From: magnus.blaudd Date: September 28 2011 7:44pm Subject: bzr push into mysql-trunk-cluster branch (magnus.blaudd:3383 to 3390) List-Archive: http://lists.mysql.com/commits/141190 Message-Id: <201109281944.p8SJisTE031549@acsmt356.oracle.com> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit 3390 magnus.blaudd@stripped 2011-09-28 [merge] Merge 3389 magnus.blaudd@stripped 2011-09-28 ndb - build AQP as part of ndbcluster - build ndbcluster without join pushdown modified: storage/ndb/CMakeLists.txt 3388 magnus.blaudd@stripped 2011-09-28 [merge] Merge modified: sql/abstract_query_plan.cc sql/ha_ndbcluster.cc sql/ha_ndbcluster.h sql/handler.h 3387 magnus.blaudd@stripped 2011-09-28 ndb - resurrect abstract_query_plan.h/.cc, they are part of ndbcluster until adopted by server team added: sql/abstract_query_plan.cc sql/abstract_query_plan.h 3386 magnus.blaudd@stripped 2011-09-28 [merge] Merge modified: sql/ha_ndbcluster_binlog.cc storage/ndb/CMakeLists.txt 3385 magnus.blaudd@stripped 2011-09-28 Remove MCP patch for reduced footprint from mtr.pl modified: mysql-test/mysql-test-run.pl 3384 magnus.blaudd@stripped 2011-09-28 [merge] Merge trunk-cluster -> trunk-cluster-merge modified: sql/ha_ndbcluster.cc 3383 magnus.blaudd@stripped 2011-09-28 remove compiler warnings ignored by Cluster team modified: support-files/compiler_warnings.supp === modified file 'mysql-test/mysql-test-run.pl' --- a/mysql-test/mysql-test-run.pl 2011-09-27 12:11:16 +0000 +++ b/mysql-test/mysql-test-run.pl 2011-09-28 13:20:31 +0000 @@ -327,13 +327,6 @@ my $opt_max_test_fail= env_or_val(MTR_MA my $opt_parallel= $ENV{MTR_PARALLEL} || 1; -## MCP temp fixes to reduce footprint of large number of failing tests -$opt_max_save_core = 1; # Don't save many cores -$opt_max_save_datadir = 1; # Dont save many datadirs -$opt_max_test_fail = 0; # Allow many tests to fail -$opt_retry = 1; # Don't retry failed tests -## MCP temp fixes end - select(STDOUT); $| = 1; # Automatically flush STDOUT === added file 'sql/abstract_query_plan.cc' --- a/sql/abstract_query_plan.cc 1970-01-01 00:00:00 +0000 +++ b/sql/abstract_query_plan.cc 2011-09-28 19:25:31 +0000 @@ -0,0 +1,573 @@ +/* + Copyright 2010 Sun Microsystems, Inc. + All rights reserved. Use is subject to license terms. + + 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 + the Free Software Foundation; version 2 of the License. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA +*/ + +#ifdef USE_PRAGMA_IMPLEMENTATION +#pragma implementation // gcc: Class implementation +#endif + +#ifndef MYSQL_SERVER +#define MYSQL_SERVER +#endif + +#include "sql_priv.h" +#include "sql_select.h" +#include "abstract_query_plan.h" + + +#ifdef NDB_WITHOUT_JOIN_PUSHDOWN +static bool +test_if_skip_sort_order(JOIN_TAB *tab,ORDER *order,ha_rows select_limit, + bool no_changes, const key_map *map) +{ + // Stub used when compiling AQP as part of ndbcluster, never called + assert(false); + return false; +} +#endif + +namespace AQP +{ + + /** + @param join_tab Array of access methods constituting the nested loop join. + @param access_count Length of array. + */ + Join_plan::Join_plan(const JOIN* join) + : m_join_tabs(join->join_tab), + m_access_count(join->tables), + m_table_accesses(NULL) + { + /* + This combination is assumed not to appear. If it does, code must + be written to handle it. + */ + DBUG_ASSERT((m_join_tabs[0].use_quick != 2) + || (m_join_tabs[0].type == JT_ALL) + || (m_join_tabs[0].select == NULL) + || (m_join_tabs[0].select->quick == NULL)); + + m_table_accesses= new Table_access[m_access_count]; + for(uint i= 0; i < m_access_count; i++) + { + m_table_accesses[i].m_join_plan= this; + m_table_accesses[i].m_tab_no= i; + } + } + + Join_plan::~Join_plan() + { + delete[] m_table_accesses; + m_table_accesses= NULL; + } + + /** Get the JOIN_TAB of the n'th table access operation.