Merge branch 'ps/upload-pack-buffer-more-writes'

Reduce system overhead "git upload-pack" spends on relaying "git
pack-objects" output to the "git fetch" running on the other end of
the connection.

* ps/upload-pack-buffer-more-writes:
  builtin/pack-objects: reduce lock contention when writing packfile data
  csum-file: drop `hashfd_throughput()`
  csum-file: introduce `hashfd_ext()`
  sideband: use writev(3p) to send pktlines
  wrapper: introduce writev(3p) wrappers
  compat/posix: introduce writev(3p) wrapper
  upload-pack: reduce lock contention when writing packfile data
  upload-pack: prefer flushing data over sending keepalive
  upload-pack: adapt keepalives based on buffering
  upload-pack: fix debug statement when flushing packfile data

Author: gitster

Makefile

@@ -2027,6 +2027,10 @@ ifdef NO_PREAD
 	COMPAT_CFLAGS += -DNO_PREAD
 	COMPAT_OBJS += compat/pread.o
 endif
+ifdef NO_WRITEV
+	COMPAT_CFLAGS += -DNO_WRITEV
+	COMPAT_OBJS += compat/writev.o
+endif
 ifdef NO_FAST_WORKING_DIRECTORY
 	BASIC_CFLAGS += -DNO_FAST_WORKING_DIRECTORY
 endif

builtin/pack-objects.c

@@ -41,6 +41,7 @@
 #include "promisor-remote.h"
 #include "pack-mtimes.h"
 #include "parse-options.h"
+#include "pkt-line.h"
 #include "blob.h"
 #include "tree.h"
 #include "path-walk.h"
@@ -1330,11 +1331,25 @@ static void write_pack_file(void)
 		unsigned char hash[GIT_MAX_RAWSZ];
 		char *pack_tmp_name = NULL;
 
-		if (pack_to_stdout)
-			f = hashfd_throughput(the_repository->hash_algo, 1,
-					      "<stdout>", progress_state);
-		else
+		if (pack_to_stdout) {
+			/*
+			 * This command is most often invoked via
+			 * git-upload-pack(1), which will typically chunk data
+			 * into pktlines. As such, we use the maximum data
+			 * length of them as buffer length.
+			 *
+			 * Note that we need to subtract one though to
+			 * accomodate for the sideband byte.
+			 */
+			struct hashfd_options opts = {
+				.progress = progress_state,
+				.buffer_len = LARGE_PACKET_DATA_MAX - 1,
+			};
+			f = hashfd_ext(the_repository->hash_algo, 1,
+				       "<stdout>", &opts);
+		} else {
 			f = create_tmp_packfile(the_repository, &pack_tmp_name);
+		}
 
 		offset = write_pack_header(f, nr_remaining);
 

compat/posix.h

@@ -137,6 +137,9 @@
 #include <sys/socket.h>
 #include <sys/ioctl.h>
 #include <sys/statvfs.h>
+#ifndef NO_WRITEV
+#include <sys/uio.h>
+#endif
 #include <termios.h>
 #ifndef NO_SYS_SELECT_H
 #include <sys/select.h>
@@ -323,6 +326,17 @@ int git_lstat(const char *, struct stat *);
 ssize_t git_pread(int fd, void *buf, size_t count, off_t offset);
 #endif
 
+#ifdef NO_WRITEV
+#define writev git_writev
+#define iovec git_iovec
+struct git_iovec {
+	void *iov_base;
+	size_t iov_len;
+};
+
+ssize_t git_writev(int fd, const struct iovec *iov, int iovcnt);
+#endif
+
 #ifdef NO_SETENV
 #define setenv gitsetenv
 int gitsetenv(const char *, const char *, int);

compat/writev.c

@@ -0,0 +1,44 @@
+#include "../git-compat-util.h"
+#include "../wrapper.h"
+
+ssize_t git_writev(int fd, const struct iovec *iov, int iovcnt)
+{
+	size_t total_written = 0;
+	size_t sum = 0;
+
+	/*
+	 * According to writev(3p), the syscall shall error with EINVAL in case
+	 * the sum of `iov_len` overflows `ssize_t`.
+	 */
+	 for (int i = 0; i < iovcnt; i++) {
+		if (iov[i].iov_len > maximum_signed_value_of_type(ssize_t) ||
+		    iov[i].iov_len + sum > maximum_signed_value_of_type(ssize_t)) {
+			errno = EINVAL;
+			return -1;
+		}
+
+		sum += iov[i].iov_len;
+	}
+
+	for (int i = 0; i < iovcnt; i++) {
+		const char *bytes = iov[i].iov_base;
+		size_t iovec_written = 0;
+
+		while (iovec_written < iov[i].iov_len) {
+			ssize_t bytes_written = xwrite(fd, bytes + iovec_written,
+						       iov[i].iov_len - iovec_written);
+			if (bytes_written < 0) {
+				if (total_written)
+					goto out;
+				return bytes_written;
+			}
+			if (!bytes_written)
+				goto out;
+			iovec_written += bytes_written;
+			total_written += bytes_written;
+		}
+	}
+
+out:
+	return (ssize_t) total_written;
+}

