[PATCH 0/3] read-cache: use stack ondisk struct when writing index

[PATCH 0/3] read-cache: use stack ondisk struct when writing index

[Kevin Willford]

When writing out the index, the ondisk struct was being allocated and freed
within the loop of cache entries.  A better way would be to use a ondisk
struct on the stack and reuse it avoiding the alloc and free calls.

A test has been added to measure the performance of writing the index
(p0007-write-cache).  Running this test on smaller repos showed no
degradation in performance.  On larger repos there was ~3-7% improvement.

0007.2: write_locked_index 10 times (1000001 files)   5.98(0.00+0.04)   5.75(0.01+0.04) -3.8%
0007.2: write_locked_index 10 times (1000001 files)   6.20(0.00+0.06)   5.86(0.01+0.03) -5.5%
0007.2: write_locked_index 3 times (4394531 files)   10.29(0.04+0.03)   9.75(0.04+0.01) -5.2%
0007.2: write_locked_index 3 times (4394531 files)   10.52(0.00+0.04)   9.79(0.03+0.03) -6.9%

Kevin Willford (3):
  perf: add test for writing the index
  read-cache: fix memory leak in do_write_index
  read-cache: avoid allocating every ondisk entry when writing

 Makefile                    |  1 +
 read-cache.c                | 62 +++++++++++++++++++++++++--------------------
 t/helper/test-write-cache.c | 23 +++++++++++++++++
 t/perf/p0007-write-cache.sh | 29 +++++++++++++++++++++
 4 files changed, 87 insertions(+), 28 deletions(-)
 create mode 100644 t/helper/test-write-cache.c
 create mode 100755 t/perf/p0007-write-cache.sh

-- 
2.14.1.205.g2812f3410d

[PATCH 1/3] perf: add test for writing the index

[Kevin Willford]

A performance test for writing the index to be able to
determine if changes to allocating ondisk structure help.

Signed-off-by: Kevin Willford <kewillf@microsoft.com>
---
 Makefile                    |  1 +
 t/helper/test-write-cache.c | 23 +++++++++++++++++++++++
 t/perf/p0007-write-cache.sh | 29 +++++++++++++++++++++++++++++
 3 files changed, 53 insertions(+)
 create mode 100644 t/helper/test-write-cache.c
 create mode 100755 t/perf/p0007-write-cache.sh

diff --git a/Makefile b/Makefile
index 86ec29202b..c6b061086f 100644
--- a/Makefile
+++ b/Makefile
@@ -655,6 +655,7 @@ TEST_PROGRAMS_NEED_X += test-parse-options
 TEST_PROGRAMS_NEED_X += test-path-utils
 TEST_PROGRAMS_NEED_X += test-prio-queue
 TEST_PROGRAMS_NEED_X += test-read-cache
+TEST_PROGRAMS_NEED_X += test-write-cache
 TEST_PROGRAMS_NEED_X += test-ref-store
 TEST_PROGRAMS_NEED_X += test-regex
 TEST_PROGRAMS_NEED_X += test-revision-walking
diff --git a/t/helper/test-write-cache.c b/t/helper/test-write-cache.c
new file mode 100644
index 0000000000..b7ee039669
--- /dev/null
+++ b/t/helper/test-write-cache.c
@@ -0,0 +1,23 @@
+#include "cache.h"
+#include "lockfile.h"
+
+static struct lock_file index_lock;
+
+int cmd_main(int argc, const char **argv)
+{
+	int i, cnt = 1, lockfd;
+	if (argc == 2)
+		cnt = strtol(argv[1], NULL, 0);
+	setup_git_directory();
+	read_cache();
+	for (i = 0; i < cnt; i++) {
+		lockfd = hold_locked_index(&index_lock, LOCK_DIE_ON_ERROR);
+		if (0 <= lockfd) {
+			write_locked_index(&the_index, &index_lock, COMMIT_LOCK);
+		} else {
+			rollback_lock_file(&index_lock);
+		}
+	}
+
+	return 0;
+}
diff --git a/t/perf/p0007-write-cache.sh b/t/perf/p0007-write-cache.sh
new file mode 100755
index 0000000000..261fe92fd9
--- /dev/null
+++ b/t/perf/p0007-write-cache.sh
@@ -0,0 +1,29 @@
+#!/bin/sh
+
+test_description="Tests performance of writing the index"
+
+. ./perf-lib.sh
+
+test_perf_default_repo
+
+test_expect_success "setup repo" '
+	if git rev-parse --verify refs/heads/p0006-ballast^{commit}
+	then
+		echo Assuming synthetic repo from many-files.sh
+		git config --local core.sparsecheckout 1
+		cat >.git/info/sparse-checkout <<-EOF
+		/*
+		!ballast/*
+		EOF
+	else
+		echo Assuming non-synthetic repo...
+	fi &&
+	nr_files=$(git ls-files | wc -l)
+'
+
+count=3
+test_perf "write_locked_index $count times ($nr_files files)" "
+	test-write-cache $count
+"
+
+test_done
-- 
2.14.1.205.g2812f3410d

[PATCH 2/3] read-cache: fix memory leak in do_write_index

[Kevin Willford]

The previous_name_buf was never getting released when there
was an error in ce_write_entry or allow was false and execution
was returned to the caller.

