diff mbox series

upload-pack: add tracing for fetches

Message ID pull.1598.git.1697040242703.gitgitgadget@gmail.com (mailing list archive)
State Superseded
Headers show
Series upload-pack: add tracing for fetches | expand

Commit Message

Robert Coup Oct. 11, 2023, 4:04 p.m. UTC
From: Robert Coup <robert@coup.net.nz>

Information on how users are accessing hosted repositories can be
helpful to server operators. For example, being able to broadly
differentiate between fetches and initial clones; the use of shallow
repository features; or partial clone filters.

a29263c (fetch-pack: add tracing for negotiation rounds, 2022-08-02)
added some information on have counts to fetch-pack itself to help
diagnose negotiation; but from a git-upload-pack (server) perspective,
there's no means of accessing such information without using
GIT_TRACE_PACKET to examine the protocol packets.

Improve this by emitting a Trace2 JSON event from upload-pack with
summary information on the contents of a fetch request.

* haves, wants, and want-ref counts can help determine (broadly) between
  fetches and clones, and the use of single-branch, etc.
* shallow clone depth, tip counts, and deepening options.
* any partial clone filter type.

Signed-off-by: Robert Coup <robert@coup.net.nz>
---
    upload-pack: add tracing for fetches

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1598%2Frcoup%2Frc-upload-pack-trace-info-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1598/rcoup/rc-upload-pack-trace-info-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/1598

 t/t5500-fetch-pack.sh | 38 +++++++++++++++++++++++++++-----------
 upload-pack.c         | 29 +++++++++++++++++++++++++++++
 2 files changed, 56 insertions(+), 11 deletions(-)


base-commit: aab89be2eb6ca51eefeb8c8066f673f447058856

Comments

Taylor Blau Oct. 11, 2023, 9:26 p.m. UTC | #1
On Wed, Oct 11, 2023 at 04:04:02PM +0000, Robert Coup via GitGitGadget wrote:
> diff --git a/t/t5500-fetch-pack.sh b/t/t5500-fetch-pack.sh
> index d18f2823d86..bb15ac34f77 100755
> --- a/t/t5500-fetch-pack.sh
> +++ b/t/t5500-fetch-pack.sh
> @@ -132,13 +132,18 @@ test_expect_success 'single branch object count' '
>  '
>
>  test_expect_success 'single given branch clone' '
> -	git clone --single-branch --branch A "file://$(pwd)/." branch-a &&
> -	test_must_fail git --git-dir=branch-a/.git rev-parse origin/B
> +	GIT_TRACE2_EVENT="$(pwd)/branch-a/trace2_event" \
> +		git clone --single-branch --branch A "file://$(pwd)/." branch-a &&
> +	test_must_fail git --git-dir=branch-a/.git rev-parse origin/B &&
> +	grep \"fetch-info\".*\"haves\":0 branch-a/trace2_event &&
> +	grep \"fetch-info\".*\"wants\":1 branch-a/trace2_event

Not at all related to your patch here, but I feel like we have a bunch
of these greps sprinkled throughout the test suite which serve to
inspect some key(s) of a JSON object printed to the trace2 event stream.

It might be nice to have a more robust test-tool that could do this
heavy lifting for us instead of having to write these grep expressions
ourselves.

But definitely outside the scope of this patch ;-).

Thanks,
Taylor
Taylor Blau Oct. 11, 2023, 9:32 p.m. UTC | #2
On Wed, Oct 11, 2023 at 04:04:02PM +0000, Robert Coup via GitGitGadget wrote:
> From: Robert Coup <robert@coup.net.nz>
>
> Information on how users are accessing hosted repositories can be
> helpful to server operators. For example, being able to broadly
> differentiate between fetches and initial clones; the use of shallow
> repository features; or partial clone filters.

Indeed. One of the custom patches that GitHub is carrying in its private
fork is something similar to this that dumps information from
upload-pack into a custom logging system specific to GitHub (called
"sockstat", in case anybody was curious).

I suspect that we would still live with that patch because we depend on
some of the custom logging infrastructure provided by sockstat, but this
is definitely a good direction to be pursuing for git.git nonetheless.