config.mak.uname

@@ -459,6 +459,7 @@ ifeq ($(uname_S),Windows)
 	SANE_TOOL_PATH ?= $(msvc_bin_dir_msys)
 	HAVE_ALLOCA_H = YesPlease
 	NO_PREAD = YesPlease
+	NO_WRITEV = YesPlease
 	NEEDS_CRYPTO_WITH_SSL = YesPlease
 	NO_LIBGEN_H = YesPlease
 	NO_POLL = YesPlease
@@ -674,6 +675,7 @@ ifeq ($(uname_S),MINGW)
 	pathsep = ;
 	HAVE_ALLOCA_H = YesPlease
 	NO_PREAD = YesPlease
+	NO_WRITEV = YesPlease
 	NEEDS_CRYPTO_WITH_SSL = YesPlease
 	NO_LIBGEN_H = YesPlease
 	NO_POLL = YesPlease

csum-file.c

@@ -161,26 +161,25 @@ struct hashfile *hashfd_check(const struct git_hash_algo *algop,
 	return f;
 }
 
-static struct hashfile *hashfd_internal(const struct git_hash_algo *algop,
-					int fd, const char *name,
-					struct progress *tp,
-					size_t buffer_len)
+struct hashfile *hashfd_ext(const struct git_hash_algo *algop,
+			    int fd, const char *name,
+			    const struct hashfd_options *opts)
 {
 	struct hashfile *f = xmalloc(sizeof(*f));
 	f->fd = fd;
 	f->check_fd = -1;
 	f->offset = 0;
 	f->total = 0;
-	f->tp = tp;
+	f->tp = opts->progress;
 	f->name = name;
 	f->do_crc = 0;
 	f->skip_hash = 0;
 
 	f->algop = unsafe_hash_algo(algop);
 	f->algop->init_fn(&f->ctx);
 
-	f->buffer_len = buffer_len;
-	f->buffer = xmalloc(buffer_len);
+	f->buffer_len = opts->buffer_len ? opts->buffer_len : 128 * 1024;
+	f->buffer = xmalloc(f->buffer_len);
 	f->check_buffer = NULL;
 
 	return f;
@@ -194,19 +193,8 @@ struct hashfile *hashfd(const struct git_hash_algo *algop,
 	 * measure the rate of data passing through this hashfile,
 	 * use a larger buffer size to reduce fsync() calls.
 	 */
-	return hashfd_internal(algop, fd, name, NULL, 128 * 1024);
-}
-
-struct hashfile *hashfd_throughput(const struct git_hash_algo *algop,
-				   int fd, const char *name, struct progress *tp)
-{
-	/*
-	 * Since we are expecting to report progress of the
-	 * write into this hashfile, use a smaller buffer
-	 * size so the progress indicators arrive at a more
-	 * frequent rate.
-	 */
-	return hashfd_internal(algop, fd, name, tp, 8 * 1024);
+	struct hashfd_options opts = { 0 };
+	return hashfd_ext(algop, fd, name, &opts);
 }
 
 void hashfile_checkpoint_init(struct hashfile *f,

csum-file.h

@@ -45,12 +45,24 @@ int hashfile_truncate(struct hashfile *, struct hashfile_checkpoint *);
 #define CSUM_FSYNC		2
 #define CSUM_HASH_IN_STREAM	4
 
+struct hashfd_options {
+	/*
+	 * Throughput progress that counts the number of bytes that have been
+	 * hashed.
+	 */
+	struct progress *progress;
+
+	/* The length of the buffer that shall be used read read data. */
+	size_t buffer_len;
+};
+
+struct hashfile *hashfd_ext(const struct git_hash_algo *algop,
+			    int fd, const char *name,
+			    const struct hashfd_options *opts);
 struct hashfile *hashfd(const struct git_hash_algo *algop,
 			int fd, const char *name);
 struct hashfile *hashfd_check(const struct git_hash_algo *algop,
 			      const char *name);
-struct hashfile *hashfd_throughput(const struct git_hash_algo *algop,
-				   int fd, const char *name, struct progress *tp);
 
 /*
  * Free the hashfile without flushing its contents to disk. This only

meson.build

@@ -1426,6 +1426,7 @@ checkfuncs = {
   'initgroups' : [],
   'strtoumax' : ['strtoumax.c', 'strtoimax.c'],
   'pread' : ['pread.c'],
+  'writev' : ['writev.c'],
 }
 
 if host_machine.system() == 'windows'

sideband.c

@@ -264,6 +264,7 @@ void send_sideband(int fd, int band, const char *data, ssize_t sz, int packet_ma
 	const char *p = data;
 
 	while (sz) {
+		struct iovec iov[2];
 		unsigned n;
 		char hdr[5];
 
@@ -273,12 +274,19 @@ void send_sideband(int fd, int band, const char *data, ssize_t sz, int packet_ma
 		if (0 <= band) {
 			xsnprintf(hdr, sizeof(hdr), "%04x", n + 5);
 			hdr[4] = band;
-			write_or_die(fd, hdr, 5);
+			iov[0].iov_base = hdr;
+			iov[0].iov_len = 5;
 		} else {
 			xsnprintf(hdr, sizeof(hdr), "%04x", n + 4);
-			write_or_die(fd, hdr, 4);
+			iov[0].iov_base = hdr;
+			iov[0].iov_len = 4;
 		}
-		write_or_die(fd, p, n);
+
+		iov[1].iov_base = (void *) p;
+		iov[1].iov_len = n;
+
+		writev_or_die(fd, iov, ARRAY_SIZE(iov));
+
 		p += n;
 		sz -= n;
 	}