git/list[1] front-page[2] threads[3] people[4] search[5] about
 

[PATCH v2] Perform cheaper connectivity check when pack is used as medium

From
Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Date
Feb 28, 2012, 13:18 UTC
Message-ID
<1330435109-4437-1-git-send-email-pclouds@gmail.com>
In-Reply-To
<7vmx849ma8.fsf@alter.siamese.dyndns.org>

When we fetch or push, usually "git rev-list --verify-objects --not --all --stdin" is used to make sure that there are no gaps between existing refs and new refs. --verify-objects calls parse_object(), which internally calls check_sha1_signature() to verify object content matches its SHA-1 signature.

check_sha1_signature() is an expensive operation, especially when new refs are far away from existing ones because all objects in between are re-hashed. However, if we receive new objects by pack, we can skip the operation because packs themselves do not contain SHA-1 signatures. All signatures are recreated by unpack-objects/index-pack's hashing objects in the pack, which we can trust.

Detect pack transfer cases and turn --verify-objects to --objects. --objects is similar to --verify-objects except that it does not call check_sha1_signature().

As an (extreme) example, a repository is created with only one commit:
e83c516 (Initial revision of "git", the information manager from hell
- 2005-04-07). The rest of git.git is fetched on top. Without the
patch:
$ time git fetch file:///home/pclouds/w/git/.git
remote: Counting objects: 125638, done.
remote: Compressing objects: 100% (33201/33201), done.
remote: Total 125638 (delta 92568), reused 123517 (delta 90743)
Receiving objects: 100% (125638/125638), 34.58 MiB | 8.07 MiB/s, done.
Resolving deltas: 100% (92568/92568), done.
From file:///home/pclouds/w/git/
 * branch            HEAD       -> FETCH_HEAD

real 1m30.972s user 1m31.410s sys 0m1.757s

With the patch:
$ time git fetch file:///home/pclouds/w/git/.git
remote: Counting objects: 125647, done.
remote: Compressing objects: 100% (33209/33209), done.
remote: Total 125647 (delta 92576), reused 123516 (delta 90744)
Receiving objects: 100% (125647/125647), 34.58 MiB | 7.99 MiB/s, done.
Resolving deltas: 100% (92576/92576), done.
From file:///home/pclouds/w/git/
 * branch            HEAD       -> FETCH_HEAD

real 0m51.456s user 0m52.737s sys 0m1.548s

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
---
 combine "quiet" and "strict" to a flag word
 builtin/fetch.c |   14 +++++++++-----
 connected.c     |   11 +++++++----
 connected.h     |    6 +++++-
 transport.c     |    5 +++++
 transport.h     |    1 +
 5 files changed, 27 insertions(+), 10 deletions(-)
diff --git a/builtin/fetch.c b/builtin/fetch.c
index 65f5f9b..70c9ca3 100644
--- a/builtin/fetch.c
+++ b/builtin/fetch.c
@@ -368,7 +368,7 @@ static int iterate_ref_map(void *cb_data, unsigned char sha1[20])
 }
 
 static int store_updated_refs(const char *raw_url, const char *remote_name,
-		struct ref *ref_map)
+			      struct ref *ref_map, int use_pack)
 {
 	FILE *fp;
 	struct commit *commit;
@@ -389,7 +389,9 @@ static int store_updated_refs(const char *raw_url, const char *remote_name,
 		url = xstrdup("foreign");
 
 	rm = ref_map;
-	if (check_everything_connected(iterate_ref_map, 0, &rm)) {
+	if (check_everything_connected(iterate_ref_map,
+				       use_pack ? 0 : CHECK_CONNECT_STRICT,
+				       &rm)) {
 		rc = error(_("%s did not send all necessary objects\n"), url);
 		goto abort;
 	}
@@ -516,7 +518,8 @@ static int quickfetch(struct ref *ref_map)
 	 */
 	if (depth)
 		return -1;
-	return check_everything_connected(iterate_ref_map, 1, &rm);
+	return check_everything_connected(iterate_ref_map,
+					  CHECK_CONNECT_QUIET, &rm);
 }
 
 static int fetch_refs(struct transport *transport, struct ref *ref_map)
@@ -526,8 +529,9 @@ static int fetch_refs(struct transport *transport, struct ref *ref_map)
 		ret = transport_fetch_refs(transport, ref_map);
 	if (!ret)
 		ret |= store_updated_refs(transport->url,
-				transport->remote->name,
-				ref_map);
+					  transport->remote->name,
+					  ref_map,
+					  is_pack_transport(transport));
 	transport_unlock_pack(transport);
 	return ret;
 }
diff --git a/connected.c b/connected.c
index d762423..9df357d 100644
--- a/connected.c
+++ b/connected.c
@@ -14,10 +14,11 @@
  *
  * Returns 0 if everything is connected, non-zero otherwise.
  */
-int check_everything_connected(sha1_iterate_fn fn, int quiet, void *cb_data)
+int check_everything_connected(sha1_iterate_fn fn, unsigned int flags,
+			       void *cb_data)
 {
 	struct child_process rev_list;
-	const char *argv[] = {"rev-list", "--verify-objects",
+	const char *argv[] = {"rev-list", "--objects",
 			      "--stdin", "--not", "--all", NULL, NULL};
 	char commit[41];
 	unsigned char sha1[20];
@@ -26,7 +27,9 @@ int check_everything_connected(sha1_iterate_fn fn, int quiet, void *cb_data)
 	if (fn(cb_data, sha1))
 		return err;
 
