Feature: In first write if we trim we set write_size to actual bytes

written and rest of the page is trimmed. In following writes
there is no need to trim again if write_size only increases
because rest of the page is already trimmed. If actual write
size decreases we need to trim again. Need to research if this
can happen frequently enough to make any effect.

7 files changed, 44 insertions, 21 deletions

diff --git a/storage/innobase/buf/buf0dblwr.cc b/storage/innobase/buf/buf0dblwr.cc
index 933b56eaf88..2ae67d8a41e 100644
--- a/storage/innobase/buf/buf0dblwr.cc
+++ b/storage/innobase/buf/buf0dblwr.cc
@@ -728,7 +728,7 @@ buf_dblwr_write_block_to_datafile(
 	fil_io(OS_FILE_WRITE | OS_AIO_SIMULATED_WAKE_LATER,
 	       FALSE, buf_block_get_space(block), 0,
 	       buf_block_get_page_no(block), 0, UNIV_PAGE_SIZE,
-	       (void*) block->frame, (void*) block, 0);
+		(void*) block->frame, (void*) block, (ulint *)&bpage->write_size);
 }
 
 /********************************************************************//**
diff --git a/storage/innobase/buf/buf0flu.cc b/storage/innobase/buf/buf0flu.cc
index 06ae7b5375c..b5f1aeef597 100644
--- a/storage/innobase/buf/buf0flu.cc
+++ b/storage/innobase/buf/buf0flu.cc
@@ -942,7 +942,7 @@ buf_flush_write_block_low(
 		       FALSE, buf_page_get_space(bpage), zip_size,
 		       buf_page_get_page_no(bpage), 0,
 		       zip_size ? zip_size : UNIV_PAGE_SIZE,
-		       frame, bpage, 0);
+		       frame, bpage, &bpage->write_size);
 	} else if (flush_type == BUF_FLUSH_SINGLE_PAGE) {
 		buf_dblwr_write_single_page(bpage);
 	} else {
diff --git a/storage/innobase/fil/fil0fil.cc b/storage/innobase/fil/fil0fil.cc
index 8a416d09c94..0bec85c699a 100644
--- a/storage/innobase/fil/fil0fil.cc
+++ b/storage/innobase/fil/fil0fil.cc
@@ -441,7 +441,7 @@ fil_read(
 				in aio this must be appropriately aligned */
 	void*	message,	/*!< in: message for aio handler if non-sync
 				aio used, else ignored */
-	ulint	write_size)	/*!< in/out: Actual write size initialized
+	ulint*	write_size)	/*!< in/out: Actual write size initialized
 				after fist successfull trim
 				operation for this page and if
 				initialized we do not trim again if
@@ -475,7 +475,7 @@ fil_write(
 				this must be appropriately aligned */
 	void*	message,	/*!< in: message for aio handler if non-sync
 				aio used, else ignored */
-	ulint	write_size)	/*!< in/out: Actual write size initialized
+	ulint*	write_size)	/*!< in/out: Actual write size initialized
 				after fist successfull trim
 				operation for this page and if
 				initialized we do not trim again if
@@ -5288,7 +5288,7 @@ fil_io(
 				appropriately aligned */
 	void*	message,	/*!< in: message for aio handler if non-sync
 				aio used, else ignored */
-	ulint	write_size)	/*!< in/out: Actual write size initialized
+	ulint*	write_size)	/*!< in/out: Actual write size initialized
 				after fist successfull trim
 				operation for this page and if
 				initialized we do not trim again if
diff --git a/storage/innobase/include/fil0fil.h b/storage/innobase/include/fil0fil.h
index c5edd33f46b..01084d52365 100644
--- a/storage/innobase/include/fil0fil.h
+++ b/storage/innobase/include/fil0fil.h
@@ -753,7 +753,7 @@ fil_io(
 				appropriately aligned */
 	void*	message,	/*!< in: message for aio handler if non-sync
 				aio used, else ignored */
-	ulint	write_size)	/*!< in/out: Actual write size initialized
+	ulint*	write_size)	/*!< in/out: Actual write size initialized
 			       after fist successfull trim
 			       operation for this page and if
 			       initialized we do not trim again if
diff --git a/storage/innobase/include/os0file.h b/storage/innobase/include/os0file.h
index 3c70f9925fe..eb5e1dddaf5 100644
--- a/storage/innobase/include/os0file.h
+++ b/storage/innobase/include/os0file.h
@@ -724,7 +724,11 @@ pfs_os_aio_func(
 				(can be used to identify a completed
 				aio operation); ignored if mode is
                                 OS_AIO_SYNC */
-	ibool		atomic_writes, /*!<in TRUE if atomic writes are used */
+	ulint*		write_size,/*!< in/out: Actual write size initialized
+			       after fist successfull trim
+			       operation for this page and if
+			       initialized we do not trim again if
+			       actual page size does not decrease. */
 	const char*	src_file,/*!< in: file name where func invoked */
 	ulint		src_line);/*!< in: line where the func invoked */
 /*******************************************************************//**
@@ -1057,7 +1061,7 @@ os_aio_func(
 				(can be used to identify a completed
 				aio operation); ignored if mode is
 				OS_AIO_SYNC */
-	ulint		write_size);/*!< in/out: Actual write size initialized
+	ulint*		write_size);/*!< in/out: Actual write size initialized
 			       after fist successfull trim
 			       operation for this page and if
 			       initialized we do not trim again if
diff --git a/storage/innobase/include/os0file.ic b/storage/innobase/include/os0file.ic
index 2be0f6a8d97..ca98428dd49 100644
--- a/storage/innobase/include/os0file.ic
+++ b/storage/innobase/include/os0file.ic
@@ -214,7 +214,11 @@ pfs_os_aio_func(
 				(can be used to identify a completed
 				aio operation); ignored if mode is
                                 OS_AIO_SYNC */
-	ibool		atomic_writes, /*!<in TRUE if atomic writes are used */
+	ulint*		write_size,/*!< in/out: Actual write size initialized
+			       after fist successfull trim
+			       operation for this page and if
+			       initialized we do not trim again if
+			       actual page size does not decrease. */
 	const char*	src_file,/*!< in: file name where func invoked */
 	ulint		src_line)/*!< in: line where the func invoked */
 {
@@ -230,7 +234,7 @@ pfs_os_aio_func(
 				   src_file, src_line);
 