*/ + const JOIN_TAB* Join_plan::get_join_tab(uint join_tab_no) const + { + DBUG_ASSERT(join_tab_no < m_access_count); + return m_join_tabs + join_tab_no; + } + + void + Join_plan::find_skippabable_group_or_order() const + { + const + JOIN* const join= m_join_tabs->join; + + if (join->const_tables < join->tables) + { + JOIN_TAB* join_head= join->join_tab+join->const_tables; + + m_group_by_filesort_is_skippable= join->group_optimized_away; + m_order_by_filesort_is_skippable= join->skip_sort_order; + + /* A single row don't have to be sorted */ + if (join_head->type == JT_CONST || + join_head->type == JT_SYSTEM || + join_head->type == JT_EQ_REF) + { + m_group_by_filesort_is_skippable= true; + m_order_by_filesort_is_skippable= true; + } + else if (join->select_options & SELECT_BIG_RESULT) + { + /* Excluded from ordered index optimization */ + } + else if (join->group_list && !m_group_by_filesort_is_skippable) + { + if (!join->tmp_table_param.quick_group || join->procedure) + { + /* Unsure how to handle - Is disabled in ::compute_type_and_index() */ + } + if (join->simple_group) + { + /** + test_if_skip_sort_order(...group_list...) already done by JOIN::optimize(). + As we still have a 'simple_group', GROUP BY has been optimized through an + access path providing an ordered sequence as required by GROUP BY: + + Verify this assumption in ASSERT below: + */ + DBUG_ASSERT(test_if_skip_sort_order(join_head, join->group_list, + join->unit->select_limit_cnt, true, + &join_head->table->keys_in_use_for_group_by)); + m_group_by_filesort_is_skippable= true; + } + } + else if (join->order && !m_order_by_filesort_is_skippable) + { + if (join->simple_order) + { + m_order_by_filesort_is_skippable= + test_if_skip_sort_order(join_head, + join->order, + join->unit->select_limit_cnt, false, + &join_head->table->keys_in_use_for_order_by); + } + } + } + } + + bool + Join_plan::group_by_filesort_is_skippable() const + { + return (m_group_by_filesort_is_skippable == true); + } + + bool + Join_plan::order_by_filesort_is_skippable() const + { + return (m_order_by_filesort_is_skippable == true); + } + + /** + Determine join type between this table access and some other table + access that preceeds it in the join plan.. + */ + enum_join_type + Table_access::get_join_type(const Table_access* predecessor) const + { + DBUG_ENTER("get_join_type"); + DBUG_ASSERT(get_access_no() > predecessor->get_access_no()); + + if (get_join_tab()->table->pos_in_table_list->outer_join != 0) + { + /* + This cover unnested outer joins such as + 'select * from t1 left join t2 on t1.attr=t1.pk'. + */ + DBUG_PRINT("info", ("JT_OUTER_JOIN between %s and %s", + predecessor->get_join_tab()->table->alias, + get_join_tab()->table->alias)); + DBUG_RETURN(JT_OUTER_JOIN); + } + + const TABLE_LIST* const child_embedding= + get_join_tab()->table->pos_in_table_list->embedding; + + if (child_embedding == NULL) + { + // 'this' is not on the inner side of any left join. + DBUG_PRINT("info", ("JT_INNER_JOIN between %s and %s", + predecessor->get_join_tab()->table->alias, + get_join_tab()->table->alias)); + DBUG_RETURN(JT_INNER_JOIN); + } + + DBUG_ASSERT(child_embedding->outer_join != 0); + + const TABLE_LIST *predecessor_embedding= + predecessor->get_join_tab()->table->pos_in_table_list->embedding; + + /* + This covers the nested join case, i.e: + LEFT JOIN (). + + TABLE_LIST objects form a tree where TABLE_LIST::emebedding points to + the parent object. Now if child_embedding is non null and not an + ancestor of predecessor_embedding in the embedding tree, then 'this' + must be on the inner side of some left join where 'predecessor' is on + the outer side. + */ + while (true) + { + if (predecessor_embedding == child_embedding) + { + DBUG_PRINT("info", ("JT_INNER_JOIN between %s and %s", + predecessor->get_join_tab()->table->alias, + get_join_tab()->table->alias)); + DBUG_RETURN(JT_INNER_JOIN); + } + else if (predecessor_embedding == NULL) + { + /* + We reached the root of the tree without finding child_embedding, + so it must be in another branch and hence on the inner side of some + left join where 'predecessor' is on the outer side. + */ + DBUG_PRINT("info", ("JT_OUTER_JOIN between %s and %s", + predecessor->get_join_tab()->table->alias, + get_join_tab()->table->alias)); + DBUG_RETURN(JT_OUTER_JOIN); + } + // Iterate through ancestors of predecessor_embedding. + predecessor_embedding = predecessor_embedding->embedding; + } + } + + /** + Get the number of key values for this operation. It is an error + to call this method on an operation that is not an index lookup + operation. + */ + uint Table_access::get_no_of_key_fields() const + { + DBUG_ASSERT(m_access_type == AT_PRIMARY_KEY || + m_access_type == AT_UNIQUE_KEY || + m_access_type == AT_MULTI_PRIMARY_KEY || + m_access_type == AT_MULTI_UNIQUE_KEY || + m_access_type == AT_ORDERED_INDEX_SCAN); // Used as 'range scan' + return get_join_tab()->ref.key_parts; + } + + /** + Get the field_no'th key values for this operation. It is an error + to call this method on an operation that is not an index lookup + operation. + */ + const Item* Table_access::get_key_field(uint field_no) const + { + DBUG_ASSERT(field_no < get_no_of_key_fields()); + return get_join_tab()->ref.items[field_no]; + } + + /** + Get the field_no'th KEY_PART_INFO for this operation. It is an error + to call this method on an operation that is not an index lookup + operation. + */ + const KEY_PART_INFO* Table_access::get_key_part_info(uint field_no) const + { + DBUG_ASSERT(field_no < get_no_of_key_fields()); + const KEY* key= &get_join_tab()->table->key_info[get_join_tab()->ref.key]; + return &key->key_part[field_no]; + } + + /** + Get the table that this operation accesses. + */ + TABLE* Table_access::get_table() const + { + return get_join_tab()->table; + } + + /** Get the JOIN_TAB object that corresponds to this operation.*/ + const JOIN_TAB* Table_access::get_join_tab() const + { + return m_join_plan->get_join_tab(m_tab_no); + } + + /** Get the Item_equal's set relevant for the specified 'Item_field' */ + Item_equal* + Table_access::get_item_equal(const Item_field* field_item) const + { + DBUG_ASSERT(field_item->type() == Item::FIELD_ITEM); + + COND_EQUAL* const cond_equal = get_join_tab()->join->cond_equal; + if (cond_equal!=NULL) + { + return (field_item->item_equal != NULL) + ? field_item->item_equal + : const_cast(field_item)->find_item_equal(cond_equal); + } + return NULL; + } + + /** + Write an entry in the trace file about the contents of this object. + */ + void Table_access::dbug_print() const + { + DBUG_PRINT("info", ("type:%d", get_join_tab()->type)); + DBUG_PRINT("info", ("ref.key:%d", get_join_tab()->ref.key)); + DBUG_PRINT("info", ("ref.key_parts:%d", get_join_tab()->ref.key_parts)); + DBUG_PRINT("info", ("ref.key_length:%d", get_join_tab()->ref.key_length)); + + DBUG_PRINT("info", ("order:%p", get_join_tab()->join->order)); + DBUG_PRINT("info", ("skip_sort_order:%d", + get_join_tab()->join->skip_sort_order)); + DBUG_PRINT("info", ("no_order:%d", get_join_tab()->join->no_order)); + DBUG_PRINT("info", ("simple_order:%d", get_join_tab()->join->simple_order)); + + DBUG_PRINT("info", ("group:%d", get_join_tab()->join->group)); + DBUG_PRINT("info", ("group_list:%p", get_join_tab()->join->group_list)); + DBUG_PRINT("info", ("simple_group:%d", get_join_tab()->join->simple_group)); + DBUG_PRINT("info", ("group_optimized_away:%d", + get_join_tab()->join->group_optimized_away)); + + DBUG_PRINT("info", ("full_join:%d", get_join_tab()->join->full_join)); + DBUG_PRINT("info", ("need_tmp:%d", get_join_tab()->join->need_tmp)); + DBUG_PRINT("info", ("select_distinct:%d", + get_join_tab()->join->select_distinct)); + + DBUG_PRINT("info", ("use_quick:%d", get_join_tab()->use_quick)); + DBUG_PRINT("info", ("index:%d", get_join_tab()->index)); + DBUG_PRINT("info", ("quick:%p", get_join_tab()->quick)); + DBUG_PRINT("info", ("select:%p", get_join_tab()->select)); + if (get_join_tab()->select && get_join_tab()->select->quick) + { + DBUG_PRINT("info", ("select->quick->get_type():%d", + get_join_tab()->select->quick->get_type())); + } + } + + + /** + Compute the access type and index (if apliccable) of this operation . + */ + void Table_access::compute_type_and_index() const + { + DBUG_ENTER("Table_access::compute_type_and_index"); + const JOIN_TAB* const join_tab= get_join_tab(); + JOIN* const join= join_tab->join; + + /** + * There are some JOIN arguments we don't fully understand or has + * not yet invested time into exploring pushability of: + */ + if (join->procedure) + { + m_access_type= AT_OTHER; + m_other_access_reason = + "'PROCEDURE'-clause post processing cannot be pushed."; + DBUG_VOID_RETURN; + } + + if (join->group_list && !join->tmp_table_param.quick_group) + { + m_access_type= AT_OTHER; + m_other_access_reason = + "GROUP BY cannot be done using index