misplaced_rows(uint read_part_id, bool repair)
+{
+  int result= 0;
+  uint32 correct_part_id;
+  longlong func_value;
+  longlong num_misplaced_rows= 0;
+
+  DBUG_ENTER("ha_partition::check_misplaced_rows");
+
+  DBUG_ASSERT(m_file);
+
+  if (repair)
+  {
+    /* We must read the full row, if we need to move it! */
+    bitmap_set_all(table->read_set);
+    bitmap_set_all(table->write_set);
+  }
+  else
+  {
+    /* Only need to read the partitioning fields. */
+    bitmap_union(table->read_set, &m_part_info->full_part_field_set);
+  }
+
+  if ((result= m_file[read_part_id]->ha_rnd_init(1)))
+    DBUG_RETURN(result);
+
+  while (true)
+  {
+    if ((result= m_file[read_part_id]->rnd_next(m_rec0)))
+    {
+      if (result == HA_ERR_RECORD_DELETED)
+        continue;
+      if (result != HA_ERR_END_OF_FILE)
+        break;
+
+      if (num_misplaced_rows > 0)
+      {
+        print_admin_msg(ha_thd(), "warning", table_share->db.str, table->alias,
+                        opt_op_name[REPAIR_PARTS],
+                        "Moved %lld misplaced rows",
+                        num_misplaced_rows);
+      }
+      /* End-of-file reached, all rows are now OK, reset result and break. */
+      result= 0;
+      break;
+    }
+
+    result= m_part_info->get_partition_id(m_part_info, &correct_part_id,
+                                          &func_value);
+    if (result)
+      break;
+
+    if (correct_part_id != read_part_id)
+    {
+      num_misplaced_rows++;
+      if (!repair)
+      {
+        /* Check. */
+        print_admin_msg(ha_thd(), "error", table_share->db.str, table->alias,
+                        opt_op_name[CHECK_PARTS],
+                        "Found a misplaced row");
+        /* Break on first misplaced row! */
+        result= HA_ADMIN_NEEDS_UPGRADE;
+        break;
+      }
+      else
+      {
+        DBUG_PRINT("info", ("Moving row from partition %d to %d",
+                            read_part_id, correct_part_id));
+
+        /*
+          Insert row into correct partition. Notice that there are no commit
+          for every N row, so the repair will be one large transaction!
+        */
+        if ((result= m_file[correct_part_id]->ha_write_row(m_rec0)))
+        {
+          /*
+            We have failed to insert a row, it might have been a duplicate!
+          */
+          char buf[MAX_KEY_LENGTH];
+          String str(buf,sizeof(buf),system_charset_info);
+          str.length(0);
+          if (result == HA_ERR_FOUND_DUPP_KEY)
+          {
+            str.append("Duplicate key found, "
+                       "please update or delete the record:\n");
+            result= HA_ADMIN_CORRUPT;
+          }
+          m_err_rec= NULL;
+          append_row_to_str(str);
+
+          /*
+            If the engine supports transactions, the failure will be
+            rollbacked.
+          */
+          if (!m_file[correct_part_id]->has_transactions())
+          {
+            /* Log this error, so the DBA can notice it and fix it! */
+            sql_print_error("Table '%-192s' failed to move/insert a row"
+                            " from part %d into part %d:\n%s",
+                            table->s->table_name.str,
+                            read_part_id,
+                            correct_part_id,
+                            str.c_ptr_safe());
+          }
+          print_admin_msg(ha_thd(), "error", table_share->db.str, table->alias,
+                          opt_op_name[REPAIR_PARTS],
+                          "Failed to move/insert a row"
+                          " from part %d into part %d:\n%s",
+                          read_part_id,
+                          correct_part_id,
+                          str.c_ptr_safe());
+          break;
+        }
+
+        /* Delete row from wrong partition. */
+        if ((result= m_file[read_part_id]->ha_delete_row(m_rec0)))
+        {
+          if (m_file[correct_part_id]->has_transactions())
+            break;
+          /*
+            We have introduced a duplicate, since we failed to remove it
+            from the wrong partition.
+          */
+          char buf[MAX_KEY_LENGTH];
+          String str(buf,sizeof(buf),system_charset_info);
+          str.length(0);
+          m_err_rec= NULL;
+          append_row_to_str(str);
+
+          /* Log this error, so the DBA can notice it and fix it! */
+          sql_print_error("Table '%-192s': Delete from part %d failed with"
+                          " error %d. But it was already inserted into"
+                          " part %d, when moving the misplaced row!"
+                          "\nPlease manually fix the duplicate row:\n%s",
+                          table->s->table_name.str,
+                          read_part_id,
+                          result,
+                          correct_part_id,
+                          str.c_ptr_safe());
+          break;
+        }
+      }
+    }
+  }
+
+  int tmp_result= m_file[read_part_id]->ha_rnd_end();
+  DBUG_RETURN(result ? result : tmp_result);
+}
+
+
+#define KEY_PARTITIONING_CHANGED_STR \
+  "KEY () partitioning changed, please run:\nALTER TABLE %s.%s %s"
+
+int ha_partition::check_for_upgrade(HA_CHECK_OPT *check_opt)
+{
+  int error= HA_ADMIN_NEEDS_CHECK;
+  DBUG_ENTER("ha_partition::check_for_upgrade");
+
+  /*
+    This is called even without FOR UPGRADE,
+    if the .frm version is lower than the current version.
+    In that case return that it needs checking!
+  */
+  if (!(check_opt->sql_flags & TT_FOR_UPGRADE))
+    DBUG_RETURN(error);
+
+  /*
+    Partitions will be checked for during their ha_check!
+
+    Check if KEY (sub)partitioning was used and any field's hash calculation
+    differs from 5.1, see bug#14521864.
+  */
+  if (table->s->mysql_version < 50503 &&              // 5.1 table (<5.5.3)
+      ((m_part_info->part_type == HASH_PARTITION &&   // KEY partitioned
+        m_part_info->list_of_part_fields) ||
+       (m_is_sub_partitioned &&                       // KEY subpartitioned
+        m_part_info->list_of_subpart_fields)))
+  {
+    Field **field;
+    if (m_is_sub_partitioned)
+    {
+      field= m_part_info->subpart_field_array;
+    }
+    else
+    {
+      field= m_part_info->part_field_array;
+    }
+    for (; *field; field++)
+    {
+      switch ((*field)->real_type()) {
+      case MYSQL_TYPE_TINY:
+      case MYSQL_TYPE_SHORT:
+      case MYSQL_TYPE_LONG:
+      case MYSQL_TYPE_FLOAT:
+      case MYSQL_TYPE_DOUBLE:
+      case MYSQL_TYPE_NEWDECIMAL:
+      case MYSQL_TYPE_TIMESTAMP:
+      case MYSQL_TYPE_LONGLONG:
+      case MYSQL_TYPE_INT24:
+      case MYSQL_TYPE_TIME:
+      case MYSQL_TYPE_DATETIME:
+      case MYSQL_TYPE_YEAR:
+      case MYSQL_TYPE_NEWDATE:
+      case MYSQL_TYPE_ENUM:
+      case MYSQL_TYPE_SET:
+        {
+          THD *thd= ha_thd();
+          char *part_buf;
+          String db_name, table_name;
+          uint part_buf_len;
+          bool skip_generation= false;
+          partition_info::enum_key_algorithm old_algorithm;
+          old_algorithm= m_part_info->key_algorithm;
+          error= HA_ADMIN_FAILED;
+          append_identifier(ha_thd(), &db_name, table_share->db.str,
+                            table_share->db.length);
+          append_identifier(ha_thd(), &table_name, table_share->table_name.str,
+                            table_share->table_name.length);
+          if (m_part_info->key_algorithm != partition_info::KEY_ALGORITHM_NONE)
+          {
+            /*
+              Only possible when someone tampered with .frm files,
+              like during tests :)
+            */
+            skip_generation= true;
+          }
+          m_part_info->key_algorithm= partition_info::KEY_ALGORITHM_51;
+          if (skip_generation ||
+              !(part_buf= generate_partition_syntax(m_part_info,
+                                                    &part_buf_len,
+                                                    true,
+                                                    true,
+                                                    NULL,
+                                                    NULL)) ||
+          /* Also check that the length is smaller than the output field! */
+              (part_buf_len + db_name.length() + table_name.length()) >=
+                (SQL_ADMIN_MSG_TEXT_SIZE -
+                  (strlen(KEY_PARTITIONING_CHANGED_STR) - 3)))
+          {
+            print_admin_msg(thd, "error", table_share->db.str, table->alias,
+                            opt_op_name[CHECK_PARTS],
+                            KEY_PARTITIONING_CHANGED_STR,
+                            db_name.c_ptr_safe(), table_name.c_ptr_safe(),
+                            "<old partition clause>, but add ALGORITHM = 1"
+                            " between 'KEY' and '(' to change the metadata"
+                            " without the need of a full table rebuild.");
+          }
+          else
+          {
+            print_admin_msg(thd, "error", table_share->db.str, table->alias,
+                            opt_op_name[CHECK_PARTS],
+                            KEY_PARTITIONING_CHANGED_STR,
+                            db_name.c_ptr_safe(), table_name.c_ptr_safe(),
+                            part_buf);
+          }
+          m_part_info->key_algorithm= old_algorithm;
+        }
+        break;
+      default:
+        /* Not affected! */
+        ;
+      }
+    }
+  }
+
+  DBUG_RETURN(error);
+}
+
+
 struct st_mysql_storage_engine partition_storage_engine=
 { MYSQL_HANDLERTON_INTERFACE_VERSION };
 
