MDEV-16283 ALTER TABLE...DISCARD TABLESPACE still takes long on a large buffer pool

Also fixes MDEV-14727, MDEV-14491
InnoDB: Error: Waited for 5 secs for hash index ref_count (1) to drop to 0
by replacing the flawed wait logic in dict_index_remove_from_cache_low().

On DISCARD TABLESPACE, there is no need to drop the adaptive hash index.
We must drop it on IMPORT TABLESPACE, and eventually on DROP TABLE or
DROP INDEX. As long as the dict_index_t object remains in the cache
and the table remains inaccessible, the adaptive hash index entries
to orphaned pages would not do any harm. They would be dropped when
buffer pool pages are reused for something else.

btr_search_drop_page_hash_when_freed(), buf_LRU_drop_page_hash_batch():
Remove the parameter zip_size, and pass 0 to buf_page_get_gen().

buf_page_get_gen(): Ignore zip_size if mode==BUF_PEEK_IF_IN_POOL.

buf_LRU_drop_page_hash_for_tablespace(): Drop the adaptive hash index
even if the tablespace is inaccessible.

buf_LRU_drop_page_hash_for_tablespace(): New global function, to drop
the adaptive hash index.

buf_LRU_flush_or_remove_pages(), fil_delete_tablespace():
Remove the parameter drop_ahi.

dict_index_remove_from_cache_low(): Actively drop the adaptive hash index
if entries exist. This should prevent InnoDB hangs on DROP TABLE or
DROP INDEX.

row_import_for_mysql(): Drop any adaptive hash index entries for the table.

row_drop_table_for_mysql(): Drop any adaptive hash index for the table,
except if the table resides in the system tablespace. (DISCARD TABLESPACE
does not apply to the system tablespace, and we do no want to drop the
adaptive hash index for other tables than the one that is being dropped.)

row_truncate_table_for_mysql(): Drop any adaptive hash index entries for
the table, except if the table resides in the system tablespace.

Author: dr-m

storage/innobase/btr/btr0sea.cc

@@ -2,6 +2,7 @@
 
 Copyright (c) 1996, 2016, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2008, Google Inc.
+Copyright (c) 2018, MariaDB Corporation.
 
 Portions of this file contain modifications contributed and copyrighted by
 Google, Inc. Those modifications are gratefully acknowledged and are described
@@ -1250,17 +1251,11 @@ btr_search_drop_page_hash_index(
 	mem_free(folds);
 }
 
-/********************************************************************//**
-Drops a possible page hash index when a page is evicted from the buffer pool
-or freed in a file segment. */
+/** Drop possible adaptive hash index entries when a page is evicted
+from the buffer pool or freed in a file, or the index is being dropped. */
 UNIV_INTERN
 void
-btr_search_drop_page_hash_when_freed(
-/*=================================*/
-	ulint	space,		/*!< in: space id */
-	ulint	zip_size,	/*!< in: compressed page size in bytes
-				or 0 for uncompressed pages */
-	ulint	page_no)	/*!< in: page number */
+btr_search_drop_page_hash_when_freed(ulint space, ulint page_no)
 {
 	buf_block_t*	block;
 	mtr_t		mtr;
@@ -1273,7 +1268,7 @@ btr_search_drop_page_hash_when_freed(
 	are possibly holding, we cannot s-latch the page, but must
 	(recursively) x-latch it, even though we are only reading. */
 
-	block = buf_page_get_gen(space, zip_size, page_no, RW_X_LATCH, NULL,
+	block = buf_page_get_gen(space, 0, page_no, RW_X_LATCH, NULL,
 				 BUF_PEEK_IF_IN_POOL, __FILE__, __LINE__,
 				 &mtr);
 

storage/innobase/buf/buf0buf.cc

@@ -3075,17 +3075,18 @@ buf_page_get_gen(
 #ifdef UNIV_DEBUG
 	switch (mode) {
 	case BUF_EVICT_IF_IN_POOL:
+	case BUF_PEEK_IF_IN_POOL:
 		/* After DISCARD TABLESPACE, the tablespace would not exist,
 		but in IMPORT TABLESPACE, PageConverter::operator() must
 		replace any old pages, which were not evicted during DISCARD.
-		Skip the assertion on zip_size. */
+		Similarly, btr_search_drop_page_hash_when_freed() must
+		remove any old pages. Skip the assertion on zip_size. */
 		break;
 	case BUF_GET_NO_LATCH:
 		ut_ad(rw_latch == RW_NO_LATCH);
 		/* fall through */
 	case BUF_GET:
 	case BUF_GET_IF_IN_POOL:
-	case BUF_PEEK_IF_IN_POOL:
 	case BUF_GET_IF_IN_POOL_OR_WATCH:
 	case BUF_GET_POSSIBLY_FREED:
 		ut_ad(zip_size == fil_space_get_zip_size(space));
@@ -3257,7 +3258,8 @@ buf_page_get_gen(
 
 	fix_mutex = buf_page_get_mutex(&fix_block->page);
 
