MDEV-15019 - InnoDB: store ReadView on trx

This will allow us to reduce critical section protected by trx_sys.mutex: - no need to maintain global m_free list - eliminate if (trx->read_view == NULL) condition. On x86_64 sizeof(Readview) is 144 mostly due to padding, sizeof(trx_t) with ReadView is 1200. Also don't close ReadView for read-write transactions, just mark it closed similarly to read-only. Clean-up: removed n_prepared_recovered_trx and n_prepared_trx, which accidentally re-appeared after some rebase.

Author: svoj

storage/innobase/handler/ha_innodb.cc

@@ -3503,7 +3503,7 @@ ha_innobase::init_table_handle_for_HANDLER(void)
 
 /* Assign a read view if the transaction does not have it yet */
 
-trx_assign_read_view(m_prebuilt->trx);
+trx_sys.mvcc.view_open(m_prebuilt->trx);
 
 innobase_register_trx(ht, m_user_thd, m_prebuilt->trx);
 
@@ -4394,7 +4394,7 @@ innobase_start_trx_and_assign_read_view(
 thd_get_trx_isolation(thd));
 
 if (trx->isolation_level == TRX_ISO_REPEATABLE_READ) {
-trx_assign_read_view(trx);
+trx_sys.mvcc.view_open(trx);
 } else {
 push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN,
  HA_ERR_UNSUPPORTED,
@@ -16220,7 +16220,7 @@ ha_innobase::external_lock(
 }
 
 } else if (trx->isolation_level <= TRX_ISO_READ_COMMITTED
- && MVCC::is_view_active(trx->read_view)) {
+ && trx->read_view.is_open()) {
 mutex_enter(&trx_sys.mutex);
 trx_sys.mvcc.view_close(trx->read_view);
 mutex_exit(&trx_sys.mutex);
@@ -16885,7 +16885,7 @@ ha_innobase::store_lock(
 (enum_tx_isolation) thd_tx_isolation(thd));
 
 if (trx->isolation_level <= TRX_ISO_READ_COMMITTED
- && MVCC::is_view_active(trx->read_view)) {
+ && trx->read_view.is_open()) {
 
 /* At low transaction isolation levels we let
 each consistent read set its own snapshot */

storage/innobase/handler/handler0alter.cc

@@ -5562,7 +5562,7 @@ prepare_inplace_alter_table_dict(
 if (ctx->online && ctx->num_to_add_index) {
 /* Assign a consistent read view for
 row_merge_read_clustered_index(). */
-trx_assign_read_view(ctx->prebuilt->trx);
+trx_sys.mvcc.view_open(ctx->prebuilt->trx);
 }
 
 if (fts_index) {

storage/innobase/include/read0read.h

@@ -31,60 +31,51 @@ Created 2/16/1997 Heikki Tuuri
 #include "read0types.h"
 