> @@ -1552,6 +1553,32 @@ static int parse_have(const char *line, struct oid_array *haves)
>  	return 0;
>  }
>
> +static void trace2_fetch_info(struct upload_pack_data *data)
> +{
> +	struct json_writer jw = JSON_WRITER_INIT;
> +
> +	jw_object_begin(&jw, 0);
> +	{

Is there a reason that we have a separate scope here? I think we may
want to drop this as unnecessary, but it's entirely possible that I'm
missing something here...

> +		jw_object_intmax(&jw, "haves", data->haves.nr);
> +		jw_object_intmax(&jw, "wants", data->want_obj.nr);
> +		jw_object_intmax(&jw, "want-refs", data->wanted_refs.nr);
> +		jw_object_intmax(&jw, "depth", data->depth);
> +		jw_object_intmax(&jw, "shallows", data->shallows.nr);
> +		jw_object_bool(&jw, "deepen-since", data->deepen_since);
> +		jw_object_intmax(&jw, "deepen-not", data->deepen_not.nr);
> +		jw_object_bool(&jw, "deepen-relative", data->deepen_relative);
> +		if (data->filter_options.choice)
> +			jw_object_string(&jw, "filter", list_object_filter_config_name(data->filter_options.choice));

I'm pretty sure that list_object_filter_config_name() returns characters
that are safe for JSON-encoding, and/or that jw_object_string() does any
quoting beforehand, but worth checking nonetheless.

> +		else
> +			jw_object_null(&jw, "filter");

These all seem like useful things to have.

Thanks,
Taylor
Jeff King Oct. 11, 2023, 10:27 p.m. UTC | #3
On Wed, Oct 11, 2023 at 04:04:02PM +0000, Robert Coup via GitGitGadget wrote:

> Improve this by emitting a Trace2 JSON event from upload-pack with
> summary information on the contents of a fetch request.
> 
> * haves, wants, and want-ref counts can help determine (broadly) between
>   fetches and clones, and the use of single-branch, etc.
> * shallow clone depth, tip counts, and deepening options.
> * any partial clone filter type.

I think this is a reasonable thing to have. I see Taylor left a more
detailed review, but I did notice one thing...

> +static void trace2_fetch_info(struct upload_pack_data *data)
> +{
> +	struct json_writer jw = JSON_WRITER_INIT;
> +
> +	jw_object_begin(&jw, 0);
> +	{
> +		jw_object_intmax(&jw, "haves", data->haves.nr);
> +		jw_object_intmax(&jw, "wants", data->want_obj.nr);
> +		jw_object_intmax(&jw, "want-refs", data->wanted_refs.nr);
> +		jw_object_intmax(&jw, "depth", data->depth);
> +		jw_object_intmax(&jw, "shallows", data->shallows.nr);
> +		jw_object_bool(&jw, "deepen-since", data->deepen_since);
> +		jw_object_intmax(&jw, "deepen-not", data->deepen_not.nr);
> +		jw_object_bool(&jw, "deepen-relative", data->deepen_relative);
> +		if (data->filter_options.choice)
> +			jw_object_string(&jw, "filter", list_object_filter_config_name(data->filter_options.choice));
> +		else
> +			jw_object_null(&jw, "filter");
> +	}
> +	jw_end(&jw);
> +
> +	trace2_data_json("upload-pack", the_repository, "fetch-info", &jw);
> +
> +	jw_release(&jw);
> +}

Generating the json output isn't entirely trivial (and certainly
involves allocations), but we throw it away unused if tracing isn't
enabled. Maybe we'd want something like:

  if (!trace2_is_enabled())
          return;

at the top of the function? It looks like other callers of
jw_object_begin() have a similar issue, and this is probably premature
optimization to some degree. It just feels like it should be easy for
tracing to be zero-cost (beyond a single conditional) when it's
disabled.

-Peff
Robert Coup Oct. 11, 2023, 10:33 p.m. UTC | #4
Hi Taylor,

Thanks for taking a look.

On Wed, 11 Oct 2023 at 22:32, Taylor Blau <me@ttaylorr.com> wrote:
>
> One of the custom patches that GitHub is carrying in its private
> fork is something similar to this that dumps information from
> upload-pack into a custom logging system specific to GitHub

I suspected as much. My initial goal is clones vs fetches, filter use, and
shallow use... since I'm changing code is there anything else that you think is
worthwhile instrumenting? I thought about casting the ints to booleans
(ie: >0 haves; >0 wants; etc) but I figured the actual numbers might come in
handy at some point, and changing them back in future would be incompatible. I
deliberately didn't add thin/ofs/etc flags: most of that stuff is on by default
now so it felt like noise.