diff --git a/sql/ha_partition.h b/sql/ha_partition.h
index 24f04ee596b..3895e99d327 100644
--- a/sql/ha_partition.h
+++ b/sql/ha_partition.h
@@ -2,7 +2,7 @@
 #define HA_PARTITION_INCLUDED
 
 /*
-   Copyright (c) 2005, 2012, Oracle and/or its affiliates. All rights reserved.
+   Copyright (c) 2005, 2013, Oracle and/or its affiliates. All rights reserved.
 
   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -25,14 +25,16 @@
 #include "queues.h"             /* QUEUE */
 
 enum partition_keywords
-{ 
+{
   PKW_HASH= 0, PKW_RANGE, PKW_LIST, PKW_KEY, PKW_MAXVALUE, PKW_LINEAR,
-  PKW_COLUMNS
+  PKW_COLUMNS, PKW_ALGORITHM
 };
 
 
 #define PARTITION_BYTES_IN_POS 2
-#define PARTITION_ENABLED_TABLE_FLAGS (HA_FILE_BASED | HA_REC_NOT_IN_SEQ)
+#define PARTITION_ENABLED_TABLE_FLAGS (HA_FILE_BASED | \
+                                       HA_REC_NOT_IN_SEQ | \
+                                       HA_CAN_REPAIR)
 #define PARTITION_DISABLED_TABLE_FLAGS (HA_CAN_GEOMETRY | \
                                         HA_CAN_FULLTEXT | \
                                         HA_DUPLICATE_POS | \
@@ -84,6 +86,7 @@ private:
   */
   KEY *m_curr_key_info[3];              // Current index
   uchar *m_rec0;                        // table->record[0]
+  const uchar *m_err_rec;               // record which gave error
   QUEUE m_queue;                        // Prio queue used by sorted read
   /*
     Since the partition handler is a handler on top of other handlers, it
@@ -1102,9 +1105,18 @@ public:
     virtual bool check_and_repair(THD *thd);
     virtual bool auto_repair() const;
     virtual bool is_crashed() const;
+    virtual int check_for_upgrade(HA_CHECK_OPT *check_opt);
 
     private:
     int handle_opt_partitions(THD *thd, HA_CHECK_OPT *check_opt, uint flags);
+    int handle_opt_part(THD *thd, HA_CHECK_OPT *check_opt, uint part_id,
+                        uint flag);
+    /**
+      Check if the rows are placed in the correct partition.  If the given
+      argument is true, then move the rows to the correct partition.
+    */
+    int check_misplaced_rows(uint read_part_id, bool repair);
+    void append_row_to_str(String &str);
     public:
   /*
     -------------------------------------------------------------------------
diff --git a/sql/handler.cc b/sql/handler.cc
index 6d022630508..dc4fc9ce210 100644
--- a/sql/handler.cc
+++ b/sql/handler.cc
@@ -1,4 +1,4 @@
-/* Copyright (c) 2000, 2012, Oracle and/or its affiliates. All rights reserved.
+/* Copyright (c) 2000, 2013, Oracle and/or its affiliates. All rights reserved.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -5176,6 +5176,8 @@ int handler::ha_write_row(uchar *buf)
   int error;
   Log_func *log_func= Write_rows_log_event::binlog_row_logging_function;
   DBUG_ENTER("handler::ha_write_row");
+  DBUG_EXECUTE_IF("inject_error_ha_write_row",
+                  DBUG_RETURN(HA_ERR_INTERNAL_ERROR); );
 
