os0file.c Use O_SYNC if possible, second choice fdatasync, and last choice fsync

configure.in	Check if fdatasync exists
manual.texi	Updated defragmenting doc
manual.texi	Corrected fillfactor contract threashold for a page to 1/2

Docs/manual.texi

@@ -25283,7 +25283,7 @@ of the index records.
 If index records are inserted in a sequential (ascending or descending)
 order, the resulting index pages will be about 15/16 full.
 If records are inserted in a random order, then the pages will be
-1/2 - 15/16 full. If the fillfactor of an index page drops below 1/4,
+1/2 - 15/16 full. If the fillfactor of an index page drops below 1/2,
 InnoDB will try to contract the index tree to free the page.
 
 @subsubsection Insert buffering
@@ -25440,13 +25440,20 @@ consistent read.
 
 If there are random insertions or deletions
 in the indexes of a table, the indexes
-may become fragmented. By this we mean that the physical ordering
+may become fragmented. By fragmentation we mean that the physical ordering
 of the index pages on the disk is not close to the alphabetical ordering
-of the records on the pages. It can speed up index scans if you
+of the records on the pages, or that there are many unused pages in the
+64-page blocks which were allocated to the index.
+
+It can speed up index scans if you
 periodically use @code{mysqldump} to dump the table to
 a text file, drop the table, and reload it from the dump.
+Another way to do the defragmenting is to @code{ALTER} the table type to
+@code{MyISAM} and back to @code{InnoDB} again.
+Note that a @code{MyISAM} table must fit in a single file
+on your operating system.
 
-Note that if the insertions to and index are always ascending and
+If the insertions to and index are always ascending and
 records are deleted only from the end, then the the file space management
 algorithm of InnoDB guarantees that fragmentation in the index will
 not occur.

innobase/configure.in

@@ -37,6 +37,7 @@ AC_PROG_INSTALL
 AC_CHECK_HEADERS(aio.h sched.h)
 AC_CHECK_SIZEOF(int, 4)
 AC_CHECK_FUNCS(sched_yield)
+AC_CHECK_FUNCS(fdatasync)
 #AC_C_INLINE  Already checked in MySQL
 AC_C_BIGENDIAN
 

innobase/os/os0file.c

@@ -299,6 +299,13 @@ try_again:
 
 	UT_NOT_USED(purpose);
 
+	/* On Linux opening a file in the O_SYNC mode seems to be much
+        more efficient than calling an explicit fsync or fdatasync after
+        each write */
+
+#ifdef O_SYNC
+	create_flag = create_flag | O_SYNC;
+#endif
 	if (create_mode == OS_FILE_CREATE) {
 	        file = open(name, create_flag, S_IRUSR | S_IWUSR | S_IRGRP
 			                     | S_IWGRP | S_IROTH | S_IWOTH);
@@ -492,8 +499,18 @@ os_file_flush(
 #else
 	int	ret;
 	
-	ret = fsync(file);
+#ifdef O_SYNC
+	/* We open all files with the O_SYNC option, which means there
+        should be no need for fsync or fdatasync. In practice such a need
+	may be because on a Linux Xeon computer "donna" the OS seemed to be
+	fooled to believe that 500 disk writes/second are possible. */
 
+	ret = 0;
+#elif defined(HAVE_FDATASYNC)
+	ret = fdatasync(file);
+#else
+	ret = fsync(file);
+#endif
 	if (ret == 0) {
 		return(TRUE);
 	}