Apply patch for test_fsync to add tests for O_DIRECT.

Adjusted patch by Josh Berkus

src/tools/fsync/Makefile

@@ -16,9 +16,9 @@ override CPPFLAGS := -I$(libpq_srcdir) $(CPPFLAGS)
 
 OBJS= test_fsync.o
 
-all: test_fsync
+all: submake-libpq submake-libpgport test_fsync
 
-test_fsync: test_fsync.o | submake-libpq submake-libpgport
+test_fsync: test_fsync.o $(libpq_builddir)/libpq.a
 	$(CC) $(CFLAGS) test_fsync.o $(libpq_pgport) $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o $@$(X)
 
 clean distclean maintainer-clean:

src/tools/fsync/README

@@ -1,11 +1,20 @@
-src/tools/fsync/README
+test_fsync
-
+==========
-fsync
-=====
 
 This program tests fsync.  The tests are described as part of the program output.
 
 	Usage:	test_fsync [-f filename] [loops]
+	
+test_fsync is intended to give you a reasonable idea of what the fastest
+fsync_method is on your specific system, as well as supplying diagnostic
+information in the event of an identified I/O problem.  However,
+differences shown by test_fsync might not make any difference in real
+database throughput, especially since many database servers are not
+speed-limited by their transaction logs.
+
+The output filename defaults to test_fsync.out in the current directory.
+test_fsync should be run in the same filesystem as your transaction log
+directory (pg_xlog).
+
+Loops default to 2000.  Increase this to get more accurate measurements.
 
-Loops defaults to 5000.  The default output file is /var/tmp/test_fsync.out.
-Consider that /tmp or /var/tmp might be memory-based file systems.

src/tools/fsync/test_fsync.c

@@ -3,7 +3,7 @@
  *
  *
  *	test_fsync.c
- *		test various fsync() methods
+ *		tests all supported fsync() methods
  */
 
 #include "postgres.h"
@@ -22,19 +22,18 @@
 #include <unistd.h>
 #include <string.h>
 
-
+/* 
-#ifdef WIN32
+ * put the temp files in the local directory
+ * unless the user specifies otherwise 
+ */
 #define FSYNC_FILENAME	"./test_fsync.out"
-#else
-/* /tmp might be a memory file system */
-#define FSYNC_FILENAME	"/var/tmp/test_fsync.out"
-#endif
 
 #define WRITE_SIZE	(8 * 1024)	/* 8k */
 
 #define LABEL_FORMAT	"\t%-30s"
 
-int			loops = 10000;
+
+int			loops = 2000;
 
 void		die(char *str);
 void		print_elapse(struct timeval start_t, struct timeval stop_t);
@@ -42,14 +41,14 @@ void		print_elapse(struct timeval start_t, struct timeval stop_t);
 int
 main(int argc, char *argv[])
 {
-	struct timeval start_t;
+	struct timeval start_t, stop_t;
-	struct timeval stop_t;
+	int			tmpfile, i;
-	int			tmpfile,
-				i;
 	char	   *full_buf = (char *) malloc(XLOG_SEG_SIZE),
-			   *buf;
+			   *buf, *filename = FSYNC_FILENAME;
-	char	   *filename = FSYNC_FILENAME;
 
+	/* 
+	 * arguments: loops and filename (optional) 
+	 */
 	if (argc > 2 && strcmp(argv[1], "-f") == 0)
 	{
 		filename = argv[2];
@@ -57,17 +56,22 @@ main(int argc, char *argv[])
 		argc -= 2;
 	}
 
-	if (argc > 1)
+	if (argc > 1) 
 		loops = atoi(argv[1]);
 
 	for (i = 0; i < XLOG_SEG_SIZE; i++)
 		full_buf[i] = random();
 
+	/* 
+	 * test if we can open the target file 
+	 */
 	if ((tmpfile = open(filename, O_RDWR | O_CREAT, S_IRUSR | S_IWUSR)) == -1)
 		die("Cannot open output file.");
 	if (write(tmpfile, full_buf, XLOG_SEG_SIZE) != XLOG_SEG_SIZE)
 		die("write failed");
-	/* fsync now so later fsync's don't have to do it */
+	/*
+	 * fsync now so that dirty buffers don't skew later tests
+	 */
 	if (fsync(tmpfile) != 0)
 		die("fsync failed");
 	close(tmpfile);
@@ -77,7 +81,7 @@ main(int argc, char *argv[])
 	printf("Loops = %d\n\n", loops);
 
