percona-server-5.5.36-34.0.tar.gz

buf/buf0flu.c

@@ -1605,17 +1605,8 @@ buf_flush_page_and_try_neighbors(
 		ut_ad(block_mutex);
 	}
 
-	if (UNIV_UNLIKELY(buf_page_get_state(bpage)
-			  == BUF_BLOCK_REMOVE_HASH)) {
-
-		/* In case we don't hold the LRU list mutex, we may see a page
-		that is about to be relocated on the flush list.  Do not
-		attempt to flush it.  */
-		ut_ad(flush_type == BUF_FLUSH_LIST);
-		return (flushed);
-	}
-
-	ut_a(buf_page_in_file(bpage));
+	ut_a(buf_page_in_file(bpage)
+	     || buf_page_get_state(bpage) == BUF_BLOCK_REMOVE_HASH);
 
 	if (buf_flush_ready_for_flush(bpage, flush_type)) {
 		ulint		space;
@@ -1631,8 +1622,10 @@ buf_flush_page_and_try_neighbors(
 
 		/* These fields are protected by both the
 		buffer pool mutex and block mutex. */
-		space = buf_page_get_space(bpage);
-		offset = buf_page_get_page_no(bpage);
+		/* Read the fields directly in order to avoid asserting on
+		BUF_BLOCK_REMOVE_HASH pages. */
+		space = bpage->space;
+		offset = bpage->offset;
 
 		if (flush_type == BUF_FLUSH_LRU) {
 			mutex_exit(block_mutex);

dict/dict0dict.c

@@ -65,6 +65,7 @@ UNIV_INTERN uint	ibuf_debug;
 #include "srv0start.h" /* SRV_LOG_SPACE_FIRST_ID */
 #include "m_string.h"
 #include "my_sys.h"
+#include "lock0lock.h"
 
 #include <ctype.h>
 
@@ -1331,6 +1332,7 @@ dict_table_remove_from_cache(
 	ut_ad(table);
 	ut_ad(mutex_own(&(dict_sys->mutex)));
 	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+	ut_ad(lock_table_has_locks(table) == FALSE);
 
 #if 0
 	fputs("Removing table ", stderr);
@@ -1383,6 +1385,53 @@ dict_table_remove_from_cache(
 	dict_mem_table_free(table);
 }
 
+/**********************************************************************//**
+Test whether a table can be evicted from the LRU cache.
+@return TRUE if table can be evicted. */
+static
+ibool
+dict_table_can_be_evicted(
+/*======================*/
+	const dict_table_t*	table)	/*!< in: table to test */
+{
+	dict_index_t*	index;
+	dict_foreign_t*	foreign;
+	ibool		has_locks;
+
+	ut_ad(mutex_own(&dict_sys->mutex));
+
+	/* bug 758788: A table may not have an active handle opened on it
+	   but may still have locks held on it across multiple statements
+	   within an individual transaction. So a table is not evictable if
+	   there are locks held on it */
+	has_locks = lock_table_has_locks(table);
+
+	if (table->n_mysql_handles_opened || table->is_corrupt || has_locks)
+		return(FALSE);
+
+	/* bug 758788: We are not allowed to free the in-memory index struct
+	   dict_index_t if there are any locks held on any of its indexes. */
+	for (index = dict_table_get_first_index(table); index != NULL;
+	     index = dict_table_get_next_index(index)) {
+
+		rw_lock_t* lock = dict_index_get_lock(index);
+
+		if (rw_lock_is_locked(lock, RW_LOCK_SHARED)
+		    || rw_lock_is_locked(lock, RW_LOCK_EX)) {
+			return(FALSE);
+		} 
+	}
+
+	for (foreign = UT_LIST_GET_FIRST(table->foreign_list); foreign != NULL;
+	    foreign = UT_LIST_GET_NEXT(foreign_list, foreign)) {
+
+		if (foreign->referenced_table) {
+			return(FALSE);
+		}
+	}
+	return(TRUE);
+}
+
 /**************************************************************************
 Frees tables from the end of table_LRU if the dictionary cache occupies
 too much space. */
@@ -1394,51 +1443,65 @@ dict_table_LRU_trim(
 {
 	dict_table_t*	table;
 	dict_table_t*	prev_table;
-	dict_foreign_t*	foreign;
-	ulint		n_removed;
-	ulint		n_have_parent;
-	ulint		cached_foreign_tables;
+	ulint		dict_size;
+	ulint		max_depth;
+	ulint		max_evict;
+	ulint		visited;
+	ulint		evicted;
 
 #ifdef UNIV_SYNC_DEBUG
 	ut_ad(mutex_own(&(dict_sys->mutex)));
 #endif /* UNIV_SYNC_DEBUG */
 
