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

[PATCHv7 3/9] send-pack.c: add --atomic command line argument

From
Stefan Beller <sbeller@google.com>
Date
Dec 29, 2014, 21:35 UTC
Message-ID
<1419888915-31760-4-git-send-email-sbeller@google.com>
In-Reply-To
<1419888915-31760-1-git-send-email-sbeller@google.com>
From: Ronnie Sahlberg <sahlberg@google.com>

This adds support to send-pack to negotiate and use atomic pushes iff the server supports it. Atomic pushes are activated by a new command line flag --atomic.

In order to do this we also need to change the semantics for send_pack() slightly. The existing send_pack() function actually doesn't send all the refs back to the server when multiple refs are involved, for example when using --all. Several of the failure modes for pushes can already be detected locally in the send_pack client based on the information from the initial server side list of all the refs as generated by receive-pack. Any such refs that we thus know would fail to push are thus pruned from the list of refs we send to the server to update.

For atomic pushes, we have to deal thus with both failures that are detected locally as well as failures that are reported back from the server. In order to do so we treat all local failures as push failures too.

We introduce a new status code REF_STATUS_ATOMIC_PUSH_FAILED so we can flag all refs that we would normally have tried to push to the server but we did not due to local failures. This is to improve the error message back to the end user to flag that "these refs failed to update since the atomic push operation failed."

Signed-off-by: Ronnie Sahlberg <sahlberg@google.com>
Signed-off-by: Stefan Beller <sbeller@google.com>
---
Notes:
    Notes:
        Changes v1 -> v2:
         * Now we only need a very small change in the existing code and have
           a new static function, which cares about error reporting.
              Junio wrote:
              > Hmph.  Is "atomic push" so special that it deserves a separate
              > parameter?  When we come up with yet another mode of failure, would
              > we add another parameter to the callers to this function?
         * error messages are worded differently (lower case!),
         * use of error function instead of fprintf
    
         * undashed the printed error message ("atomic push failed");
    
    Changes v2 -> v3:
    > We avoid assignment inside a conditional.
    
    Ok I switched to using a switch statement.
    
    skipped v4 v5
    
    v6:
    * realign to one error string:
    +	error("atomic push failed for ref %s. status: %d\n",
    +	      failing_ref->name, failing_ref->status);
    
    * Use correct value now (negative defined from previous patch)
    
    v7:
     * return error(...); instead of error(...); return -1;
 Documentation/git-send-pack.txt |  7 ++++++-
 builtin/send-pack.c             |  6 +++++-
 remote.h                        |  3 ++-
 send-pack.c                     | 39 +++++++++++++++++++++++++++++++++++++--
 transport.c                     |  4 ++++
 5 files changed, 54 insertions(+), 5 deletions(-)
diff --git a/Documentation/git-send-pack.txt b/Documentation/git-send-pack.txt
index 2a0de42..45c7725 100644
--- a/Documentation/git-send-pack.txt
+++ b/Documentation/git-send-pack.txt
@@ -9,7 +9,7 @@ git-send-pack - Push objects over Git protocol to another repository
 SYNOPSIS
 --------
 [verse]
-'git send-pack' [--all] [--dry-run] [--force] [--receive-pack=<git-receive-pack>] [--verbose] [--thin] [<host>:]<directory> [<ref>...]
+'git send-pack' [--all] [--dry-run] [--force] [--receive-pack=<git-receive-pack>] [--verbose] [--thin] [--atomic] [<host>:]<directory> [<ref>...]
 
 DESCRIPTION
 -----------
@@ -62,6 +62,11 @@ be in a separate packet, and the list must end with a flush packet.
 	Send a "thin" pack, which records objects in deltified form based
 	on objects not included in the pack to reduce network traffic.
 
+--atomic::
+	Use an atomic transaction for updating the refs. If any of the refs
+	fails to update then the entire push will fail without changing any
+	refs.
+
 <host>::
 	A remote host to house the repository.  When this
 	part is specified, 'git-receive-pack' is invoked via
diff --git a/builtin/send-pack.c b/builtin/send-pack.c
index b564a77..b961e5a 100644
--- a/builtin/send-pack.c
+++ b/builtin/send-pack.c
@@ -13,7 +13,7 @@
 #include "sha1-array.h"
 
 static const char send_pack_usage[] =
-"git send-pack [--all | --mirror] [--dry-run] [--force] [--receive-pack=<git-receive-pack>] [--verbose] [--thin] [<host>:]<directory> [<ref>...]\n"
+"git send-pack [--all | --mirror] [--dry-run] [--force] [--receive-pack=<git-receive-pack>] [--verbose] [--thin] [--atomic] [<host>:]<directory> [<ref>...]\n"
 "  --all and explicit <ref> specification are mutually exclusive.";
 