-	if (quiet)
+	if (flags & CHECK_CONNECT_STRICT)
+		argv[1] = "--verify-objects";
+	if (flags & CHECK_CONNECT_QUIET)
 		argv[5] = "--quiet";
 
 	memset(&rev_list, 0, sizeof(rev_list));
@@ -34,7 +37,7 @@ int check_everything_connected(sha1_iterate_fn fn, int quiet, void *cb_data)
 	rev_list.git_cmd = 1;
 	rev_list.in = -1;
 	rev_list.no_stdout = 1;
-	rev_list.no_stderr = quiet;
+	rev_list.no_stderr = flags & CHECK_CONNECT_QUIET;
 	if (start_command(&rev_list))
 		return error(_("Could not run 'git rev-list'"));
 
diff --git a/connected.h b/connected.h
index 7e4585a..52636b9 100644
--- a/connected.h
+++ b/connected.h
@@ -1,6 +1,9 @@
 #ifndef CONNECTED_H
 #define CONNECTED_H
 
+#define CHECK_CONNECT_QUIET  1
+#define CHECK_CONNECT_STRICT 2
+
 /*
  * Take callback data, and return next object name in the buffer.
  * When called after returning the name for the last object, return -1
@@ -15,6 +18,7 @@ typedef int (*sha1_iterate_fn)(void *, unsigned char [20]);
  *
  * Return 0 if Ok, non zero otherwise (i.e. some missing objects)
  */
-extern int check_everything_connected(sha1_iterate_fn, int quiet, void *cb_data);
+extern int check_everything_connected(sha1_iterate_fn, unsigned int flags,
+				      void *cb_data);
 
 #endif /* CONNECTED_H */
diff --git a/transport.c b/transport.c
index 181f8f2..cd5e0ca 100644
--- a/transport.c
+++ b/transport.c
@@ -1248,3 +1248,8 @@ void for_each_alternate_ref(alternate_ref_fn fn, void *data)
 	cb.data = data;
 	foreach_alt_odb(refs_from_alternate_cb, &cb);
 }
+
+int is_pack_transport(const struct transport *transport)
+{
+	return transport->fetch == fetch_refs_via_pack;
+}
diff --git a/transport.h b/transport.h
index ce99ef8..7cf72ff 100644
--- a/transport.h
+++ b/transport.h
@@ -150,6 +150,7 @@ int transport_disconnect(struct transport *transport);
 char *transport_anonymize_url(const char *url);
 void transport_take_over(struct transport *transport,
 			 struct child_process *child);
+int is_pack_transport(const struct transport *transport);
 
 int transport_connect(struct transport *transport, const char *name,
 		      const char *exec, int fd[2]);
-- 
1.7.8.36.g69ee2
Previous: Junio C HamanoNext: Johannes Sixt
Message 3 of 23 in “Perform cheaper connectivity check when pack is used as medium”
  1. Perform cheaper connectivity check when pack is used as mediumNguyễn Thái Ngọc Duy, Feb 27, 2012
  2. Junio C HamanoFeb 27, 2012
  3. Perform cheaper connectivity check when pack is used as mediumNguyễn Thái Ngọc Duy, Feb 28, 2012
  4. Johannes SixtFeb 28, 2012
  5. Andreas EricssonFeb 28, 2012
  6. Nguyen Thai Ngoc DuyFeb 29, 2012
  7. Junio C HamanoMar 2, 2012
  8. Nguyen Thai Ngoc DuyMar 2, 2012
  9. Junio C HamanoMar 2, 2012
  10. Nguyen Thai Ngoc DuyMar 3, 2012
  11. Junio C HamanoMar 3, 2012
  12. Nguyen Thai Ngoc DuyMar 3, 2012
  13. Nguyen Thai Ngoc DuyMar 6, 2012
  14. Junio C HamanoMar 6, 2012
  15. {fetch,receive}-pack: skip sha-1 integrity test on objects from new packNguyễn Thái Ngọc Duy, Mar 14, 2012
  16. Junio C HamanoMar 14, 2012
  17. Junio C HamanoMar 15, 2012
  18. Junio C HamanoMar 15, 2012
  19. Nguyen Thai Ngoc DuyMar 16, 2012
  20. Junio C HamanoMar 16, 2012
  21. Nguyen Thai Ngoc DuyMar 16, 2012
  22. Nguyen Thai Ngoc DuyMar 16, 2012
  23. Junio C HamanoMar 16, 2012

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.