   MYSQL_INSERT_ROW_START(table_share->db.str, table_share->table_name.str);
   mark_trx_read_write();
@@ -5203,6 +5205,7 @@ int handler::ha_update_row(const uchar *old_data, uchar *new_data)
     (and the old record is in record[1]).
    */
   DBUG_ASSERT(new_data == table->record[0]);
+  DBUG_ASSERT(old_data == table->record[1]);
 
   MYSQL_UPDATE_ROW_START(table_share->db.str, table_share->table_name.str);
   mark_trx_read_write();
@@ -5220,6 +5223,13 @@ int handler::ha_delete_row(const uchar *buf)
 {
   int error;
   Log_func *log_func= Delete_rows_log_event::binlog_row_logging_function;
+  /*
+    Normally table->record[0] is used, but sometimes table->record[1] is used.
+  */
+  DBUG_ASSERT(buf == table->record[0] ||
+              buf == table->record[1]);
+  DBUG_EXECUTE_IF("inject_error_ha_delete_row",
+                  return HA_ERR_INTERNAL_ERROR; );
 
   MYSQL_DELETE_ROW_START(table_share->db.str, table_share->table_name.str);
   mark_trx_read_write();
diff --git a/sql/key.cc b/sql/key.cc
index 9d0f86f2565..f13377146c8 100644
--- a/sql/key.cc
+++ b/sql/key.cc
@@ -1,4 +1,4 @@
-/* Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved.
+/* Copyright (c) 2000, 2013, Oracle and/or its affiliates. All rights reserved.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -312,6 +312,70 @@ bool key_cmp_if_same(TABLE *table,const uchar *key,uint idx,uint key_length)
   return 0;
 }
 
+
+/**
+  Unpack a field and append it.
+
+  @param[inout] to           String to append the field contents to.
+  @param        field        Field to unpack.
+  @param        rec          Record which contains the field data.
+  @param        max_length   Maximum length of field to unpack
+                             or 0 for unlimited.
+  @param        prefix_key   The field is used as a prefix key.
+*/
+
+void field_unpack(String *to, Field *field, const uchar *rec, uint max_length,
+                  bool prefix_key)
+{
+  String tmp;
+  DBUG_ENTER("field_unpack");
+  if (!max_length)
+    max_length= field->pack_length();
+  if (field)
+  {
+    if (field->is_null())
+    {
+      to->append(STRING_WITH_LEN("NULL"));
+      DBUG_VOID_RETURN;
+    }
+    CHARSET_INFO *cs= field->charset();
+    field->val_str(&tmp);
+    /*
+      For BINARY(N) strip trailing zeroes to make
+      the error message nice-looking
+    */
+    if (field->binary() &&  field->type() == MYSQL_TYPE_STRING && tmp.length())
+    {
+      const char *tmp_end= tmp.ptr() + tmp.length();
+      while (tmp_end > tmp.ptr() && !*--tmp_end) ;
+      tmp.length(tmp_end - tmp.ptr() + 1);
+    }
+    if (cs->mbmaxlen > 1 && prefix_key)
+    {
+      /*
+        Prefix key, multi-byte charset.
+        For the columns of type CHAR(N), the above val_str()
+        call will return exactly "key_part->length" bytes,
+        which can break a multi-byte characters in the middle.
+        Align, returning not more than "char_length" characters.
+      */
+      uint charpos, char_length= max_length / cs->mbmaxlen;
+      if ((charpos= my_charpos(cs, tmp.ptr(),
+                               tmp.ptr() + tmp.length(),
+                               char_length)) < tmp.length())
+        tmp.length(charpos);
+    }
+    if (max_length < field->pack_length())
+      tmp.length(min(tmp.length(),max_length));
+    ErrConvString err(&tmp);
+    to->append(err.ptr());
+  }
+  else
+    to->append(STRING_WITH_LEN("???"));
+  DBUG_VOID_RETURN;
+}
+
+
 /*
   unpack key-fields from record to some buffer.
 
@@ -329,8 +393,6 @@ bool key_cmp_if_same(TABLE *table,const uchar *key,uint idx,uint key_length)
 void key_unpack(String *to,TABLE *table,uint idx)
 {
   KEY_PART_INFO *key_part,*key_part_end;
-  Field *field;
-  String tmp;
   my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
   DBUG_ENTER("key_unpack");
 