on grouped columns."; + DBUG_VOID_RETURN; + } + + /* Tables below 'const_tables' has been const'ified, or entirely + * optimized away due to 'impossible WHERE/ON' + */ + if (join_tab < join->join_tab+join->const_tables) + { + DBUG_PRINT("info", ("Operation %d is const-optimized.", m_tab_no)); + m_access_type= AT_FIXED; + DBUG_VOID_RETURN; + } + + /* First non-const table may provide 'simple' ordering for entire join */ + if (join_tab == join->join_tab+join->const_tables) + { + m_join_plan->find_skippabable_group_or_order(); + } + + /* + Identify the type of access operation and the index to use (if any). + */ + switch (join_tab->type) + { + case JT_EQ_REF: + case JT_CONST: + m_index_no= join_tab->ref.key; + + if (m_index_no == static_cast(join_tab->table->s->primary_key)) + { + DBUG_PRINT("info", ("Operation %d is a primary key lookup.", m_tab_no)); + m_access_type= AT_PRIMARY_KEY; + } + else + { + DBUG_PRINT("info", ("Operation %d is a unique index lookup.", + m_tab_no)); + m_access_type= AT_UNIQUE_KEY; + } + break; + + case JT_REF: + { + DBUG_ASSERT(join_tab->ref.key >= 0); + DBUG_ASSERT(join_tab->ref.key < MAX_KEY); + m_index_no= join_tab->ref.key; + + /* + All parts of a key are specified for an unique index -> access is a key lookup. + */ + const KEY *key_info= join_tab->table->s->key_info; + if (key_info[m_index_no].key_parts == join_tab->ref.key_parts && + key_info[m_index_no].flags & HA_NOSAME) + { + m_access_type= + (m_index_no == static_cast(join_tab->table->s->primary_key)) + ? AT_PRIMARY_KEY + : AT_UNIQUE_KEY; + DBUG_PRINT("info", ("Operation %d is an unique key referrence.", m_tab_no)); + } + else + { + DBUG_ASSERT(join_tab->ref.key_parts > 0); + DBUG_ASSERT(join_tab->ref.key_parts <= key_info[m_index_no].key_parts); + m_access_type= AT_ORDERED_INDEX_SCAN; + DBUG_PRINT("info", ("Operation %d is an ordered index scan.", m_tab_no)); + } + break; + } + case JT_NEXT: + DBUG_ASSERT(join_tab->index < MAX_KEY); + m_index_no= join_tab->index; + m_access_type= AT_ORDERED_INDEX_SCAN; + DBUG_PRINT("info", ("Operation %d is an ordered index scan.", m_tab_no)); + break; + + case JT_ALL: + if (join_tab->use_quick == 2) + { + /* + use_quick == 2 means that the decision on which access method to use + will be taken late (as rows from the preceeding operation arrive). + This operation is therefor not pushable. + */ + DBUG_PRINT("info", + ("Operation %d has 'use_quick == 2' -> not pushable", + m_tab_no)); + m_access_type= AT_UNDECIDED; + m_index_no= -1; + } + else + { + if (join_tab->select != NULL && + join_tab->select->quick != NULL) + { + QUICK_SELECT_I *quick= join_tab->select->quick; + + /** QUICK_SELECT results in execution of MRR (Multi Range Read). + * Depending on each range, it may require execution of + * either a PK-lookup or a range scan. To cover both of + * these we may need to prepare both a pushed lookup join + * and a pushed range scan. Currently we handle it as + * a range scan and convert e PK lookup to a (closed-) range + * whenever required. + **/ + + const KEY *key_info= join_tab->table->s->key_info; + DBUG_EXECUTE("info", quick->dbug_dump(0, TRUE);); + + // Temporary assert as we are still investigation the relation between + // 'quick->index == MAX_KEY' and the different quick_types + DBUG_ASSERT ((quick->index == MAX_KEY) == + ((quick->get_type() == QUICK_SELECT_I::QS_TYPE_INDEX_MERGE) || + (quick->get_type() == QUICK_SELECT_I::QS_TYPE_ROR_INTERSECT) || + (quick->get_type() == QUICK_SELECT_I::QS_TYPE_ROR_UNION))); + + // JT_INDEX_MERGE: We have a set of qualifying PKs as root of pushed joins + if (quick->index == MAX_KEY) + { + m_index_no= join_tab->table->s->primary_key; + m_access_type= AT_MULTI_PRIMARY_KEY; // Multiple PKs are produced by merge + } + + // Else JT_RANGE: May be both exact PK and/or index scans when sorted index available + else if (quick->index == join_tab->table->s->primary_key) + { + m_index_no= quick->index; + if (key_info[m_index_no].algorithm == HA_KEY_ALG_HASH) + m_access_type= AT_MULTI_PRIMARY_KEY; // MRR w/ multiple PK's + else + m_access_type= AT_MULTI_MIXED; // MRR w/ both range and PKs + } + else + { + m_index_no= quick->index; + if (key_info[m_index_no].algorithm == HA_KEY_ALG_HASH) + m_access_type= AT_MULTI_UNIQUE_KEY; // MRR