> > + jw_object_begin(&jw, 0);
> > + {
>
> Is there a reason that we have a separate scope here? I think we may
> want to drop this as unnecessary, but it's entirely possible that I'm
> missing something here...

I just followed the patterns in test-json-writer.c - personally I think since
JSON objects/arrays have starts & ends then it's easy to read. But one person's
readable is another's coding style violation ;-)

> > + jw_object_string(&jw, "filter", list_object_filter_config_name(data->filter_options.choice));
>
> I'm pretty sure that list_object_filter_config_name() returns characters
> that are safe for JSON-encoding, and/or that jw_object_string() does any
> quoting beforehand, but worth checking nonetheless.

Yes, jw_object_string() calls append_quoted_string() which deals with escaping
and quoting.

Thanks,

Rob :)
Robert Coup Oct. 11, 2023, 10:36 p.m. UTC | #5
Hi Jeff,

On Wed, 11 Oct 2023 at 23:27, Jeff King <peff@peff.net> wrote:
> I think this is a reasonable thing to have.

Thanks for taking a look

> Generating the json output isn't entirely trivial (and certainly
> involves allocations), but we throw it away unused if tracing isn't
> enabled. Maybe we'd want something like:
>
>   if (!trace2_is_enabled())
>           return;
>
> at the top of the function? It looks like other callers of
> jw_object_begin() have a similar issue, and this is probably premature
> optimization to some degree. It just feels like it should be easy for
> tracing to be zero-cost (beyond a single conditional) when it's
> disabled.

This definitely makes sense to me. Will add to a v2.

Thanks, Rob :)
diff mbox series

Patch

diff --git a/t/t5500-fetch-pack.sh b/t/t5500-fetch-pack.sh
index d18f2823d86..bb15ac34f77 100755
--- a/t/t5500-fetch-pack.sh
+++ b/t/t5500-fetch-pack.sh
@@ -132,13 +132,18 @@  test_expect_success 'single branch object count' '
 '
 
 test_expect_success 'single given branch clone' '
-	git clone --single-branch --branch A "file://$(pwd)/." branch-a &&
-	test_must_fail git --git-dir=branch-a/.git rev-parse origin/B
+	GIT_TRACE2_EVENT="$(pwd)/branch-a/trace2_event" \
+		git clone --single-branch --branch A "file://$(pwd)/." branch-a &&
+	test_must_fail git --git-dir=branch-a/.git rev-parse origin/B &&
+	grep \"fetch-info\".*\"haves\":0 branch-a/trace2_event &&
+	grep \"fetch-info\".*\"wants\":1 branch-a/trace2_event
 '
 
 test_expect_success 'clone shallow depth 1' '
-	git clone --no-single-branch --depth 1 "file://$(pwd)/." shallow0 &&
-	test "$(git --git-dir=shallow0/.git rev-list --count HEAD)" = 1
+	GIT_TRACE2_EVENT="$(pwd)/shallow0/trace2_event" \
+		git clone --no-single-branch --depth 1 "file://$(pwd)/." shallow0 &&
+	test "$(git --git-dir=shallow0/.git rev-list --count HEAD)" = 1 &&
+	grep \"fetch-info\".*\"depth\":1 shallow0/trace2_event
 '
 
 test_expect_success 'clone shallow depth 1 with fsck' '
@@ -235,7 +240,10 @@  test_expect_success 'add two more (part 2)' '
 test_expect_success 'deepening pull in shallow repo' '
 	(
 		cd shallow &&
-		git pull --depth 4 .. B
+		GIT_TRACE2_EVENT="$(pwd)/trace2_event" \
+			git pull --depth 4 .. B &&
+		grep \"fetch-info\".*\"depth\":4 trace2_event &&
+		grep \"fetch-info\".*\"shallows\":2 trace2_event
 	)
 '
 