@@ -346,47 +408,13 @@ void key_unpack(String *to,TABLE *table,uint idx)
     {
       if (table->record[0][key_part->null_offset] & key_part->null_bit)
       {
-	to->append(STRING_WITH_LEN("NULL"));
-	continue;
+        to->append(STRING_WITH_LEN("NULL"));
+        continue;
       }
     }
-    if ((field=key_part->field))
-    {
-      CHARSET_INFO *cs= field->charset();
-      field->val_str(&tmp);
-      /*
-        For BINARY(N) strip trailing zeroes to make
-        the error message nice-looking
-      */
-      if (field->binary() &&  field->type() == MYSQL_TYPE_STRING && tmp.length())
-      {
-        const char *tmp_end= tmp.ptr() + tmp.length();
-        while (tmp_end > tmp.ptr() && !*--tmp_end) ;
-        tmp.length(tmp_end - tmp.ptr() + 1);
-      }
-      if (cs->mbmaxlen > 1 && (key_part->key_part_flag & HA_PART_KEY_SEG))
-      {
-        /* 
-          Prefix key, multi-byte charset. 
-          For the columns of type CHAR(N), the above val_str() 
-          call will return exactly "key_part->length" bytes, 
-          which can break a multi-byte characters in the middle. 
-          Align, returning not more than "char_length" characters. 
-        */
-        uint charpos, char_length= key_part->length / cs->mbmaxlen;
-        if ((charpos= my_charpos(cs, tmp.ptr(),
-                                 tmp.ptr() + tmp.length(),
-                                 char_length)) < tmp.length())
-          tmp.length(charpos);
-      }
-      if (key_part->length < field->pack_length())
-	tmp.length(min(tmp.length(),key_part->length));
-      ErrConvString err(&tmp);
-      to->append(err.ptr());
-    }
-    else
-      to->append(STRING_WITH_LEN("???"));
-  }
+    field_unpack(to, key_part->field, table->record[0], key_part->length,
+                 test(key_part->key_part_flag & HA_PART_KEY_SEG));
+ }
   dbug_tmp_restore_column_map(table->read_set, old_map);
   DBUG_VOID_RETURN;
 }
diff --git a/sql/key.h b/sql/key.h
index 0c8a4b2db27..90296427dac 100644
--- a/sql/key.h
+++ b/sql/key.h
@@ -1,4 +1,4 @@
-/* Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
+/* Copyright (c) 2006, 2013, Oracle and/or its affiliates. All rights reserved.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -32,6 +32,8 @@ void key_restore(uchar *to_record, uchar *from_key, KEY *key_info,
                  uint key_length);
 bool key_cmp_if_same(TABLE *form,const uchar *key,uint index,uint key_length);
 void key_unpack(String *to,TABLE *form,uint index);
+void field_unpack(String *to, Field *field, const uchar *rec, uint max_length,
+                  bool prefix_key);
 bool is_key_used(TABLE *table, uint idx, const MY_BITMAP *fields);
 int key_cmp(KEY_PART_INFO *key_part, const uchar *key, uint key_length);
 extern "C" int key_rec_cmp(void *key_info, uchar *a, uchar *b);
diff --git a/sql/partition_element.h b/sql/partition_element.h
index d6ee44078f5..f4eb282073b 100644
--- a/sql/partition_element.h
+++ b/sql/partition_element.h
@@ -1,7 +1,7 @@
 #ifndef PARTITION_ELEMENT_INCLUDED
 #define PARTITION_ELEMENT_INCLUDED
 
-/* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+/* Copyright (c) 2005, 2013, Oracle and/or its affiliates. All rights reserved.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -106,9 +106,8 @@ public:
   enum partition_state part_state;
   uint16 nodegroup_id;
   bool has_null_value;
-  /* signed_flag and max_value only relevant for subpartitions */
-  bool signed_flag;
-  bool max_value;
+  bool signed_flag;                          // Range value signed
+  bool max_value;                            // MAXVALUE range
 
   partition_element()
   : part_max_rows(0), part_min_rows(0), range_value(0),
diff --git a/sql/partition_info.cc b/sql/partition_info.cc
index cbf22d18c87..5a7d0bf0c43 100644
--- a/sql/partition_info.cc
+++ b/sql/partition_info.cc
@@ -1,4 +1,4 @@
-/* Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
+/* Copyright (c) 2006, 2013, Oracle and/or its affiliates. All rights reserved.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -2180,9 +2180,36 @@ int partition_info::fix_parser_data(THD *thd)
   if (!(part_type == RANGE_PARTITION ||
         part_type == LIST_PARTITION))
   {
-    /* Nothing to do for HASH/KEY partitioning */
+    if (part_type == HASH_PARTITION && list_of_part_fields)
+    {
+      /* KEY partitioning, check ALGORITHM = N. Should not pass the parser! */
+      if (key_algorithm > KEY_ALGORITHM_55)
+      {
+        my_error(ER_PARTITION_FUNCTION_IS_NOT_ALLOWED, MYF(0));
+        DBUG_RETURN(true);
+      }
+      /* If not set, use DEFAULT = 2 for CREATE and ALTER! */
+      if ((thd_sql_command(thd) == SQLCOM_CREATE_TABLE ||
+           thd_sql_command(thd) == SQLCOM_ALTER_TABLE) &&
+          key_algorithm == KEY_ALGORITHM_NONE)
+        key_algorithm= KEY_ALGORITHM_55;
+    }
     DBUG_RETURN(FALSE);
   }
+  if (is_sub_partitioned() && list_of_subpart_fields)
+  {
+    /* KEY subpartitioning, check ALGORITHM = N. Should not pass the parser! */
+    if (key_algorithm > KEY_ALGORITHM_55)
+    {
+      my_error(ER_PARTITION_FUNCTION_IS_NOT_ALLOWED, MYF(0));
+      DBUG_RETURN(true);
+    }
+    /* If not set, use DEFAULT = 2 for CREATE and ALTER! */
+    if ((thd_sql_command(thd) == SQLCOM_CREATE_TABLE ||
+         thd_sql_command(thd) == SQLCOM_ALTER_TABLE) &&
+        key_algorithm == KEY_ALGORITHM_NONE)
+      key_algorithm= KEY_ALGORITHM_55;
+  }
   do
   {
     part_elem= it++;
@@ -2231,6 +2258,255 @@ int partition_info::fix_parser_data(THD *thd)
   DBUG_RETURN(FALSE);
 }
 