with multiple unique keys + else + m_access_type= AT_MULTI_MIXED; // MRR w/ both range and unique keys + } + } + else + { + DBUG_PRINT("info", ("Operation %d is a table scan.", m_tab_no)); + m_access_type= AT_TABLE_SCAN; + } + } + break; + + default: + /* + Other join_types either cannot be pushed or the code analyze them is + not yet in place. + */ + DBUG_PRINT("info", + ("Operation %d has join_type %d. -> Not pushable.", + m_tab_no, join_tab->type)); + m_access_type= AT_OTHER; + m_index_no= -1; + m_other_access_reason = "This table access method can not be pushed."; + break; + } + DBUG_VOID_RETURN; + } + // Table_access::compute_type_and_index() + + + Table_access::Table_access() + :m_join_plan(NULL), + m_tab_no(0), + m_access_type(AT_VOID), + m_other_access_reason(NULL), + m_index_no(-1) + {} + + /** + @return True iff ordered index access is *required* from this operation. + */ + bool Table_access::is_fixed_ordered_index() const + { + const JOIN_TAB* const join_tab= get_join_tab(); + + /* For the QUICK_SELECT_I classes we can disable ordered index usage by + * setting 'QUICK_SELECT_I::sorted = false'. + * However, QUICK_SELECT_I::QS_TYPE_RANGE_DESC is special as its + * internal implementation requires its 'multi-ranges' to be retrieved + * in (descending) sorted order from the underlying table. + */ + if (join_tab->select != NULL && + join_tab->select->quick != NULL) + { + QUICK_SELECT_I *quick= join_tab->select->quick; + return (quick->get_type() == QUICK_SELECT_I::QS_TYPE_RANGE_DESC); + } + return false; + } + + /** + Check if the results from this operation will joined with results + from the next operation using a join buffer (instead of plain nested loop). + @return True if using a join buffer. + */ + bool Table_access::uses_join_cache() const + { + return get_join_tab()->next_select == sub_select_cache; + } + +}; +// namespace AQP === added file 'sql/abstract_query_plan.h' --- a/sql/abstract_query_plan.h 1970-01-01 00:00:00 +0000 +++ b/sql/abstract_query_plan.h 2011-09-28 19:25:31 +0000 @@ -0,0 +1,315 @@ +/* + Copyright 2010 Sun Microsystems, Inc. + All rights reserved. Use is subject to license terms. + + 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 + the Free Software Foundation; version 2 of the License. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA +*/ + +#ifndef ABSTRACT_QUERY_PLAN_H_INCLUDED +#define ABSTRACT_QUERY_PLAN_H_INCLUDED + +struct TABLE; +struct st_join_table; +typedef st_join_table JOIN_TAB; +class JOIN; +class Item; +class Item_field; +class Item_equal_iterator; + +/** + Abstract query plan (AQP) is an interface for examining certain aspects of + query plans without accessing mysqld internal classes (JOIN_TAB, SQL_SELECT + etc.) directly. + + AQP maps join execution plans, as represented by mysqld internals, to a set + of facade classes. Non-join operations such as sorting and aggregation is + currently *not* modelled in the AQP. + + The AQP models an n-way join as a sequence of the n table access operations + that the MySQL server would execute as part of its nested loop join + execution. (Each such table access operation is a scan of a table or index, + or an index lookup.) For each lookup operation, it is possible to examine + the expression that represents each field in the key. + + A storage enging will typically use the AQP for finding sections of a join + execution plan that may be executed in the engine rather than in mysqld. By + using the AQP rather than the mysqld internals directly, the coupling between + the engine and mysqld is reduced. +*/ +namespace AQP +{ + class Table_access; + + /** + This class represents a query plan for an n-way join, in the form a + sequence of n table access operations that will execute as a nested loop + join. + */ + class Join_plan + { + friend class Equal_set_iterator; + friend class Table_access; + public: + + explicit Join_plan(const JOIN* join); + + ~Join_plan(); + + const Table_access* get_table_access(uint access_no) const; + + uint get_access_count() const; + + /** + Can filesort(), normally required by execution of GROUP BY + or ORDER BY, be skipped due to the columns already being + accessible in required sorted order. + */ + bool