-retry:
-	n_removed = n_have_parent = 0;
+	if (srv_dict_size_limit == 0)
+		return;
+
+	/* Calculate this once here and then once on every eviction */
+	dict_size = (dict_sys->table_hash->n_cells
+		    + dict_sys->table_id_hash->n_cells)
+		      * sizeof(hash_cell_t) + dict_sys->size;
+
+	/* Just some magic numbers to help keep us from holding the dict mutex
+	   for too long as well as preventing constant full scans of the list in
+	   a memory pressure situation. */
+
+	/* Don't go scanning into the front 50% of the list, chances are very
+	   good that there is nothing to be evicted in there and if there is,
+	   it should quickly get pushed into the back 50% */
+	max_depth = UT_LIST_GET_LEN(dict_sys->table_LRU) / 2;
+
+	/* Don't try to evict any more than 10% of evictable tables at once. */
+	max_evict = UT_LIST_GET_LEN(dict_sys->table_LRU) / 10;
+
+	visited = evicted = 0;
+
 	table = UT_LIST_GET_LAST(dict_sys->table_LRU);
 
-	while ( srv_dict_size_limit && table
-		&& ((dict_sys->table_hash->n_cells
-		     + dict_sys->table_id_hash->n_cells) * sizeof(hash_cell_t)
-		    + dict_sys->size) > srv_dict_size_limit ) {
+	while (table && dict_size > srv_dict_size_limit
+	       && visited <= max_depth
+	       && srv_shutdown_state == SRV_SHUTDOWN_NONE) {
+
 		prev_table = UT_LIST_GET_PREV(table_LRU, table);
 
-		if (table == self || table->n_mysql_handles_opened || table->is_corrupt)
-			goto next_loop;
+		if (table != self && dict_table_can_be_evicted(table)) {
+			dict_table_remove_from_cache(table);
 
-		cached_foreign_tables = 0;
-		foreign = UT_LIST_GET_FIRST(table->foreign_list);
-		while (foreign != NULL) {
-			if (foreign->referenced_table)
-				cached_foreign_tables++;
-			foreign = UT_LIST_GET_NEXT(foreign_list, foreign);
-		}
+			evicted++;
 
-		if (cached_foreign_tables == 0) {
-			dict_table_remove_from_cache(table);
-			n_removed++;
-		} else {
-			n_have_parent++;
+			if (evicted >= max_evict)
+				break;
+
+			/* Only need to recalculate this when something
+			   has been evicted */
+			dict_size = (dict_sys->table_hash->n_cells
+				    + dict_sys->table_id_hash->n_cells)
+				      * sizeof(hash_cell_t) + dict_sys->size;
 		}
-next_loop:
+
+		visited++;
+
 		table = prev_table;
 	}
-
-	if ( srv_dict_size_limit && n_have_parent && n_removed
-		&& ((dict_sys->table_hash->n_cells
-		     + dict_sys->table_id_hash->n_cells) * sizeof(hash_cell_t)
-		    + dict_sys->size) > srv_dict_size_limit )
-		goto retry;
 }
 
 /****************************************************************//**

dict/dict0mem.c

@@ -115,6 +115,7 @@ dict_mem_table_free(
 	ut_ad(table);
 	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
 	ut_d(table->cached = FALSE);
+	ut_d(table->magic_n = 0);
 
 #ifndef UNIV_HOTBACKUP
 	mutex_free(&(table->autoinc_mutex));

handler/ha_innodb.cc

@@ -8675,16 +8675,6 @@ innobase_rename_table(
 	error = row_rename_table_for_mysql(
 		norm_from, norm_to, trx, lock_and_commit);
 
-	if (error != DB_SUCCESS) {
-		FILE* ef = dict_foreign_err_file;
-
-		fputs("InnoDB: Renaming table ", ef);
-		ut_print_name(ef, trx, TRUE, norm_from);
-		fputs(" to ", ef);
-		ut_print_name(ef, trx, TRUE, norm_to);
-		fputs(" failed!\n", ef);
-	}
-
 	if (lock_and_commit) {
 		row_mysql_unlock_data_dictionary(trx);
 

include/lock0lock.h

@@ -459,6 +459,16 @@ lock_sec_rec_cons_read_sees(
 					by a read cursor */
 	const read_view_t*	view);	/*!< in: consistent read view */
 /*********************************************************************//**
+Check if there are any locks (table or rec) against table.
+@return	TRUE if locks exist */
+UNIV_INLINE
+ibool
+lock_table_has_locks(
+/*=================*/
+	const dict_table_t*	table);	/*!< in: check if there are any locks
+					held on records in this table or on the
+					table itself */
+/*********************************************************************//**
 Locks the specified database table in the mode given. If the lock cannot
 be granted immediately, the query thread is put to wait.
 @return	DB_SUCCESS, DB_LOCK_WAIT, DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */

include/lock0lock.ic

@@ -119,3 +119,17 @@ lock_get_min_heap_no(
 						   FALSE)));
 	}
 }
+
+/*******************************************************************//**
+Check if there are any locks (table or rec) against table.
+@return	TRUE if table has either table or record locks. */
+UNIV_INLINE
+ibool
+lock_table_has_locks(
+/*=================*/
+	const dict_table_t*	table)	/*!< in: check if there are any locks
+					held on records in this table or on the
+					table itself */
+{
+	return(UT_LIST_GET_LEN(table->locks) > 0);
+}

include/univ.i

@@ -64,7 +64,7 @@ component, i.e. we show M.N.P as M.N */
 	(INNODB_VERSION_MAJOR << 8 | INNODB_VERSION_MINOR)
 