+
+/**
+  helper function to compare strings that can also be
+  a NULL pointer.
+
+  @param a  char pointer (can be NULL).
+  @param b  char pointer (can be NULL).
+
+  @return false if equal
+    @retval true  strings differs
+    @retval false strings is equal
+*/
+
+static bool strcmp_null(const char *a, const char *b)
+{
+  if (!a && !b)
+    return false;
+  if (a && b && !strcmp(a, b))
+    return false;
+  return true;
+}
+
+
+/**
+  Check if the new part_info has the same partitioning.
+
+  @param new_part_info  New partition definition to compare with.
+
+  @return True if not considered to have changed the partitioning.
+    @retval true  Allowed change (only .frm change, compatible distribution).
+    @retval false Different partitioning, will need redistribution of rows.
+
+  @note Currently only used to allow changing from non-set key_algorithm
+  to a specified key_algorithm, to avoid rebuild when upgrading from 5.1 of
+  such partitioned tables using numeric colums in the partitioning expression.
+  For more info see bug#14521864.
+  Does not check if columns etc has changed, i.e. only for
+  alter_info->flags == ALTER_PARTITION.
+*/
+
+bool partition_info::has_same_partitioning(partition_info *new_part_info)
+{
+  DBUG_ENTER("partition_info::has_same_partitioning");
+
+  if (!new_part_info ||
+      part_type != new_part_info->part_type ||
+      num_parts != new_part_info->num_parts ||
+      use_default_partitions != new_part_info->use_default_partitions ||
+      new_part_info->is_sub_partitioned() != is_sub_partitioned())
+    DBUG_RETURN(false);
+
+  if (part_type != HASH_PARTITION)
+  {
+    /*
+      RANGE or LIST partitioning, check if KEY subpartitioned.
+      Also COLUMNS partitioning was added in 5.5, so treat that as different.
+    */
+    if (!is_sub_partitioned() ||
+        !new_part_info->is_sub_partitioned() ||
+        column_list ||
+        new_part_info->column_list ||
+        !list_of_subpart_fields ||
+        !new_part_info->list_of_subpart_fields ||
+        new_part_info->num_subparts != num_subparts ||
+        new_part_info->subpart_field_list.elements !=
+          subpart_field_list.elements ||
+        new_part_info->use_default_subpartitions !=
+          use_default_subpartitions)
+      DBUG_RETURN(false);
+  }
+  else
+  {
+    /* Check if KEY partitioned. */
+    if (!new_part_info->list_of_part_fields ||
+        !list_of_part_fields ||
+        new_part_info->part_field_list.elements != part_field_list.elements)
+      DBUG_RETURN(false);
+  }
+
+  /* Check that it will use the same fields in KEY (fields) list. */
+  List_iterator<char> old_field_name_it(part_field_list);
+  List_iterator<char> new_field_name_it(new_part_info->part_field_list);
+  char *old_name, *new_name;
+  while ((old_name= old_field_name_it++))
+  {
+    new_name= new_field_name_it++;
+    if (!new_name || my_strcasecmp(system_charset_info,
+                                   new_name,
+                                   old_name))
+      DBUG_RETURN(false);
+  }
+
+  if (is_sub_partitioned())
+  {
+    /* Check that it will use the same fields in KEY subpart fields list. */
+    List_iterator<char> old_field_name_it(subpart_field_list);
+    List_iterator<char> new_field_name_it(new_part_info->subpart_field_list);
+    char *old_name, *new_name;
+    while ((old_name= old_field_name_it++))
+    {
+      new_name= new_field_name_it++;
+      if (!new_name || my_strcasecmp(system_charset_info,
+                                     new_name,
+                                     old_name))
+        DBUG_RETURN(false);
+    }
+  }
+
+  if (!use_default_partitions)
+  {
+    /*
+      Loop over partitions/subpartition to verify that they are
+      the same, including state and name.
+    */
+    List_iterator<partition_element> part_it(partitions);
+    List_iterator<partition_element> new_part_it(new_part_info->partitions);
+    uint i= 0;
+    do
+    {
+      partition_element *part_elem= part_it++;
+      partition_element *new_part_elem= new_part_it++;
+      /*
+        The following must match:
+        partition_name, tablespace_name, data_file_name, index_file_name,
+        engine_type, part_max_rows, part_min_rows, nodegroup_id.
+        (max_value, signed_flag, has_null_value only on partition level,
+        RANGE/LIST)
+        The following can differ:
+          - part_comment
+        part_state must be PART_NORMAL!
+      */
+      if (!part_elem || !new_part_elem ||
+          strcmp(part_elem->partition_name,
+                 new_part_elem->partition_name) ||
+          part_elem->part_state != PART_NORMAL ||
+          new_part_elem->part_state != PART_NORMAL ||
+          part_elem->max_value != new_part_elem->max_value ||
+          part_elem->signed_flag != new_part_elem->signed_flag ||
+          part_elem->has_null_value != new_part_elem->has_null_value)
+        DBUG_RETURN(false);
+
+      /* new_part_elem may not have engine_type set! */
+      if (new_part_elem->engine_type &&
+          part_elem->engine_type != new_part_elem->engine_type)
+        DBUG_RETURN(false);
+
+      if (is_sub_partitioned())
+      {
+        /*
+          Check that both old and new partition has the same definition
+          (VALUES IN/VALUES LESS THAN) (No COLUMNS partitioning, see above)
+        */
+        if (part_type == LIST_PARTITION)
+        {
+          List_iterator<part_elem_value> list_vals(part_elem->list_val_list);
+          List_iterator<part_elem_value>
+            new_list_vals(new_part_elem->list_val_list);
+          part_elem_value *val;
+          part_elem_value *new_val;
+          while ((val= list_vals++))
+          {
+            new_val= new_list_vals++;
+            if (!new_val)
+              DBUG_RETURN(false);
+            if ((!val->null_value && !new_val->null_value) &&
+                val->value != new_val->value)
+              DBUG_RETURN(false);
+          }
+          if (new_list_vals++)
+            DBUG_RETURN(false);
+        }
+        else
+        {
+          DBUG_ASSERT(part_type == RANGE_PARTITION);
+          if (new_part_elem->range_value != part_elem->range_value)
+            DBUG_RETURN(false);
+        }
+
+        if (!use_default_subpartitions)
+        {
+          List_iterator<partition_element>
+            sub_part_it(part_elem->subpartitions);
+          List_iterator<partition_element>
+            new_sub_part_it(new_part_elem->subpartitions);
+          uint j= 0;
+          do
+          {
+            partition_element *sub_part_elem= sub_part_it++;
+            partition_element *new_sub_part_elem= new_sub_part_it++;
+            /* new_part_elem may not have engine_type set! */
+            if (new_sub_part_elem->engine_type &&
+                sub_part_elem->engine_type != new_part_elem->engine_type)
+              DBUG_RETURN(false);
+
+            if (strcmp(sub_part_elem->partition_name,
+                       new_sub_part_elem->partition_name) ||
+                sub_part_elem->part_state != PART_NORMAL ||
+                new_sub_part_elem->part_state != PART_NORMAL ||
+                sub_part_elem->part_min_rows !=
+                  new_sub_part_elem->part_min_rows ||
+                sub_part_elem->part_max_rows !=
+                  new_sub_part_elem->part_max_rows ||
+                sub_part_elem->nodegroup_id !=
+                  new_sub_part_elem->nodegroup_id)
+              DBUG_RETURN(false);
+  
+            if (strcmp_null(sub_part_elem->data_file_name,
+                            new_sub_part_elem->data_file_name) ||
+                strcmp_null(sub_part_elem->index_file_name,
+                            new_sub_part_elem->index_file_name) ||
+                strcmp_null(sub_part_elem->tablespace_name,
+                            new_sub_part_elem->tablespace_name))
+              DBUG_RETURN(false);
+
+          } while (++j < num_subparts);
+        }
+      }
+      else
+      {
+        if (part_elem->part_min_rows != new_part_elem->part_min_rows ||
+            part_elem->part_max_rows != new_part_elem->part_max_rows ||
+            part_elem->nodegroup_id != new_part_elem->nodegroup_id)
+          DBUG_RETURN(false);
+
+        if (strcmp_null(part_elem->data_file_name,
+                        new_part_elem->data_file_name) ||
+            strcmp_null(part_elem->index_file_name,
+                        new_part_elem->index_file_name) ||
+            strcmp_null(part_elem->tablespace_name,
+                        new_part_elem->tablespace_name))
+          DBUG_RETURN(false);
+      }
+    } while (++i < num_parts);
+  }
+
+  /*
+    Only if key_algorithm was not specified before and it is now set,
+    consider this as nothing was changed!
+    But if already set, consider it as a change, and force rebuild!
+  */
+  DBUG_ASSERT(new_part_info->key_algorithm !=
+                partition_info::KEY_ALGORITHM_NONE);
+  if (key_algorithm != partition_info::KEY_ALGORITHM_NONE)
+    DBUG_RETURN(false);
+
+  DBUG_RETURN(true);
+}
+
+
 void partition_info::print_debug(const char *str, uint *value)
 {
   DBUG_ENTER("print_debug");
diff --git a/sql/partition_info.h b/sql/partition_info.h
index e59d4ec8ba4..d3706c8abf4 100644
--- a/sql/partition_info.h
+++ b/sql/partition_info.h
@@ -1,7 +1,7 @@
 #ifndef PARTITION_INFO_INCLUDED
 #define PARTITION_INFO_INCLUDED
 