group_by_filesort_is_skippable() const; + bool order_by_filesort_is_skippable() const; + + private: + /** + Array of the JOIN_TABs that are the internal representation of table + access operations. + */ + const JOIN_TAB* const m_join_tabs; + + /** Number of table access operations. */ + const uint m_access_count; + Table_access* m_table_accesses; + + mutable bool m_group_by_filesort_is_skippable; + mutable bool m_order_by_filesort_is_skippable; + + void find_skippabable_group_or_order() const; + + const JOIN_TAB* get_join_tab(uint join_tab_no) const; + + // No copying. + Join_plan(const Join_plan&); + Join_plan& operator=(const Join_plan&); + }; + // class Join_plan + + + /** + This class is an iterator for iterating over sets of fields (columns) that + should have the same value. For example, if the query is + SELECT * FROM T1, T2, T3 WHERE T1.b = T2.a AND T2.a = T3.a + then there would be such a set of {T1.b, T2.a, T3.a}. + */ + class Equal_set_iterator + { + public: + explicit Equal_set_iterator(Item_equal& item_equal) + : m_iterator(item_equal) {} + + const Item_field* next() + { return m_iterator++; } + + private: + /** + This class is implemented in terms of this mysqld internal class. + */ + Item_equal_iterator m_iterator; + + // No copying. + Equal_set_iterator(const Equal_set_iterator&); + Equal_set_iterator& operator=(const Equal_set_iterator&); + }; + // class Equal_set_iterator + + /** The type of a table access operation. */ + enum enum_access_type + { + /** For default initialization.*/ + AT_VOID, + /** Value has already been fetched / determined by optimizer.*/ + AT_FIXED, + /** Do a lookup of a single primary key.*/ + AT_PRIMARY_KEY, + /** Do a lookup of a single unique index key.*/ + AT_UNIQUE_KEY, + /** Scan an ordered index with a single upper and lower bound pair.*/ + AT_ORDERED_INDEX_SCAN, + /** Do a multi range read for a set of primary keys.*/ + AT_MULTI_PRIMARY_KEY, + /** Do a multi range read for a set of unique index keys.*/ + AT_MULTI_UNIQUE_KEY, + /** + Do a multi range read for a mix of ranges (for which there is an + ordered index), and either primary keys or unique index keys. + */ + AT_MULTI_MIXED, + /** Scan a table. (No index is assumed to be used.) */ + AT_TABLE_SCAN, + /** Access method will not be chosen before the execution phase.*/ + AT_UNDECIDED, + /** + The access method has properties that prevents it from being pushed to a + storage engine. + */ + AT_OTHER + }; + + /** The type of join operation require */ + enum enum_join_type + { + JT_OUTER_JOIN, + JT_INNER_JOIN, + JT_SEMI_JOIN + }; + + /** + This class represents an access operation on a table, such as a table + scan, or a scan or lookup via an index. A Table_access object is always + owned by a Join_plan object, such that the life time of the Table_access + object ends when the life time of the owning Join_plan object ends. + */ + class Table_access + { + friend class Join_plan; + friend inline bool equal(const Table_access*, const Table_access*); + public: + + const Join_plan* get_join_plan() const; + + enum_access_type get_access_type() const; + + const char* get_other_access_reason() const; + + enum_join_type get_join_type(const Table_access* parent) const; + + uint get_no_of_key_fields() const; + + const Item* get_key_field(uint field_no) const; + + const KEY_PART_INFO* get_key_part_info(uint field_no) const; + + uint get_access_no() const; + + int get_index_no() const; + + TABLE* get_table() const; + + Item_equal* get_item_equal(const Item_field* field_item) const; + + void dbug_print() const; + + bool is_fixed_ordered_index() const; + + bool uses_join_cache() const; + + private: + + /** Backref. to the Join_plan which this Table_access is part of */ + const Join_plan* m_join_plan; + + /** This operation corresponds to m_root_tab[m_tab_no].*/ + uint m_tab_no; + + /** The type of this operation.