 #ifndef PERCONA_INNODB_VERSION
-#define PERCONA_INNODB_VERSION 33.0
+#define PERCONA_INNODB_VERSION 34.0
 #endif
 
 #define INNODB_VERSION_STR	MYSQL_SERVER_VERSION "-" IB_TO_STR(PERCONA_INNODB_VERSION)

lock/lock0lock.c

@@ -3832,6 +3832,8 @@ lock_table_remove_low(
 	trx = lock->trx;
 	table = lock->un_member.tab_lock.table;
 
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+
 	/* Remove the table from the transaction's AUTOINC vector, if
 	the lock that is being release is an AUTOINC lock. */
 	if (lock_get_mode(lock) == LOCK_AUTO_INC) {

os/os0file.c

@@ -2365,12 +2365,13 @@ os_file_pread(
 	os_n_pending_reads++;
 	os_mutex_exit(os_file_count_mutex);
 
-	/* Handle signal interruptions correctly */
+	/* Handle partial reads and signal interruptions correctly */
 	for (n_bytes = 0; n_bytes < (ssize_t) n; ) {
-		n_read = pread(file, buf, (ssize_t)n, offs);
+		n_read = pread(file, buf, (ssize_t)n - n_bytes, offs);
 		if (n_read > 0) {
 			n_bytes += n_read;
 			offs += n_read;
+			buf = (char *)buf + n_read;
 		} else if (n_read == -1 && errno == EINTR) {
 			continue;
 		} else {
@@ -2493,12 +2494,13 @@ os_file_pwrite(
 	os_n_pending_writes++;
 	os_mutex_exit(os_file_count_mutex);
 
-	/* Handle signal interruptions correctly */
+	/* Handle partial writes and signal interruptions correctly */
 	for (ret = 0; ret < (ssize_t) n; ) {
-		n_written = pwrite(file, buf, (ssize_t)n, offs);
-		if (n_written > 0) {
+		n_written = pwrite(file, buf, (ssize_t)n - ret, offs);
+		if (n_written >= 0) {
 			ret += n_written;
 			offs += n_written;
+			buf = (char *)buf + n_written;
 		} else if (n_written == -1 && errno == EINTR) {
 			continue;
 		} else {
@@ -4849,6 +4851,7 @@ os_aio_linux_handle(
 	segment = os_aio_get_array_and_local_segment(&array, global_seg);
 	n = array->n_slots / array->n_segments;
 
+ wait_for_event:
 	/* Loop until we have found a completed request. */
 	for (;;) {
 		ibool	any_reserved = FALSE;
@@ -4918,6 +4921,43 @@ found:
 			ut_error;
 		}
 #endif /* UNIV_DO_FLUSH */
+	} else if ((slot->ret == 0) && (slot->n_bytes > 0)
+		   && (slot->n_bytes < (long) slot->len)) {
+		/* Partial read or write scenario */
+		int submit_ret;
+		struct iocb*    iocb;
+		slot->buf = (byte*)slot->buf + slot->n_bytes;
+		slot->offset = slot->offset + slot->n_bytes;
+		slot->len = slot->len - slot->n_bytes;
+		/* Resetting the bytes read/written */
+		slot->n_bytes = 0;
+		slot->io_already_done = FALSE;
+		iocb = &(slot->control);
+
+		if (slot->type == OS_FILE_READ) {
+			io_prep_pread(&slot->control, slot->file, slot->buf,
+				      slot->len, (off_t) slot->offset);
+		} else {
+			ut_a(slot->type == OS_FILE_WRITE);
+			io_prep_pwrite(&slot->control, slot->file, slot->buf,
+				       slot->len, (off_t) slot->offset);
+		}
+		/* Resubmit an I/O request */
+		submit_ret = io_submit(array->aio_ctx[segment], 1, &iocb);
+		if (submit_ret < 0 ) {
+			/* Aborting in case of submit failure */
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"InnoDB: Error: Native Linux AIO interface. "
+				"io_submit() call failed when resubmitting a "
+				"partial I/O request on the file %s.",
+				slot->name);
+			ut_error;
+		} else {
+			ret = FALSE;
+			os_mutex_exit(array->mutex);
+			goto wait_for_event;
+		}
 	} else {
 		errno = -slot->ret;