 static struct send_pack_args args;
@@ -170,6 +170,10 @@ int cmd_send_pack(int argc, const char **argv, const char *prefix)
 				args.use_thin_pack = 1;
 				continue;
 			}
+			if (!strcmp(arg, "--atomic")) {
+				args.atomic = 1;
+				continue;
+			}
 			if (!strcmp(arg, "--stateless-rpc")) {
 				args.stateless_rpc = 1;
 				continue;
diff --git a/remote.h b/remote.h
index 8b62efd..f346524 100644
--- a/remote.h
+++ b/remote.h
@@ -115,7 +115,8 @@ struct ref {
 		REF_STATUS_REJECT_SHALLOW,
 		REF_STATUS_UPTODATE,
 		REF_STATUS_REMOTE_REJECT,
-		REF_STATUS_EXPECTING_REPORT
+		REF_STATUS_EXPECTING_REPORT,
+		REF_STATUS_ATOMIC_PUSH_FAILED
 	} status;
 	char *remote_status;
 	struct ref *peer_ref; /* when renaming */
diff --git a/send-pack.c b/send-pack.c
index b7d8e01..e8f60df 100644
--- a/send-pack.c
+++ b/send-pack.c
@@ -282,6 +282,29 @@ free_return:
 	return update_seen;
 }
 
+
+static int atomic_push_failure(struct send_pack_args *args,
+			       struct ref *remote_refs,
+			       struct ref *failing_ref)
+{
+	struct ref *ref;
+	/* Mark other refs as failed */
+	for (ref = remote_refs; ref; ref = ref->next) {
+		if (!ref->peer_ref && !args->send_mirror)
+			continue;
+
+		switch (ref->status) {
+		case REF_STATUS_EXPECTING_REPORT:
+			ref->status = REF_STATUS_ATOMIC_PUSH_FAILED;
+			continue;
+		default:
+			; /* do nothing */
+		}
+	}
+	return error("atomic push failed for ref %s. status: %d\n",
+		     failing_ref->name, failing_ref->status);
+}
+
 int send_pack(struct send_pack_args *args,
 	      int fd[], struct child_process *conn,
 	      struct ref *remote_refs,
@@ -373,9 +396,21 @@ int send_pack(struct send_pack_args *args,
 	 * the pack data.
 	 */
 	for (ref = remote_refs; ref; ref = ref->next) {
-		if (check_to_send_update(ref, args) < 0)
+		switch (check_to_send_update(ref, args)) {
+		case 0: /* no error */
+			break;
+		case CHECK_REF_STATUS_REJECTED:
+			/*
+			 * When we know the server would reject a ref update if
+			 * we were to send it and we're trying to send the refs
+			 * atomically, abort the whole operation.
+			 */
+			if (use_atomic)
+				return atomic_push_failure(args, remote_refs, ref);
+			/* Fallthrough for non atomic case. */
+		default:
 			continue;
-
+		}
 		if (!ref->deletion)
 			need_pack_data = 1;
 
diff --git a/transport.c b/transport.c
index 70d38e4..c67feee 100644
--- a/transport.c
+++ b/transport.c
@@ -728,6 +728,10 @@ static int print_one_push_status(struct ref *ref, const char *dest, int count, i
 						 ref->deletion ? NULL : ref->peer_ref,
 						 "remote failed to report status", porcelain);
 		break;
+	case REF_STATUS_ATOMIC_PUSH_FAILED:
+		print_ref_status('!', "[rejected]", ref, ref->peer_ref,
+						 "atomic push failed", porcelain);
+		break;
 	case REF_STATUS_OK:
 		print_ok_ref_status(ref, porcelain);
 		break;

-- 
2.2.1.62.g3f15098
Previous: Stefan BellerNext: Stefan Beller
Message 4 of 11 in “[PATCHv7 0/9] atomic pushes”
  1. Stefan BellerDec 29, 2014
  2. 1/9 receive-pack.c: add documentation for atomic push supportStefan Beller, Dec 29, 2014
  3. 2/9 send-pack: Rename ref_update_to_be_sent to check_to_send_updateStefan Beller, Dec 29, 2014
  4. 3/9 send-pack.c: add --atomic command line argumentStefan Beller, Dec 29, 2014
  5. 4/9 receive-pack.c: simplify execute_commandsStefan Beller, Dec 29, 2014
  6. 5/9 receive-pack.c: move transaction handling in a central placeStefan Beller, Dec 29, 2014
  7. 6/9 receive-pack.c: add execute_commands_atomic functionStefan Beller, Dec 29, 2014
  8. 7/9 receive-pack.c: enable atomic push protocol supportStefan Beller, Dec 29, 2014
  9. 8/9 push.c: add an --atomic argumentStefan Beller, Dec 29, 2014
  10. 9/9 t5543-atomic-push.sh: add basic tests for atomic pushesStefan Beller, Dec 29, 2014
  11. Stefan BellerDec 29, 2014

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.