-/* Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
+/* Copyright (c) 2006, 2013, Oracle and/or its affiliates. All rights reserved.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -192,6 +192,19 @@ public:
     but mainly of use to handlers supporting partitioning.
   */
   uint16 linear_hash_mask;
+  /*
+    PARTITION BY KEY ALGORITHM=N
+    Which algorithm to use for hashing the fields.
+    N = 1 - Use 5.1 hashing (numeric fields are hashed as binary)
+    N = 2 - Use 5.5 hashing (numeric fields are hashed like latin1 bytes)
+  */
+  enum enum_key_algorithm
+    {
+      KEY_ALGORITHM_NONE= 0,
+      KEY_ALGORITHM_51= 1,
+      KEY_ALGORITHM_55= 2
+    };
+  enum_key_algorithm key_algorithm;
 
   bool use_default_partitions;
   bool use_default_num_partitions;
@@ -232,6 +245,7 @@ public:
     count_curr_subparts(0), part_error_code(0),
     num_list_values(0), num_part_fields(0), num_subpart_fields(0),
     num_full_part_fields(0), has_null_part_id(0), linear_hash_mask(0),
+    key_algorithm(KEY_ALGORITHM_NONE),
     use_default_partitions(TRUE), use_default_num_partitions(TRUE),
     use_default_subpartitions(TRUE), use_default_num_subpartitions(TRUE),
     default_partitions_setup(FALSE), defined_max_value(FALSE),
@@ -298,6 +312,7 @@ public:
   bool add_column_list_value(THD *thd, Item *item);
   void set_show_version_string(String *packet);
   void report_part_expr_error(bool use_subpart_expr);
+  bool has_same_partitioning(partition_info *new_part_info);
 private:
   static int list_part_cmp(const void* a, const void* b);
   bool set_up_default_partitions(handler *file, HA_CREATE_INFO *info,
diff --git a/sql/sql_admin.cc b/sql/sql_admin.cc
index c66d7364f6a..16fcf120128 100644
--- a/sql/sql_admin.cc
+++ b/sql/sql_admin.cc
@@ -1,4 +1,4 @@
-/* Copyright (c) 2010, 2011, Oracle and/or its affiliates. All rights reserved.
+/* Copyright (c) 2010, 2013, Oracle and/or its affiliates. All rights reserved.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -286,7 +286,8 @@ static bool mysql_admin_table(THD* thd, TABLE_LIST* tables,
   item->maybe_null = 1;
   field_list.push_back(item = new Item_empty_string("Msg_type", 10));
   item->maybe_null = 1;
-  field_list.push_back(item = new Item_empty_string("Msg_text", 255));
+  field_list.push_back(item = new Item_empty_string("Msg_text",
+                                                    SQL_ADMIN_MSG_TEXT_SIZE));
   item->maybe_null = 1;
   if (protocol->send_result_set_metadata(&field_list,
                             Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
diff --git a/sql/sql_admin.h b/sql/sql_admin.h
index f7ec76efd5e..5398e3019f1 100644
--- a/sql/sql_admin.h
+++ b/sql/sql_admin.h
@@ -1,4 +1,4 @@
-/* Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
+/* Copyright (c) 2010, 2013, Oracle and/or its affiliates. All rights reserved.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -16,6 +16,8 @@
 #ifndef SQL_TABLE_MAINTENANCE_H
 #define SQL_TABLE_MAINTENANCE_H
 