 	/*
-	 * Simple write
+	 * Test a simple write without fsync
 	 */
 	printf("Simple write:\n");
 	printf(LABEL_FORMAT, "8k write");
@@ -95,10 +99,13 @@ main(int argc, char *argv[])
 	print_elapse(start_t, stop_t);
 
 	/*
-	 * Compare file sync methods with one 8k write
+	 * Test all fsync methods using single 8k writes
 	 */
 	printf("\nCompare file sync methods using one write:\n");
 
+	/*
+	 * Test open_datasync if available
+	 */
 #ifdef OPEN_DATASYNC_FLAG
 	printf(LABEL_FORMAT, "open_datasync 8k write");
 	fflush(stdout);
@@ -115,10 +122,40 @@ main(int argc, char *argv[])
 	gettimeofday(&stop_t, NULL);
 	close(tmpfile);
 	print_elapse(start_t, stop_t);
+
+	/*
+	 * If O_DIRECT is enabled, test that with open_datasync
+	 */
+#if PG_O_DIRECT != 0
+	fflush(stdout);
+	if ((tmpfile = open(filename, O_RDWR | O_DSYNC | PG_O_DIRECT, 0)) == -1)
+		printf("\t(unavailable: o_direct on this filesystem)\n");
+	else
+	{
+		printf(LABEL_FORMAT, "open_datasync 8k direct I/O write");
+		gettimeofday(&start_t, NULL);
+		for (i = 0; i < loops; i++)
+		{
+			if (write(tmpfile, buf, WRITE_SIZE) != WRITE_SIZE)
+				die("write failed");
+			if (lseek(tmpfile, 0, SEEK_SET) == -1)
+				die("seek failed");
+		}
+		gettimeofday(&stop_t, NULL);
+		close(tmpfile);
+		print_elapse(start_t, stop_t);
+	}
+#else
+		printf("\t(unavailable: o_direct)\n");
+#endif
+
 #else
 	printf("\t(unavailable: open_datasync)\n");
 #endif
 
+/*
+ * Test open_sync if available
+ */
 #ifdef OPEN_SYNC_FLAG
 	printf(LABEL_FORMAT, "open_sync 8k write");
 	fflush(stdout);
@@ -135,10 +172,40 @@ main(int argc, char *argv[])
 	gettimeofday(&stop_t, NULL);
 	close(tmpfile);
 	print_elapse(start_t, stop_t);
+
+	/*
+	 * If O_DIRECT is enabled, test that with open_sync
+	 */
+#if PG_O_DIRECT != 0
+	printf(LABEL_FORMAT, "open_sync 8k direct I/O write");
+	fflush(stdout);
+	if ((tmpfile = open(filename, O_RDWR | OPEN_SYNC_FLAG | PG_O_DIRECT, 0)) == -1)
+		printf("\t(unavailable: o_direct on this filesystem)\n");
+	else
+	{
+		gettimeofday(&start_t, NULL);
+		for (i = 0; i < loops; i++)
+		{
+			if (write(tmpfile, buf, WRITE_SIZE) != WRITE_SIZE)
+				die("write failed");
+			if (lseek(tmpfile, 0, SEEK_SET) == -1)
+				die("seek failed");
+		}
+		gettimeofday(&stop_t, NULL);
+		close(tmpfile);
+		print_elapse(start_t, stop_t);
+	}
+#else
+	printf("\t(unavailable: o_direct)\n");
+#endif
+
 #else
 	printf("\t(unavailable: open_sync)\n");
 #endif
 
+/*
+ * Test fdatasync if available
+ */
 #ifdef HAVE_FDATASYNC
 	printf(LABEL_FORMAT, "8k write, fdatasync");
 	fflush(stdout);
@@ -160,6 +227,9 @@ main(int argc, char *argv[])
 	printf("\t(unavailable: fdatasync)\n");
 #endif
 