@@ -306,9 +314,12 @@  test_expect_success 'fetch --depth --no-shallow' '
 test_expect_success 'turn shallow to complete repository' '
 	(
 		cd shallow &&
-		git fetch --unshallow &&
+		GIT_TRACE2_EVENT="$(pwd)/trace2_event" \
+			git fetch --unshallow &&
 		! test -f .git/shallow &&
-		git fsck --full
+		git fsck --full &&
+		grep \"fetch-info\".*\"shallows\":2 trace2_event &&
+		grep \"fetch-info\".*\"depth\":2147483647 trace2_event
 	)
 '
 
@@ -826,13 +837,15 @@  test_expect_success 'clone shallow since ...' '
 '
 
 test_expect_success 'fetch shallow since ...' '
-	git -C shallow11 fetch --shallow-since "200000000 +0700" origin &&
+	GIT_TRACE2_EVENT=$(pwd)/shallow11/trace2_event \
+		git -C shallow11 fetch --shallow-since "200000000 +0700" origin &&
 	git -C shallow11 log --pretty=tformat:%s origin/main >actual &&
 	cat >expected <<-\EOF &&
 	three
 	two
 	EOF
-	test_cmp expected actual
+	test_cmp expected actual &&
+	grep \"fetch-info\".*\"deepen-since\":true shallow11/trace2_event
 '
 
 test_expect_success 'clone shallow since selects no commits' '
@@ -987,13 +1000,16 @@  test_expect_success 'filtering by size' '
 	test_config -C server uploadpack.allowfilter 1 &&
 
 	test_create_repo client &&
-	git -C client fetch-pack --filter=blob:limit=0 ../server HEAD &&
+	GIT_TRACE2_EVENT=$(pwd)/client/trace2_event \
+		git -C client fetch-pack --filter=blob:limit=0 ../server HEAD &&
 
 	# Ensure that object is not inadvertently fetched
 	commit=$(git -C server rev-parse HEAD) &&
 	blob=$(git hash-object server/one.t) &&
 	git -C client rev-list --objects --missing=allow-any "$commit" >oids &&
-	! grep "$blob" oids
+	! grep "$blob" oids &&
+
+	grep \"fetch-info\".*\"filter\":\"blob:limit\" client/trace2_event
 '
 
 test_expect_success 'filtering by size has no effect if support for it is not advertised' '
diff --git a/upload-pack.c b/upload-pack.c
index 83f3d2651ab..165c1c6350a 100644
--- a/upload-pack.c
+++ b/upload-pack.c
@@ -33,6 +33,7 @@ 
 #include "commit-reach.h"
 #include "shallow.h"
 #include "write-or-die.h"
+#include "json-writer.h"
 
 /* Remember to update object flag allocation in object.h */
 #define THEY_HAVE	(1u << 11)
@@ -1552,6 +1553,32 @@  static int parse_have(const char *line, struct oid_array *haves)
 	return 0;
 }
 
+static void trace2_fetch_info(struct upload_pack_data *data)
+{
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	{
+		jw_object_intmax(&jw, "haves", data->haves.nr);
+		jw_object_intmax(&jw, "wants", data->want_obj.nr);
+		jw_object_intmax(&jw, "want-refs", data->wanted_refs.nr);
+		jw_object_intmax(&jw, "depth", data->depth);
+		jw_object_intmax(&jw, "shallows", data->shallows.nr);
+		jw_object_bool(&jw, "deepen-since", data->deepen_since);
+		jw_object_intmax(&jw, "deepen-not", data->deepen_not.nr);
+		jw_object_bool(&jw, "deepen-relative", data->deepen_relative);
+		if (data->filter_options.choice)
+			jw_object_string(&jw, "filter", list_object_filter_config_name(data->filter_options.choice));
+		else
+			jw_object_null(&jw, "filter");
+	}
+	jw_end(&jw);
+
+	trace2_data_json("upload-pack", the_repository, "fetch-info", &jw);
+
+	jw_release(&jw);
+}
+
 static void process_args(struct packet_reader *request,
 			 struct upload_pack_data *data)
 {
@@ -1640,6 +1667,8 @@  static void process_args(struct packet_reader *request,
 
 	if (request->status != PACKET_READ_FLUSH)
 		die(_("expected flush after fetch arguments"));
+
+	trace2_fetch_info(data);
 }
 
 static int process_haves(struct upload_pack_data *data, struct oid_array *common)