-	ut_ad(page_zip_get_size(&block->page.zip) == zip_size);
+	ut_ad(page_zip_get_size(&block->page.zip) == zip_size
+	      || mode == BUF_PEEK_IF_IN_POOL);
 
 	switch (mode) {
 	case BUF_GET_IF_IN_POOL:

storage/innobase/buf/buf0lru.cc

@@ -241,8 +241,6 @@ void
 buf_LRU_drop_page_hash_batch(
 /*=========================*/
 	ulint		space_id,	/*!< in: space id */
-	ulint		zip_size,	/*!< in: compressed page size in bytes
-					or 0 for uncompressed pages */
 	const ulint*	arr,		/*!< in: array of page_no */
 	ulint		count)		/*!< in: number of entries in array */
 {
@@ -252,8 +250,7 @@ buf_LRU_drop_page_hash_batch(
 	ut_ad(count <= BUF_LRU_DROP_SEARCH_SIZE);
 
 	for (i = 0; i < count; ++i) {
-		btr_search_drop_page_hash_when_freed(space_id, zip_size,
-						     arr[i]);
+		btr_search_drop_page_hash_when_freed(space_id, arr[i]);
 	}
 }
 
@@ -272,15 +269,6 @@ buf_LRU_drop_page_hash_for_tablespace(
 	buf_page_t*	bpage;
 	ulint*		page_arr;
 	ulint		num_entries;
-	ulint		zip_size;
-
-	zip_size = fil_space_get_zip_size(id);
-
-	if (UNIV_UNLIKELY(zip_size == ULINT_UNDEFINED)) {
-		/* Somehow, the tablespace does not exist.  Nothing to drop. */
-		ut_ad(0);
-		return;
-	}
 
 	page_arr = static_cast<ulint*>(ut_malloc(
 		sizeof(ulint) * BUF_LRU_DROP_SEARCH_SIZE));
@@ -333,8 +321,7 @@ buf_LRU_drop_page_hash_for_tablespace(
 		the latching order. */
 		buf_pool_mutex_exit(buf_pool);
 
-		buf_LRU_drop_page_hash_batch(
-			id, zip_size, page_arr, num_entries);
+		buf_LRU_drop_page_hash_batch(id, page_arr, num_entries);
 
 		num_entries = 0;
 
@@ -365,10 +352,32 @@ buf_LRU_drop_page_hash_for_tablespace(
 	buf_pool_mutex_exit(buf_pool);
 
 	/* Drop any remaining batch of search hashed pages. */
-	buf_LRU_drop_page_hash_batch(id, zip_size, page_arr, num_entries);
+	buf_LRU_drop_page_hash_batch(id, page_arr, num_entries);
 	ut_free(page_arr);
 }
 
+/** Drop the adaptive hash index for a tablespace.
+@param[in,out]	table	table */
+UNIV_INTERN void buf_LRU_drop_page_hash_for_tablespace(dict_table_t* table)
+{
+	for (dict_index_t* index = dict_table_get_first_index(table);
+	     index != NULL;
+	     index = dict_table_get_next_index(index)) {
+		if (btr_search_info_get_ref_count(
+			    btr_search_get_info(index))) {
+			goto drop_ahi;
+		}
+	}
+
+	return;
+drop_ahi:
+	ulint id = table->space;
+	for (ulint i = 0; i < srv_buf_pool_instances; i++) {
+		buf_LRU_drop_page_hash_for_tablespace(buf_pool_from_array(i),
+						      id);
+	}
+}
+
 /******************************************************************//**
 While flushing (or removing dirty) pages from a tablespace we don't
 want to hog the CPU and resources. Release the buffer pool and block
@@ -675,18 +684,11 @@ buf_flush_dirty_pages(buf_pool_t* buf_pool, ulint id, const trx_t* trx)
 /** Empty the flush list for all pages belonging to a tablespace.
 @param[in]	id		tablespace identifier
 @param[in]	trx		transaction, for checking for user interrupt;
-				or NULL if nothing is to be written
-@param[in]	drop_ahi	whether to drop the adaptive hash index */
-UNIV_INTERN
-void
-buf_LRU_flush_or_remove_pages(ulint id, const trx_t* trx, bool drop_ahi)
+				or NULL if nothing is to be written */
+UNIV_INTERN void buf_LRU_flush_or_remove_pages(ulint id, const trx_t* trx)
 {
 	for (ulint i = 0; i < srv_buf_pool_instances; i++) {
-		buf_pool_t* buf_pool = buf_pool_from_array(i);
-		if (drop_ahi) {
-			buf_LRU_drop_page_hash_for_tablespace(buf_pool, id);
-		}
-		buf_flush_dirty_pages(buf_pool, id, trx);
+		buf_flush_dirty_pages(buf_pool_from_array(i), id, trx);
 	}
 
 	if (trx && !trx_is_interrupted(trx)) {

storage/innobase/dict/dict0dict.cc

@@ -2,7 +2,7 @@
 
 Copyright (c) 1996, 2016, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2012, Facebook Inc.
-Copyright (c) 2013, 2017, MariaDB Corporation.
+Copyright (c) 2013, 2018, MariaDB Corporation.
 