+/*
+ * Test fsync
+ */
 	printf(LABEL_FORMAT, "8k write, fsync");
 	fflush(stdout);
 	if ((tmpfile = open(filename, O_RDWR, 0)) == -1)
@@ -177,14 +247,43 @@ main(int argc, char *argv[])
 	gettimeofday(&stop_t, NULL);
 	close(tmpfile);
 	print_elapse(start_t, stop_t);
+	
+/*
+ * If fsync_writethrough is available, test as well
+ */	
+#ifdef HAVE_FSYNC_WRITETHROUGH
+	printf(LABEL_FORMAT, "8k write, fsync_writethrough");
+	fflush(stdout);
+	if ((tmpfile = open(filename, O_RDWR, 0)) == -1)
+		die("Cannot open output file.");
+	gettimeofday(&start_t, NULL);
+	for (i = 0; i < loops; i++)
+	{
+		if (write(tmpfile, buf, WRITE_SIZE) != WRITE_SIZE)
+			die("write failed");
+		if (fcntl(tmpfile, F_FULLFSYNC ) != 0)
+			die("fsync failed");
+		if (lseek(tmpfile, 0, SEEK_SET) == -1)
+			die("seek failed");
+	}
+	gettimeofday(&stop_t, NULL);
+	close(tmpfile);
+	print_elapse(start_t, stop_t);
+#else
+	printf("\t(unavailable: fsync_writethrough)\n");
+#endif
 
 	/*
-	 * Compare file sync methods with two 8k write
+	 * Compare some of the file sync methods with 
+	 * two 8k writes to see if timing is different
 	 */
 	printf("\nCompare file sync methods using two writes:\n");
 
+/*
+ * Test open_datasync with and without o_direct
+ */
 #ifdef OPEN_DATASYNC_FLAG
-	printf(LABEL_FORMAT, "2 open_datasync 8k writes");
+ 	printf(LABEL_FORMAT, "2 open_datasync 8k writes");
 	fflush(stdout);
 	if ((tmpfile = open(filename, O_RDWR | O_DSYNC, 0)) == -1)
 		die("Cannot open output file.");
@@ -201,10 +300,36 @@ main(int argc, char *argv[])
 	gettimeofday(&stop_t, NULL);
 	close(tmpfile);
 	print_elapse(start_t, stop_t);
+	
+#if PG_O_DIRECT != 0
+	printf(LABEL_FORMAT, "2 open_datasync direct I/O 8k writes");
+	fflush(stdout);
+	if ((tmpfile = open(filename, O_RDWR | O_DSYNC | PG_O_DIRECT, 0)) == -1)
+		die("Cannot open output file.");
+	gettimeofday(&start_t, NULL);
+	for (i = 0; i < loops; i++)
+	{
+		if (write(tmpfile, buf, WRITE_SIZE) != WRITE_SIZE)
+			die("write failed");
+		if (write(tmpfile, buf, WRITE_SIZE) != WRITE_SIZE)
+			die("write failed");
+		if (lseek(tmpfile, 0, SEEK_SET) == -1)
+			die("seek failed");
+	}
+	gettimeofday(&stop_t, NULL);
+	close(tmpfile);
+	print_elapse(start_t, stop_t);
+#else
+		printf("\t(unavailable: o_direct)\n");
+#endif
+
 #else
 	printf("\t(unavailable: open_datasync)\n");
 #endif
 
+/*
+ * Test open_sync with and without o_direct
+ */
 #ifdef OPEN_SYNC_FLAG
 	printf(LABEL_FORMAT, "2 open_sync 8k writes");
 	fflush(stdout);
@@ -223,8 +348,36 @@ main(int argc, char *argv[])
 	gettimeofday(&stop_t, NULL);
 	close(tmpfile);
 	print_elapse(start_t, stop_t);
+	
+#if PG_O_DIRECT != 0
+	printf(LABEL_FORMAT, "2 open_sync direct I/O 8k writes");
+	fflush(stdout);
+	if ((tmpfile = open(filename, O_RDWR | OPEN_SYNC_FLAG | PG_O_DIRECT, 0)) == -1)
+		die("Cannot open output file.");
+	gettimeofday(&start_t, NULL);
+	for (i = 0; i < loops; i++)
+	{
+		if (write(tmpfile, buf, WRITE_SIZE) != WRITE_SIZE)
+			die("write failed");
+		if (write(tmpfile, buf, WRITE_SIZE) != WRITE_SIZE)
+			die("write failed");
+		if (lseek(tmpfile, 0, SEEK_SET) == -1)
+			die("seek failed");
+	}
+	gettimeofday(&stop_t, NULL);
+	close(tmpfile);
+	print_elapse(start_t, stop_t);
+#else
+	printf("\t(unavailable: o_direct)\n");
 #endif
 
+#else
+	printf("\t(unavailable: open_sync)\n");
+#endif
+
+/*
+ *	Test fdatasync
+ */
 #ifdef HAVE_FDATASYNC
 	printf(LABEL_FORMAT, "8k write, 8k write, fdatasync");
 	fflush(stdout);
@@ -248,6 +401,9 @@ main(int argc, char *argv[])
 	printf("\t(unavailable: fdatasync)\n");
 #endif
 