+/* Must be able to hold ALTER TABLE t PARTITION BY ... KEY ALGORITHM = 1 ... */
+#define SQL_ADMIN_MSG_TEXT_SIZE 128 * 1024
 
 bool mysql_assign_to_keycache(THD* thd, TABLE_LIST* table_list,
                               LEX_STRING *key_cache_name);
diff --git a/sql/sql_partition.cc b/sql/sql_partition.cc
index 8154f4cc404..a64520a298d 100644
--- a/sql/sql_partition.cc
+++ b/sql/sql_partition.cc
@@ -1,4 +1,4 @@
-/* Copyright (c) 2005, 2012, Oracle and/or its affiliates. All rights reserved.
+/* Copyright (c) 2005, 2013, Oracle and/or its affiliates. All rights reserved.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -87,7 +87,9 @@ const LEX_STRING partition_keywords[]=
   { C_STRING_WITH_LEN("KEY") },
   { C_STRING_WITH_LEN("MAXVALUE") },
   { C_STRING_WITH_LEN("LINEAR ") },
-  { C_STRING_WITH_LEN(" COLUMNS") }
+  { C_STRING_WITH_LEN(" COLUMNS") },
+  { C_STRING_WITH_LEN("ALGORITHM") }
+
 };
 static const char *part_str= "PARTITION";
 static const char *sub_str= "SUB";
@@ -368,7 +370,7 @@ int get_parts_for_update(const uchar *old_data, uchar *new_data,
   longlong old_func_value;
   DBUG_ENTER("get_parts_for_update");
 
-  DBUG_ASSERT(new_data == rec0);
+  DBUG_ASSERT(new_data == rec0);             // table->record[0]
   set_field_ptr(part_field_array, old_data, rec0);
   error= part_info->get_partition_id(part_info, old_part_id,
                                      &old_func_value);
@@ -526,12 +528,12 @@ static bool set_up_field_array(TABLE *table,
   }
   if (num_fields > MAX_REF_PARTS)
   {
-    char *ptr;
+    char *err_str;
     if (is_sub_part)
-      ptr= (char*)"subpartition function";
+      err_str= (char*)"subpartition function";
     else
-      ptr= (char*)"partition function";
-    my_error(ER_TOO_MANY_PARTITION_FUNC_FIELDS_ERROR, MYF(0), ptr);
+      err_str= (char*)"partition function";
+    my_error(ER_TOO_MANY_PARTITION_FUNC_FIELDS_ERROR, MYF(0), err_str);
     DBUG_RETURN(TRUE);
   }
   if (num_fields == 0)
@@ -2489,6 +2491,17 @@ char *generate_partition_syntax(partition_info *part_info,
       if (part_info->list_of_part_fields)
       {
         err+= add_part_key_word(fptr, partition_keywords[PKW_KEY].str);
+        if (part_info->key_algorithm != partition_info::KEY_ALGORITHM_NONE)
+        {
+          /*
+            Can't add a !50530 comment, since we are already within a comment!
+          */
+          err+= add_part_key_word(fptr, partition_keywords[PKW_ALGORITHM].str);
+          err+= add_equal(fptr);
+          err+= add_space(fptr);
+          err+= add_int(fptr, part_info->key_algorithm);
+          err+= add_space(fptr);
+        }
         err+= add_part_field_list(fptr, part_info->part_field_list);
       }
       else
@@ -2529,6 +2542,17 @@ char *generate_partition_syntax(partition_info *part_info,
     if (part_info->list_of_subpart_fields)
     {
       add_part_key_word(fptr, partition_keywords[PKW_KEY].str);
+      if (part_info->key_algorithm != partition_info::KEY_ALGORITHM_NONE)
+      {
+        /*
+          Can't add a !50530 comment, since we are already within a comment!
+        */
+        err+= add_part_key_word(fptr, partition_keywords[PKW_ALGORITHM].str);
+        err+= add_equal(fptr);
+        err+= add_space(fptr);
+        err+= add_int(fptr, part_info->key_algorithm);
+        err+= add_space(fptr);
+      }
       add_part_field_list(fptr, part_info->subpart_field_list);
     }
     else
@@ -2738,10 +2762,82 @@ static uint32 calculate_key_value(Field **field_array)
 {
   ulong nr1= 1;
   ulong nr2= 4;
+  bool use_51_hash;
+  use_51_hash= test((*field_array)->table->part_info->key_algorithm ==
+                    partition_info::KEY_ALGORITHM_51);
 