 	result = os_aio_func(type, mode, name, file, buf, offset,
-			     n, message1, message2, atomic_writes);
+		n, message1, message2, write_size);
 
 	register_pfs_file_io_end(locker, n);
 
diff --git a/storage/innobase/os/os0file.cc b/storage/innobase/os/os0file.cc
index 9f12ca86601..6bb9e47b116 100644
--- a/storage/innobase/os/os0file.cc
+++ b/storage/innobase/os/os0file.cc
@@ -196,7 +196,7 @@ struct os_aio_slot_t{
 					       freed after the write
 					       has been completed */
 
-	ulint           write_size;     /*!< Actual write size initialized
+	ulint*          write_size;     /*!< Actual write size initialized
 					after fist successfull trim
 					operation for this page and if
 					initialized we do not trim again if
@@ -4363,7 +4363,7 @@ os_aio_array_reserve_slot(
 				to write */
 	os_offset_t	offset,	/*!< in: file offset */
 	ulint		len,	/*!< in: length of the block to read or write */
-	ulint		write_size)     /*!< in: Actual write size initialized
+	ulint*		write_size)     /*!< in: Actual write size initialized
 			       after fist successfull trim
 			       operation for this page and if
 			       initialized we do not trim again if
@@ -4783,7 +4783,7 @@ os_aio_func(
 				(can be used to identify a completed
 				aio operation); ignored if mode is
 				OS_AIO_SYNC */
-	ulint		write_size)/*!< in/out: Actual write size initialized
+	ulint*		write_size)/*!< in/out: Actual write size initialized
 			       after fist successfull trim
 			       operation for this page and if
 			       initialized we do not trim again if
@@ -6163,13 +6163,20 @@ os_file_trim(
 	// because rest of the page is already trimmed. If actual write
 	// size decreases we need to trim again.
 	if (trim_len == 0 ||
-	    (slot->write_size > 0 && len >= slot->write_size)) {
+	    (slot->write_size &&
+		    *slot->write_size > 0 &&
+		    len >= *slot->write_size)) {
 
-		if (slot->write_size > 0 && len >= slot->write_size) {
+#ifdef UNIV_DEBUG
+		fprintf(stderr, "Note: TRIM: write_size %lu trim_len %lu len %lu\n",
+			*slot->write_size, trim_len, len);
+#endif
+
+		if (*slot->write_size > 0 && len >= *slot->write_size) {
 			srv_stats.page_compressed_trim_op_saved.inc();
 		}
 
-		slot->write_size = len;
+		*slot->write_size = len;
 
 		return (TRUE);
 	}
@@ -6191,11 +6198,15 @@ os_file_trim(
 			" fallocate(FALLOC_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE) ",
 			FALSE, __FILE__, __LINE__);
 
-		slot->write_size = 0;
+		if (slot->write_size) {
+			*slot->write_size = 0;
+		}
 
 		return (FALSE);
 	} else {
-		slot->write_size = len;
+		if (slot->write_size) {
+			*slot->write_size = len;
+		}
 	}
 #else
 	ut_print_timestamp(stderr);
@@ -6203,7 +6214,7 @@ os_file_trim(
 		"  InnoDB: [Warning] fallocate not supported on this installation."
 		"  InnoDB: Disabling fallocate for now.");
 	os_fallocate_failed = TRUE;
-	slot->write_size = 0;
+	slot->write_size = NULL;
 
 #endif /* HAVE_FALLOCATE ... */
 
@@ -6229,10 +6240,14 @@ os_file_trim(
 			" DeviceIOControl(FSCTL_FILE_LEVEL_TRIM) ",
 			FALSE, __FILE__, __LINE__);
 
-		slot->write_size = 0;
+		if (slot->write_size) {
+			slot->write_size = 0;
+		}
 		return (FALSE);
 	} else {
-		slot->write_size = len;
+		if (slot->write_size) {
+			slot->write_size = len;
+		}
 	}
 #endif