 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
@@ -1674,7 +1674,7 @@ dict_table_rename_in_cache(
 			filepath = fil_make_ibd_name(table->name, false);
 		}
 
-		fil_delete_tablespace(table->space, true);
+		fil_delete_tablespace(table->space);
 
 		/* Delete any temp file hanging around. */
 		if (os_file_status(filepath, &exists, &ftype)
@@ -2719,35 +2719,13 @@ dict_index_remove_from_cache_low(
 	zero. See also: dict_table_can_be_evicted() */
 
 	do {
-		ulint ref_count = btr_search_info_get_ref_count(info);
-
-		if (ref_count == 0) {
+		if (!btr_search_info_get_ref_count(info)) {
 			break;
 		}
 
-		/* Sleep for 10ms before trying again. */
-		os_thread_sleep(10000);
-		++retries;
-
-		if (retries % 500 == 0) {
-			/* No luck after 5 seconds of wait. */
-			fprintf(stderr, "InnoDB: Error: Waited for"
-				" %lu secs for hash index"
-				" ref_count (%lu) to drop"
-				" to 0.\n"
-				"index: \"%s\""
-				" table: \"%s\"\n",
-				retries/100,
-				ref_count,
-				index->name,
-				table->name);
-		}
+		buf_LRU_drop_page_hash_for_tablespace(table);
 
-		/* To avoid a hang here we commit suicide if the
-		ref_count doesn't drop to zero in 600 seconds. */
-		if (retries >= 60000) {
-			ut_error;
-		}
+		ut_a(++retries < 10000);
 	} while (srv_shutdown_state == SRV_SHUTDOWN_NONE || !lru_evict);
 
 	rw_lock_free(&index->lock);

storage/innobase/fil/fil0fil.cc

@@ -2891,7 +2891,7 @@ fil_delete_tablespace(ulint id, bool drop_ahi)
 	To deal with potential read requests by checking the
 	::stop_new_ops flag in fil_io() */
 
-	buf_LRU_flush_or_remove_pages(id, NULL, drop_ahi);
+	buf_LRU_flush_or_remove_pages(id, NULL);
 
 #endif /* !UNIV_HOTBACKUP */
 
@@ -3002,7 +3002,7 @@ fil_discard_tablespace(
 {
 	dberr_t	err;
 
-	switch (err = fil_delete_tablespace(id, true)) {
+	switch (err = fil_delete_tablespace(id)) {
 	case DB_SUCCESS:
 		break;
 

storage/innobase/fsp/fsp0fsp.cc

@@ -1,7 +1,7 @@
 /*****************************************************************************
 
 Copyright (c) 1995, 2016, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2017, MariaDB Corporation.
+Copyright (c) 2017, 2018, MariaDB Corporation.
 
 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
@@ -3027,7 +3027,7 @@ fseg_free_page_low(
 	/* Drop search system page hash index if the page is found in
 	the pool and is hashed */
 
-	btr_search_drop_page_hash_when_freed(space, zip_size, page);
+	btr_search_drop_page_hash_when_freed(space, page);
 
 	descr = xdes_get_descriptor(space, zip_size, page, mtr);
 
@@ -3247,7 +3247,7 @@ fseg_free_extent(
 			found in the pool and is hashed */
 
 			btr_search_drop_page_hash_when_freed(
-				space, zip_size, first_page_in_extent + i);
+				space, first_page_in_extent + i);
 		}
 	}
 

storage/innobase/include/btr0sea.h

@@ -1,6 +1,7 @@
 /*****************************************************************************
 
 Copyright (c) 1996, 2016, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 2018, MariaDB Corporation.
 