   do
   {
     Field *field= *field_array;
+    if (use_51_hash)
+    {
+      switch (field->real_type()) {
+      case MYSQL_TYPE_TINY:
+      case MYSQL_TYPE_SHORT:
+      case MYSQL_TYPE_LONG:
+      case MYSQL_TYPE_FLOAT:
+      case MYSQL_TYPE_DOUBLE:
+      case MYSQL_TYPE_NEWDECIMAL:
+      case MYSQL_TYPE_TIMESTAMP:
+      case MYSQL_TYPE_LONGLONG:
+      case MYSQL_TYPE_INT24:
+      case MYSQL_TYPE_TIME:
+      case MYSQL_TYPE_DATETIME:
+      case MYSQL_TYPE_YEAR:
+      case MYSQL_TYPE_NEWDATE:
+        {
+          if (field->is_null())
+          {
+            nr1^= (nr1 << 1) | 1;
+            continue;
+          }
+          /* Force this to my_hash_sort_bin, which was used in 5.1! */
+          uint len= field->pack_length();
+          my_charset_bin.coll->hash_sort(&my_charset_bin, field->ptr, len,
+                                         &nr1, &nr2);
+          /* Done with this field, continue with next one. */
+          continue;
+        }
+      case MYSQL_TYPE_STRING:
+      case MYSQL_TYPE_VARCHAR:
+      case MYSQL_TYPE_BIT:
+        /* Not affected, same in 5.1 and 5.5 */
+        break;
+      /*
+        ENUM/SET uses my_hash_sort_simple in 5.1 (i.e. my_charset_latin1)
+        and my_hash_sort_bin in 5.5!
+      */
+      case MYSQL_TYPE_ENUM:
+      case MYSQL_TYPE_SET:
+        {
+          if (field->is_null())
+          {
+            nr1^= (nr1 << 1) | 1;
+            continue;
+          }
+          /* Force this to my_hash_sort_bin, which was used in 5.1! */
+          uint len= field->pack_length();
+          my_charset_latin1.coll->hash_sort(&my_charset_latin1, field->ptr,
+                                            len, &nr1, &nr2);
+          continue;
+        }
+      /* These types should not be allowed for partitioning! */
+      case MYSQL_TYPE_NULL:
+      case MYSQL_TYPE_DECIMAL:
+      case MYSQL_TYPE_DATE:
+      case MYSQL_TYPE_TINY_BLOB:
+      case MYSQL_TYPE_MEDIUM_BLOB:
+      case MYSQL_TYPE_LONG_BLOB:
+      case MYSQL_TYPE_BLOB:
+      case MYSQL_TYPE_VAR_STRING:
+      case MYSQL_TYPE_GEOMETRY:
+        /* fall through. */
+      default:
+        DBUG_ASSERT(0);                    // New type?
+        /* Fall through for default hashing (5.5). */
+      }
+      /* fall through, use collation based hashing. */
+    }
     field->hash(&nr1, &nr2);
   } while (*(++field_array));
   return (uint32) nr1;
@@ -5484,14 +5580,25 @@ the generated partition syntax in a correct manner.
         Need to cater for engine types that can handle partition without
         using the partition handler.
       */
-      if (thd->work_part_info != table->part_info)
+      if (part_info != table->part_info)
       {
-        DBUG_PRINT("info", ("partition changed"));
-        *partition_changed= TRUE;
-        if (thd->work_part_info->fix_parser_data(thd))
+        if (part_info->fix_parser_data(thd))
         {
           goto err;
         }
+        /*
+          Compare the old and new part_info. If only key_algorithm
+          change is done, don't consider it as changed partitioning (to avoid
+          rebuild). This is to handle KEY (numeric_cols) partitioned tables
+          created in 5.1. For more info, see bug#14521864.
+        */
+        if (alter_info->flags != ALTER_PARTITION ||
+            !table->part_info ||
+            !table->part_info->has_same_partitioning(part_info))
+        {
+          DBUG_PRINT("info", ("partition changed"));
+          *partition_changed= true;
+        }
       }
       /*
         Set up partition default_engine_type either from the create_info
diff --git a/sql/sql_show.cc b/sql/sql_show.cc
index b921b2c782e..326f09e7955 100644
--- a/sql/sql_show.cc
+++ b/sql/sql_show.cc
@@ -1,4 +1,4 @@
-/* Copyright (c) 2000, 2012, Oracle and/or its affiliates. All rights reserved.
+/* Copyright (c) 2000, 2013, Oracle and/or its affiliates. All rights reserved.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -1176,6 +1176,7 @@ int store_create_info(THD *thd, TABLE_LIST *table_list, String *packet,
                                                        MODE_MYSQL323 |
                                                        MODE_MYSQL40)) != 0;
   my_bitmap_map *old_map;
+  int error= 0;
   DBUG_ENTER("store_create_info");
   DBUG_PRINT("enter",("table: %s", table->s->table_name.str));
 
@@ -1541,14 +1542,15 @@ int store_create_info(THD *thd, TABLE_LIST *table_list, String *packet,
                                                   NULL, NULL))))
     {
        table->part_info->set_show_version_string(packet);
-       packet->append(part_syntax, part_syntax_len);
-       packet->append(STRING_WITH_LEN(" */"));
+       if (packet->append(part_syntax, part_syntax_len) ||
+           packet->append(STRING_WITH_LEN(" */")))
+        error= 1;
        my_free(part_syntax);
     }
   }
 #endif
   tmp_restore_column_map(table->read_set, old_map);
-  DBUG_RETURN(0);
+  DBUG_RETURN(error);
 }
 
 
diff --git a/sql/sql_yacc.yy b/sql/sql_yacc.yy
index 7f1b5a57ac4..cf0346d6c0e 100644
--- a/sql/sql_yacc.yy
+++ b/sql/sql_yacc.yy
@@ -1,5 +1,5 @@
 /*
-   Copyright (c) 2000, 2011, Oracle and/or its affiliates. All rights reserved.
+   Copyright (c) 2000, 2013, Oracle and/or its affiliates. All rights reserved.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -4380,7 +4380,7 @@ partition:
         ;
 
 part_type_def:
-          opt_linear KEY_SYM '(' part_field_list ')'
+          opt_linear KEY_SYM opt_key_algo '(' part_field_list ')'
           {
             partition_info *part_info= Lex->part_info;
             part_info->list_of_part_fields= TRUE;
@@ -4406,6 +4406,25 @@ opt_linear:
           { Lex->part_info->linear_hash_ind= TRUE;}
         ;
 
+opt_key_algo:
+          /* empty */
+          { Lex->part_info->key_algorithm= partition_info::KEY_ALGORITHM_NONE;}
+        | ALGORITHM_SYM EQ real_ulong_num
+          {
+            switch ($3) {
+            case 1:
+              Lex->part_info->key_algorithm= partition_info::KEY_ALGORITHM_51;
+              break;
+            case 2:
+              Lex->part_info->key_algorithm= partition_info::KEY_ALGORITHM_55;
+              break;
+            default:
+              my_parse_error(ER(ER_SYNTAX_ERROR));
+              MYSQL_YYABORT;
+            }
+          }
+        ;
+
 part_field_list:
           /* empty */ {}
         | part_field_item_list {}
@@ -4487,7 +4506,7 @@ opt_sub_part:
         | SUBPARTITION_SYM BY opt_linear HASH_SYM sub_part_func
           { Lex->part_info->subpart_type= HASH_PARTITION; }
           opt_num_subparts {}
-        | SUBPARTITION_SYM BY opt_linear KEY_SYM
+        | SUBPARTITION_SYM BY opt_linear KEY_SYM opt_key_algo
           '(' sub_part_field_list ')'
           {
             partition_info *part_info= Lex->part_info;