 /** The MVCC read view manager */
-class MVCC {
+class MVCC
+{
+ /** Active views. */
+ UT_LIST_BASE_NODE_T(ReadView) m_views;
+
+
+ /** Validates a read view list. */
+ bool validate() const;
 public:
-/** Constructor
-@param size	Number of views to pre-allocate */
-void create(ulint size);
-
-/** Destructor.
-Free all the views in the m_free list */
-void close();
-
-/**
-Allocate and create a view.
-@param view	view owned by this class created for the
-caller. Must be freed by calling close()
-@param trx	transaction creating the view */
-void view_open(ReadView*& view, trx_t* trx);
-
-/**
-Close a view created by the above function.
-@param view	view allocated by trx_open. */
-void view_close(ReadView*& view);
-
-/** Clones the oldest view and stores it in view. No need to
-call view_close(). The caller owns the view that is passed in.
-It will also move the closed views from the m_views list to the
-m_free list. This function is called by Purge to create it view.
-@param view	Preallocated view, owned by the caller */
-void clone_oldest_view(ReadView* view);
-
-/**
-@return the number of active views */
-ulint size() const;
-
-/**
-@return true if the view is active and valid */
-static bool is_view_active(ReadView* view)
-{
-return view && !view->is_closed();
-}
-
-private:
-
-/**
-Validates a read view list. */
-bool validate() const;
-
-typedef UT_LIST_BASE_NODE_T(ReadView) view_list_t;
-
-/** Free views ready for reuse. */
-view_list_tm_free;
-
-/** Active and closed views, the closed views will have the
-creator trx id set to TRX_ID_MAX */
-view_list_tm_views;
+ MVCC() { UT_LIST_INIT(m_views, &ReadView::m_view_list); }
+ ~MVCC() { ut_ad(UT_LIST_GET_LEN(m_views) == 0); }
+
+
+ /**
+ Allocate and create a view.
+ @param trx transaction creating the view
+ */
+ void view_open(trx_t *trx);
+
+
+ /**
+ Close a view created by the above function.
+ @param view view allocated by view_open.
+ */
+ void view_close(ReadView &view)
+ {
+ view.close();
+ view.set_registered(false);
+ UT_LIST_REMOVE(m_views, &view);
+ ut_ad(validate());
+ }
+
+
+ /**
+ Clones the oldest view and stores it in view. No need to
+ call view_close(). The caller owns the view that is passed in.
+ This function is called by Purge to create it view.
+
+ @param view Preallocated view, owned by the caller
+ */
+ void clone_oldest_view(ReadView *view);
+
+
+ /** @return the number of active views */
+ size_t size() const;
 };
 
 #endif /* read0read_h */

storage/innobase/include/read0types.h

@@ -198,19 +198,13 @@ class ReadView {
 {
 ut_ad(m_creator_trx_id != TRX_ID_MAX);
 m_creator_trx_id = TRX_ID_MAX;
+set_open(false);
 }
 
-/**
-@return true if the view is closed */
-bool is_closed() const
-{
-return(m_closed);
-}
-
-void set_closed(bool closed)
-{
-m_closed= closed;
-}
+bool is_open() const { return(m_open); }
+void set_open(bool open) { m_open= open; }
+bool is_registered() const { return(m_registered); }
+void set_registered(bool registered) { m_registered= registered; }
 
 /**
 Write the limits to the file.
@@ -325,14 +319,15 @@ class ReadView {
 they can be removed in purge if not needed by other views */
 trx_id_tm_low_limit_no;
 
-/** AC-NL-RO transaction view that has been "closed". */
-boolm_closed;
+/** true if view is open. */
+boolm_open;
 
-typedef UT_LIST_NODE_T(ReadView) node_t;
+/** true if transaction is in MVCC::m_views. Only thread that owns
+this view may access it. */
+boolm_registered;
 
-/** List of read views in trx_sys */
-byte	pad1[64 - sizeof(node_t)];
-node_tm_view_list;
+byte	pad1[CACHE_LINE_SIZE];
+UT_LIST_NODE_T(ReadView)	m_view_list;
 };
 
 #endif

storage/innobase/include/trx0sys.h

@@ -883,17 +883,6 @@ struct trx_sys_t {
 
  MY_ALIGNED(CACHE_LINE_SIZE) rw_trx_hash_t rw_trx_hash;
 
-ulint	n_prepared_trx;/*!< Number of transactions currently
-in the XA PREPARED state */
-
-ulint	n_prepared_recovered_trx; /*!< Number of transactions
-currently in XA PREPARED state that are
-also recovered. Such transactions cannot
-be added during runtime. They can only
-occur after recovery if mysqld crashed
-while there were XA PREPARED
-transactions. We disable query cache
-if such transactions exist. */
 
  /**
  Constructor.

storage/innobase/include/trx0trx.h

@@ -45,13 +45,11 @@ Created 3/26/1996 Heikki Tuuri
 #include "ut0vec.h"
 #include "fts0fts.h"
 #include "srv0srv.h"
+#include "read0types.h"
 