*/ + mutable enum_access_type m_access_type; + + /** + The reason for getting m_access_type==AT_OTHER. Used for explain extended. + */ + mutable const char* m_other_access_reason; + + /** The index to use for this operation (if applicable )*/ + mutable int m_index_no; + + explicit Table_access(); + + const JOIN_TAB* get_join_tab() const; + + void compute_type_and_index() const; + + /** No copying*/ + Table_access(const Table_access&); + Table_access& operator=(const Table_access&); + }; + // class Table_access + + /** + Get the n'th table access operation. + @param access_no The index of the table access operation to fetch. + @return The access_no'th table access operation. + */ + inline const Table_access* Join_plan::get_table_access(uint access_no) const + { + DBUG_ASSERT(access_no < m_access_count); + return m_table_accesses + access_no; + } + + /** + @return The number of table access operations in the nested loop join. + */ + inline uint Join_plan::get_access_count() const + { + return m_access_count; + } + + /** Get the Join_plan that this Table_access belongs to.*/ + inline const Join_plan* Table_access::get_join_plan() const + { + return m_join_plan; + } + + /** Get the type of this operation.*/ + inline enum_access_type Table_access::get_access_type() const + { + if (m_access_type == AT_VOID) + compute_type_and_index(); + return m_access_type; + } + + /** + Get a description of the reason for getting access_type==AT_OTHER. To be + used for informational messages. + @return A string that should be assumed to have the same life time as the + Table_access object. + */ + inline const char* Table_access::get_other_access_reason() const + { + if (m_access_type == AT_VOID) + compute_type_and_index(); + return m_other_access_reason; + } + + /** + @return The number of the index to use for this access operation ( + or -1 for non-index operations). + */ + inline int Table_access::get_index_no() const + { + if (m_access_type == AT_VOID) + compute_type_and_index(); + + return m_index_no; + } + + /** + Get the number of this Table_access within the enclosing Join_plan. + (This number will be in the range 0 to Join_plan::get_access_count() - 1.) + */ + inline uint Table_access::get_access_no() const + { + return m_tab_no; + } + +}; +// namespace AQP + +#endif === modified file 'sql/ha_ndbcluster.cc' --- a/sql/ha_ndbcluster.cc 2011-09-27 12:11:16 +0000 +++ b/sql/ha_ndbcluster.cc 2011-09-28 19:11:28 +0000 @@ -896,7 +896,7 @@ SHOW_VAR ndb_status_index_stat_variables {NullS, NullS, SHOW_LONG} }; -#ifndef NO_PUSHED_JOIN +#ifndef NDB_WITHOUT_JOIN_PUSHDOWN static int ndbcluster_make_pushed_join(handlerton *, THD*,AQP::Join_plan*); #endif @@ -3069,17 +3069,7 @@ static const ulong index_type_flags[]= HA_READ_NEXT | HA_READ_PREV | HA_READ_RANGE | - HA_READ_ORDER | - /* - NOTE 1: our ordered indexes are not really clustered - but since accesing data when scanning index is free - it's a good approxiamtion - - NOTE 2: We really should consider DD attributes here too - (for which there is IO to read data when scanning index) - but that will need to handled later... - */ - HA_CLUSTERED_INDEX, + HA_READ_ORDER, /* UNIQUE_INDEX */ HA_ONLY_WHOLE_INDEX, @@ -3088,15 +3078,13 @@ static const ulong index_type_flags[]= HA_READ_NEXT | HA_READ_PREV | HA_READ_RANGE | - HA_READ_ORDER | - HA_CLUSTERED_INDEX, + HA_READ_ORDER, /* ORDERED_INDEX */ HA_READ_NEXT | HA_READ_PREV | HA_READ_RANGE | - HA_READ_ORDER | - HA_CLUSTERED_INDEX + HA_READ_ORDER }; static const int index_flags_size= sizeof(index_type_flags)/sizeof(ulong); @@ -3134,9 +3122,24 @@ inline ulong ha_ndbcluster::index_flags( bool ha_ndbcluster::primary_key_is_clustered() { - if (table->s->primary_key != MAX_KEY) - return test(index_flags(table->s->primary_key, 0, 0) & HA_CLUSTERED_INDEX); - return FALSE; + + if (table->s->primary_key == MAX_KEY) + return false; + + /* + NOTE 1: our ordered indexes are not really clustered + but since accesing data when scanning index is free + it's a good approximation + + NOTE 2: We really should consider DD attributes here too + (for which there is IO to read data when scanning index) + but that will need to be handled later... + */ + const ndb_index_type idx_type = + get_index_type_from_table(table->s->primary_key); + return (idx_type == PRIMARY_KEY_ORDERED_INDEX || + idx_type == UNIQUE_ORDERED_INDEX || + idx_type == ORDERED_INDEX); } bool ha_ndbcluster::check_index_fields_in_write_set(uint keyno) @@ -6474,11 +6477,7 @@ int ha_ndbcluster::index_first(uchar *bu // Start the ordered index scan and fetch the first row // Only HA_READ_ORDER indexes get called by index_first -#ifdef