+/*
+ * Test basic fsync
+ */
 	printf(LABEL_FORMAT, "8k write, 8k write, fsync");
 	fflush(stdout);
 	if ((tmpfile = open(filename, O_RDWR, 0)) == -1)
@@ -267,12 +423,44 @@ main(int argc, char *argv[])
 	gettimeofday(&stop_t, NULL);
 	close(tmpfile);
 	print_elapse(start_t, stop_t);
+	
+/*
+ * Test fsync_writethrough if available
+ */	
+#ifdef HAVE_FSYNC_WRITETHROUGH
+	printf(LABEL_FORMAT, "8k write, 8k write, fsync_writethrough");
+	fflush(stdout);
+	if ((tmpfile = open(filename, O_RDWR, 0)) == -1)
+		die("Cannot open output file.");
+	gettimeofday(&start_t, NULL);
+	for (i = 0; i < loops; i++)
+	{
+		if (write(tmpfile, buf, WRITE_SIZE) != WRITE_SIZE)
+			die("write failed");
+		if (write(tmpfile, buf, WRITE_SIZE) != WRITE_SIZE)
+			die("write failed");
+		if (fcntl(tmpfile, F_FULLFSYNC) != 0)
+			die("fsync failed");
+		if (lseek(tmpfile, 0, SEEK_SET) == -1)
+			die("seek failed");
+	}
+	gettimeofday(&stop_t, NULL);
+	close(tmpfile);
+	print_elapse(start_t, stop_t);
+#else
+	printf("\t(unavailable: fsync_writethrough)\n");
+#endif
 
 	/*
 	 * Compare 1 to 2 writes
 	 */
 	printf("\nCompare open_sync with different sizes:\n");
+	printf("(This is designed to compare the cost of one large\n");
+	printf("sync'ed write and two smaller sync'ed writes.)\n");
 
+/*
+ * Test open_sync with different size files
+ */
 #ifdef OPEN_SYNC_FLAG
 	printf(LABEL_FORMAT, "open_sync 16k write");
 	fflush(stdout);
@@ -312,12 +500,20 @@ main(int argc, char *argv[])
 #endif
 
 	/*
-	 * Fsync another file descriptor?
+	 * Test whether fsync can sync data written on a different
+	 * descriptor for the same file.  This checks the efficiency
+	 * of multi-process fsyncs against the same file.
+	 * Possibly this should be done with writethrough on platforms
+	 * which support it.
 	 */
 	printf("\nTest if fsync on non-write file descriptor is honored:\n");
 	printf("(If the times are similar, fsync() can sync data written\n");
 	printf("on a different descriptor.)\n");
 
+	/* 
+	 * first write, fsync and close, which is the 
+	 * normal behavior without multiple descriptors
+	 */
 	printf(LABEL_FORMAT, "8k write, fsync, close");
 	fflush(stdout);
 	gettimeofday(&start_t, NULL);
@@ -330,14 +526,22 @@ main(int argc, char *argv[])
 		if (fsync(tmpfile) != 0)
 			die("fsync failed");
 		close(tmpfile);
+		/*
+		 * open and close the file again to be consistent
+		 * with the following test
+		 */
 		if ((tmpfile = open(filename, O_RDWR, 0)) == -1)
 			die("Cannot open output file.");
-		/* do nothing but the open/close the tests are consistent. */
 		close(tmpfile);
 	}
 	gettimeofday(&stop_t, NULL);
 	print_elapse(start_t, stop_t);
 
+	/*
+	 * Now open, write, close, open again and fsync
+	 * This simulates processes fsyncing each other's
+	 * writes.
+	 */
  	printf(LABEL_FORMAT, "8k write, close, fsync");
  	fflush(stdout);
 	gettimeofday(&start_t, NULL);
@@ -358,18 +562,22 @@ main(int argc, char *argv[])
 	gettimeofday(&stop_t, NULL);
 	print_elapse(start_t, stop_t);
 
-	/* cleanup */
+	/* 
+	 * cleanup 
+	 */
 	free(full_buf);
 	unlink(filename);
 
 	return 0;
 }
 
+/* 
+ * print out the writes per second for tests
+ */
 void
 print_elapse(struct timeval start_t, struct timeval stop_t)
 {
 	double		total_time = (stop_t.tv_sec - start_t.tv_sec) +
-	/* usec subtraction might be negative, e.g. 5.4 - 4.8 */
 	(stop_t.tv_usec - start_t.tv_usec) * 0.000001;
 	double		per_second = loops / total_time;