Signed-off-by: Kevin Willford <kewillf@microsoft.com>
---
 read-cache.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

diff --git a/read-cache.c b/read-cache.c
index acfb028f48..47220cc30d 100644
--- a/read-cache.c
+++ b/read-cache.c
@@ -2192,7 +2192,7 @@ static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
 	int newfd = tempfile->fd;
 	git_SHA_CTX c;
 	struct cache_header hdr;
-	int i, err, removed, extended, hdr_version;
+	int i, err = 0, removed, extended, hdr_version;
 	struct cache_entry **cache = istate->cache;
 	int entries = istate->cache_nr;
 	struct stat st;
@@ -2247,15 +2247,21 @@ static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
 			if (allow)
 				warning(msg, ce->name);
 			else
-				return error(msg, ce->name);
+				err = error(msg, ce->name);
 
 			drop_cache_tree = 1;
 		}
 		if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
-			return -1;
+			err = -1;
+
+		if (err)
+			break;
 	}
 	strbuf_release(&previous_name_buf);
 
+	if (err)
+		return err;
+
 	/* Write extension data here */
 	if (!strip_extensions && istate->split_index) {
 		struct strbuf sb = STRBUF_INIT;
-- 
2.14.1.205.g2812f3410d

[PATCH 3/3] read-cache: avoid allocating every ondisk entry when writing

[Kevin Willford]

When writing the index for each entry an ondisk struct will be
allocated and freed in ce_write_entry.  We can do better by
using a ondisk struct on the stack for each entry.

This is accomplished by using a stack ondisk_cache_entry_extended
outside looping through the entries in do_write_index.  Only the
fixed fields of this struct are used when writing and depending on
whether it is extended or not the flags2 field will be written.
The name field is not used and instead the cache_entry name field
is used directly when writing out the name.  Because ce_write is
using a buffer and memcpy to fill the buffer before flushing to disk,
we don't have to worry about doing multiple ce_write calls.

Running the p0007-write-cache.sh tests would save anywhere
between 3-7% when the index had over a million entries with no
performance degradation on small repos.

Signed-off-by: Kevin Willford <kewillf@microsoft.com>
---
 read-cache.c | 50 +++++++++++++++++++++++++-------------------------
 1 file changed, 25 insertions(+), 25 deletions(-)

diff --git a/read-cache.c b/read-cache.c
index 47220cc30d..694bed8d82 100644
--- a/read-cache.c
+++ b/read-cache.c
@@ -1499,6 +1499,7 @@ struct ondisk_cache_entry_extended {
 };
 
 /* These are only used for v3 or lower */
+#define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
@@ -2032,7 +2033,7 @@ static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
 }
 
 /* Copy miscellaneous fields but not the name */
-static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
+static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
 				       struct cache_entry *ce)
 {
 	short flags;
@@ -2056,32 +2057,35 @@ static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
 		struct ondisk_cache_entry_extended *ondisk2;
 		ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
 		ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
-		return ondisk2->name;
-	}
-	else {
-		return ondisk->name;
 	}
 }
 
 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
-			  struct strbuf *previous_name)
+			  struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
 {
 	int size;
-	struct ondisk_cache_entry *ondisk;
 	int saved_namelen = saved_namelen; /* compiler workaround */
-	char *name;
 	int result;
+	static unsigned char padding[8] = { 0x00 };
 
 	if (ce->ce_flags & CE_STRIP_NAME) {
 		saved_namelen = ce_namelen(ce);
 		ce->ce_namelen = 0;
 	}
 
+	if (ce->ce_flags & CE_EXTENDED)
+		size = offsetof(struct ondisk_cache_entry_extended, name);
+	else
+		size = offsetof(struct ondisk_cache_entry, name);
+
 	if (!previous_name) {
-		size = ondisk_ce_size(ce);
-		ondisk = xcalloc(1, size);
-		name = copy_cache_entry_to_ondisk(ondisk, ce);
-		memcpy(name, ce->name, ce_namelen(ce));
+		int len = ce_namelen(ce);
+		copy_cache_entry_to_ondisk(ondisk, ce);
+		result = ce_write(c, fd, ondisk, size);
+		if (!result)
+			result = ce_write(c, fd, ce->name, len);
+		if (!result)
+			result = ce_write(c, fd, padding, align_padding_size(size, len));
 	} else {
 		int common, to_remove, prefix_size;
 		unsigned char to_remove_vi[16];
@@ -2094,16 +2098,12 @@ static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
 		to_remove = previous_name->len - common;
 		prefix_size = encode_varint(to_remove, to_remove_vi);
 