MCP_BUG11764737 - const int error= ordered_index_scan(0, 0, TRUE, FALSE, buf, NULL); -#else const int error= ordered_index_scan(0, 0, m_sorted, FALSE, buf, NULL); -#endif table->status=error ? STATUS_NOT_FOUND: 0; DBUG_RETURN(error); } @@ -6488,11 +6487,7 @@ int ha_ndbcluster::index_last(uchar *buf { DBUG_ENTER("ha_ndbcluster::index_last"); ha_statistic_increment(&SSV::ha_read_last_count); -#ifdef MCP_BUG11764737 - const int error= ordered_index_scan(0, 0, TRUE, TRUE, buf, NULL); -#else const int error= ordered_index_scan(0, 0, m_sorted, TRUE, buf, NULL); -#endif table->status=error ? STATUS_NOT_FOUND: 0; DBUG_RETURN(error); } @@ -11853,7 +11848,9 @@ static int ndbcluster_init(void *p) h->discover= ndbcluster_discover; h->find_files= ndbcluster_find_files; h->table_exists_in_engine= ndbcluster_table_exists_in_engine; +#ifndef NDB_WITHOUT_JOIN_PUSHDOWN h->make_pushed_join= ndbcluster_make_pushed_join; +#endif } // Initialize ndb interface @@ -14300,7 +14297,7 @@ ha_ndbcluster::read_multi_range_fetch_ne } #endif -#ifndef NO_PUSHED_JOIN +#ifndef NDB_WITHOUT_JOIN_PUSHDOWN /** * Try to find pushable subsets of a join plan. @@ -14357,7 +14354,9 @@ int ndbcluster_make_pushed_join(handlert } DBUG_RETURN(0); } // ndbcluster_make_pushed_join - +#endif + + /** * In case a pushed join having the table for this handler as its root * has been produced. ::assign_pushed_join() is responsible for setting @@ -14640,7 +14639,6 @@ ha_ndbcluster::test_push_flag(enum ha_pu DBUG_RETURN(false); } -#endif /** @param[in] comment table comment defined by user === modified file 'sql/ha_ndbcluster.h' --- a/sql/ha_ndbcluster.h 2011-09-23 12:53:33 +0000 +++ b/sql/ha_ndbcluster.h 2011-09-28 19:11:28 +0000 @@ -425,6 +425,13 @@ static void set_tabname(const char *path bool maybe_pushable_join(const char*& reason) const; int assign_pushed_join(const ndb_pushed_join* pushed_join); +#ifdef NDB_WITHOUT_JOIN_PUSHDOWN + enum ha_push_flag { + HA_PUSH_BLOCK_CONST_TABLE, + HA_PUSH_MULTIPLE_DEPENDENCY, + HA_PUSH_NO_ORDERED_INDEX + }; +#endif bool test_push_flag(enum ha_push_flag flag) const; uint number_of_pushed_joins() const; === modified file 'sql/ha_ndbcluster_binlog.cc' --- a/sql/ha_ndbcluster_binlog.cc 2011-09-23 12:53:33 +0000 +++ b/sql/ha_ndbcluster_binlog.cc 2011-09-28 14:58:56 +0000 @@ -644,6 +644,7 @@ ndbcluster_binlog_index_purge_file(THD * } +#ifndef NDB_WITHOUT_DIST_PRIV // Determine if privilege tables are distributed, ie. stored in NDB static bool priv_tables_are_in_ndb(THD *thd) @@ -679,7 +680,7 @@ priv_tables_are_in_ndb(THD *thd) } DBUG_RETURN(distributed); } - +#endif static void ndbcluster_binlog_log_query(handlerton *hton, THD *thd, enum_binlog_command binlog_command, @@ -731,6 +732,7 @@ ndbcluster_binlog_log_query(handlerton * type= SOT_DROP_DB; DBUG_ASSERT(FALSE); break; +#ifndef NDB_WITHOUT_DIST_PRIV case LOGCOM_CREATE_USER: type= SOT_CREATE_USER; if (priv_tables_are_in_ndb(thd)) @@ -771,6 +773,7 @@ ndbcluster_binlog_log_query(handlerton * log= 1; } break; +#endif } if (log) { === modified file 'sql/handler.h' --- a/sql/handler.h 2011-09-28 09:10:16 +0000 +++ b/sql/handler.h 2011-09-28 19:25:31 +0000 @@ -181,6 +181,7 @@ #define HA_DO_INDEX_COND_PUSHDOWN 256 /* Supports Index Condition Pushdown */ + /* bits in alter_table_flags: */ === modified file 'storage/ndb/CMakeLists.txt' --- a/storage/ndb/CMakeLists.txt 2011-09-23 12:53:33 +0000 +++ b/storage/ndb/CMakeLists.txt 2011-09-28 19:36:09 +0000 @@ -87,6 +87,16 @@ IF(NOT MYSQL_CLUSTER_VERSION) # Online alter table not supported in non MySQL Cluster # versions yet, compile ndbcluster without online alter support ADD_DEFINITIONS(-DNDB_WITHOUT_ONLINE_ALTER) + # Distributed privilege tables not supported in non + # MySQL Cluster version yet, compile ndbcluster without dist priv + ADD_DEFINITIONS(-DNDB_WITHOUT_DIST_PRIV) + # Pushdown of join queries not supported in non + # MySQL Cluster version yet, compile ndbcluster without it + ADD_DEFINITIONS(-DNDB_WITHOUT_JOIN_PUSHDOWN) + # Since the MySQL Server has no AQP support, compile + # that into ndbcluster instead + SET(NDBCLUSTER_SOURCES ${NDBCLUSTER_SOURCES} + ../../sql/abstract_query_plan.cc) ENDIF() # NDB is DEFAULT plugin in MySQL Cluster No bundle (reason: useless for push emails).