 // Forward declaration
 struct mtr_t;
 
-// Forward declaration
-class ReadView;
-
 // Forward declaration
 class FlushObserver;
 
@@ -290,23 +288,6 @@ void
 trx_mark_sql_stat_end(
 /*==================*/
 trx_t*	trx);/*!< in: trx handle */
-/********************************************************************//**
-Assigns a read view for a consistent read query. All the consistent reads
-within the same transaction will get the same read view, which is created
-when this function is first called for a new started transaction. */
-ReadView*
-trx_assign_read_view(
-/*=================*/
-trx_t*	trx);/*!< in: active transaction */
-
-/****************************************************************//**
-@return the transaction's read view or NULL if one not assigned. */
-UNIV_INLINE
-ReadView*
-trx_get_read_view(
-/*==============*/
-trx_t*	trx);
-
 /****************************************************************//**
 Prepares a transaction for commit/rollback. */
 void
@@ -571,7 +552,7 @@ Check transaction state */
 ut_ad(trx_state_eq((t), TRX_STATE_NOT_STARTED)	\
  || trx_state_eq((t), TRX_STATE_FORCED_ROLLBACK));	\
 ut_ad(!trx->has_logged());	\
-ut_ad(!MVCC::is_view_active((t)->read_view));	\
+ut_ad(!(t)->read_view.is_open());\
 ut_ad((t)->lock.wait_thr == NULL);	\
 ut_ad(UT_LIST_GET_LEN((t)->lock.trx_locks) == 0);	\
 ut_ad((t)->dict_operation == TRX_DICT_OP_NONE);	\
@@ -1000,7 +981,7 @@ struct trx_t {
 
 trx_state_tstate;
 
-ReadView*read_view;/*!< consistent read view used in the
+ReadView	read_view;/*!< consistent read view used in the
 transaction, or NULL if not yet set */
 
 UT_LIST_NODE_T(trx_t)

storage/innobase/include/trx0trx.ic

@@ -211,17 +211,6 @@ ok:
 trx->dict_operation = op;
 }
 
-/**
-@param trx	Get the active view for this transaction, if one exists
-@return the transaction's read view or NULL if one not assigned. */
-UNIV_INLINE
-ReadView*
-trx_get_read_view(
-trx_t*	trx)
-{
-return(!MVCC::is_view_active(trx->read_view) ? NULL : trx->read_view);
-}
-
 /**
 @param[in] trx	Transaction to check
 @return true if the transaction is a high priority transaction.*/

storage/innobase/lock/lock0lock.cc

@@ -415,7 +415,8 @@ lock_clust_rec_cons_read_sees(
 operate on same temp-table and so read of temp-table is
 always consistent read. */
 if (srv_read_only_mode || dict_table_is_temporary(index->table)) {
-ut_ad(view == 0 || dict_table_is_temporary(index->table));
+ut_ad(!view->is_open()
+ || dict_table_is_temporary(index->table));
 return(true);
 }
 
@@ -5719,15 +5720,13 @@ lock_trx_print_wait_and_mvcc_state(
 
 trx_print_latched(file, trx, 600);
 
-/* Note: this read_view access is data race. Further
-read_view->is_active() check is race condition. But it should
-"kind of work" because read_view is freed only at shutdown.
+/* Note: read_view->is_active() check is race condition. But it
+should "kind of work" because read_view is freed only at shutdown.
 Worst thing that may happen is that it'll get transferred to
 another thread and print wrong values. */
-const ReadView*	read_view = trx->read_view;
 
-if (read_view != NULL && !read_view->is_closed()) {
-read_view->print_limits(file);
+if (trx->read_view.is_open()) {
+trx->read_view.print_limits(file);
 }
 
 if (trx->lock.que_state == TRX_QUE_LOCK_WAIT) {