-		if (ce->ce_flags & CE_EXTENDED)
-			size = offsetof(struct ondisk_cache_entry_extended, name);
-		else
-			size = offsetof(struct ondisk_cache_entry, name);
-		size += prefix_size + (ce_namelen(ce) - common + 1);
-
-		ondisk = xcalloc(1, size);
-		name = copy_cache_entry_to_ondisk(ondisk, ce);
-		memcpy(name, to_remove_vi, prefix_size);
-		memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
+		copy_cache_entry_to_ondisk(ondisk, ce);
+		result = ce_write(c, fd, ondisk, size);
+		if (!result)
+			result = ce_write(c, fd, to_remove_vi, prefix_size);
+		if (!result)
+			result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common + 1);
 
 		strbuf_splice(previous_name, common, to_remove,
 			      ce->name + common, ce_namelen(ce) - common);
@@ -2113,8 +2113,6 @@ static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
 		ce->ce_flags &= ~CE_STRIP_NAME;
 	}
 
-	result = ce_write(c, fd, ondisk, size);
-	free(ondisk);
 	return result;
 }
 
@@ -2196,6 +2194,7 @@ static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
 	struct cache_entry **cache = istate->cache;
 	int entries = istate->cache_nr;
 	struct stat st;
+	struct ondisk_cache_entry_extended ondisk;
 	struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
 	int drop_cache_tree = 0;
 
@@ -2232,6 +2231,7 @@ static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
 		return -1;
 
 	previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
+
 	for (i = 0; i < entries; i++) {
 		struct cache_entry *ce = cache[i];
 		if (ce->ce_flags & CE_REMOVE)
@@ -2251,7 +2251,7 @@ static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
 
 			drop_cache_tree = 1;
 		}
-		if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
+		if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
 			err = -1;
 
 		if (err)
-- 
2.14.1.205.g2812f3410d

Re: [PATCH 2/3] read-cache: fix memory leak in do_write_index

[Junio C Hamano]

Kevin Willford <kewillf@microsoft.com> writes:

> The previous_name_buf was never getting released when there
> was an error in ce_write_entry or allow was false and execution
> was returned to the caller.
>
> Signed-off-by: Kevin Willford <kewillf@microsoft.com>
> ---

Thanks for spotting an old error that dates back to at least 2013
this month.  The patch looks sensible.



>  read-cache.c | 12 +++++++++---
>  1 file changed, 9 insertions(+), 3 deletions(-)
>
> diff --git a/read-cache.c b/read-cache.c
> index acfb028f48..47220cc30d 100644
> --- a/read-cache.c
> +++ b/read-cache.c
> @@ -2192,7 +2192,7 @@ static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
>  	int newfd = tempfile->fd;
>  	git_SHA_CTX c;
>  	struct cache_header hdr;
> -	int i, err, removed, extended, hdr_version;
> +	int i, err = 0, removed, extended, hdr_version;
>  	struct cache_entry **cache = istate->cache;
>  	int entries = istate->cache_nr;
>  	struct stat st;
> @@ -2247,15 +2247,21 @@ static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
>  			if (allow)
>  				warning(msg, ce->name);
>  			else
> -				return error(msg, ce->name);
> +				err = error(msg, ce->name);
>  
>  			drop_cache_tree = 1;
>  		}
>  		if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
> -			return -1;
> +			err = -1;
> +
> +		if (err)
> +			break;
>  	}
>  	strbuf_release(&previous_name_buf);
>  
> +	if (err)
> +		return err;
> +
>  	/* Write extension data here */
>  	if (!strip_extensions && istate->split_index) {
>  		struct strbuf sb = STRBUF_INIT;

Re: [PATCH 3/3] read-cache: avoid allocating every ondisk entry when writing

[Junio C Hamano]

Kevin Willford <kewillf@microsoft.com> writes:

> When writing the index for each entry an ondisk struct will be
> allocated and freed in ce_write_entry.  We can do better by
> using a ondisk struct on the stack for each entry.
>
> This is accomplished by using a stack ondisk_cache_entry_extended
> outside looping through the entries in do_write_index.  Only the
> fixed fields of this struct are used when writing and depending on
> whether it is extended or not the flags2 field will be written.
> The name field is not used and instead the cache_entry name field
> is used directly when writing out the name.  Because ce_write is
> using a buffer and memcpy to fill the buffer before flushing to disk,
> we don't have to worry about doing multiple ce_write calls.

The code already heavily assumes that the only difference between
ondisk_cache_entry and its _extended variant are identical in their
earlier parts, so having a single instance of the larger one on the
stack and passing it around makes sense as an optimization.

Multiple chomped calls to ce_write() does make me nervous, not due
to potential I/O cost (which is none) but to potential programming
mistakes for people who touch the code.  The version immediately
after this patch looks correct in that once "result" is set to non
zero to indicate an error nothing is written out, but keeping that
property over time, when a newer index format is eventually invented,
may become an additional burden.

But overall, I think this is a good change.  Will queue all three.

Thanks.