 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
@@ -141,17 +142,11 @@ btr_search_drop_page_hash_index(
 				s- or x-latched, or an index page
 				for which we know that
 				block->buf_fix_count == 0 */
-/********************************************************************//**
-Drops a possible page hash index when a page is evicted from the buffer pool
-or freed in a file segment. */
+/** Drop possible adaptive hash index entries when a page is evicted
+from the buffer pool or freed in a file, or the index is being dropped. */
 UNIV_INTERN
 void
-btr_search_drop_page_hash_when_freed(
-/*=================================*/
-	ulint	space,		/*!< in: space id */
-	ulint	zip_size,	/*!< in: compressed page size in bytes
-				or 0 for uncompressed pages */
-	ulint	page_no);	/*!< in: page number */
+btr_search_drop_page_hash_when_freed(ulint space, ulint page_no);
 /********************************************************************//**
 Updates the page hash index when a single record is inserted on a page. */
 UNIV_INTERN

storage/innobase/include/buf0lru.h

@@ -1,7 +1,7 @@
 /*****************************************************************************
 
 Copyright (c) 1995, 2016, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2017, MariaDB Corporation.
+Copyright (c) 2017, 2018, MariaDB Corporation.
 
 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
@@ -34,6 +34,7 @@ Created 11/5/1995 Heikki Tuuri
 
 // Forward declaration
 struct trx_t;
+struct dict_table_t;
 
 /******************************************************************//**
 Returns TRUE if less than 25 % of the buffer pool is available. This can be
@@ -52,14 +53,15 @@ These are low-level functions
 /** Minimum LRU list length for which the LRU_old pointer is defined */
 #define BUF_LRU_OLD_MIN_LEN	512	/* 8 megabytes of 16k pages */
 
+/** Drop the adaptive hash index for a tablespace.
+@param[in,out]	table	table */
+UNIV_INTERN void buf_LRU_drop_page_hash_for_tablespace(dict_table_t* table);
+
 /** Empty the flush list for all pages belonging to a tablespace.
 @param[in]	id		tablespace identifier
 @param[in]	trx		transaction, for checking for user interrupt;
-				or NULL if nothing is to be written
-@param[in]	drop_ahi	whether to drop the adaptive hash index */
-UNIV_INTERN
-void
-buf_LRU_flush_or_remove_pages(ulint id, const trx_t* trx, bool drop_ahi=false);
+				or NULL if nothing is to be written */
+UNIV_INTERN void buf_LRU_flush_or_remove_pages(ulint id, const trx_t* trx);
 
 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
 /********************************************************************//**

storage/innobase/row/row0import.cc

@@ -31,6 +31,7 @@ Created 2012-02-08 by Sunny Bains.
 #endif
 
 #include "btr0pcur.h"
+#include "btr0sea.h"
 #include "que0que.h"
 #include "dict0boot.h"
 #include "ibuf0ibuf.h"
@@ -3983,6 +3984,17 @@ row_import_for_mysql(
 		return(row_import_cleanup(prebuilt, trx, err));
 	}
 
+	/* On DISCARD TABLESPACE, we did not drop any adaptive hash
+	index entries. If we replaced the discarded tablespace with a
+	smaller one here, there could still be some adaptive hash
+	index entries that point to cached garbage pages in the buffer
+	pool, because PageConverter::operator() only evicted those
+	pages that were replaced by the imported pages. We must
+	discard all remaining adaptive hash index entries, because the
+	adaptive hash index must be a subset of the table contents;
+	false positives are not tolerated. */
+	buf_LRU_drop_page_hash_for_tablespace(table);
+
 	row_mysql_lock_data_dictionary(trx);
 
 	/* If the table is stored in a remote tablespace, we need to

storage/innobase/row/row0mysql.cc

@@ -3516,6 +3516,8 @@ row_truncate_table_for_mysql(
 			fil_space_release(space);
 		}
 
+		buf_LRU_drop_page_hash_for_tablespace(table);
+
 		if (flags != ULINT_UNDEFINED
 		    && fil_discard_tablespace(space_id) == DB_SUCCESS) {
 
@@ -4209,6 +4211,21 @@ row_drop_table_for_mysql(
 		rw_lock_x_unlock(dict_index_get_lock(index));
 	}
 
+	if (table->space != TRX_SYS_SPACE) {
+		/* On DISCARD TABLESPACE, we would not drop the
+		adaptive hash index entries. If the tablespace is
+		missing here, delete-marking the record in SYS_INDEXES
+		would not free any pages in the buffer pool. Thus,
+		dict_index_remove_from_cache() would hang due to
+		adaptive hash index entries existing in the buffer
+		pool.  To prevent this hang, and also to guarantee
+		that btr_search_drop_page_hash_when_freed() will avoid
+		calling btr_search_drop_page_hash_index() while we
+		hold the InnoDB dictionary lock, we will drop any
+		adaptive hash index entries upfront. */
+		buf_LRU_drop_page_hash_for_tablespace(table);
+	}
+
 	/* We use the private SQL parser of Innobase to generate the
 	query graphs needed in deleting the dictionary data from system
 	tables in Innobase. Deleting a row from SYS_INDEXES table also