threads / patch / 60682

patchpush: region_leave trace for negotiate_using_fetch

Subject: [PATCH] push: region_leave trace for negotiate_using_fetch

## tl;dr

4 messages between Jan 3, 2024 and Jan 5, 2024. Diffs are folded; open one to read it.

replies: 3people: 2as markdown or json

Sam Delmerico· Jan 3, 2024, 22:40 UTC · lore

There were two region_enter events for negotiate_using_fetch instead of one enter and one leave. This commit replaces the second region_enter event with a region_leave.

Signed-off-by: Sam Delmerico <delmerico@google.com>
---
 fetch-pack.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Show changes to fetch-pack.c +1 −1

diff --git a/fetch-pack.c b/fetch-pack.c
index 31a72d43de..dba6d79944 100644
--- a/fetch-pack.c
+++ b/fetch-pack.c
@@ -2218,7 +2218,7 @@ void negotiate_using_fetch(const struct oid_array *negotiation_tips,
 					   the_repository, "%d",
 					   negotiation_round);
 	}
-	trace2_region_enter("fetch-pack", "negotiate_using_fetch", the_repository);
+	trace2_region_leave("fetch-pack", "negotiate_using_fetch", the_repository);
 	trace2_data_intmax("negotiate_using_fetch", the_repository,
 			   "total_rounds", negotiation_round);
 	clear_common_flag(acked_commits);

base-commit: a26002b62827b89a19b1084bd75d9371d565d03c

-- 
2.43.0.472.g3155946c3a-goog

Junio C Hamano· Jan 3, 2024, 23:37 UTC · re: Sam Delmerico · lore

Re: [PATCH] push: region_leave trace for negotiate_using_fetch

Sam Delmerico <delmerico@google.com> writes:

Show 8 quoted lines

> There were two region_enter events for negotiate_using_fetch instead of
> one enter and one leave. This commit replaces the second region_enter
> event with a region_leave.
>
> Signed-off-by: Sam Delmerico <delmerico@google.com>
> ---
>  fetch-pack.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)

Looks right, after skimming a29263cf (fetch-pack: add tracing for negotiation rounds, 2022-08-02). Two questions and a half.

 * How did you find it?  Code inspection?  While writing a script to
   parse the output from around this area, your script noticed the
   ever-increasing nesting level?  Something else?

 * Would it be feasible to write some tests or tools that find
   similar problems (semi-)automatically?

 * Is the breakage (before this patch) something easily demonstrated
   in a new test in t/ somewhere?  And if so, would it be worth
   doing?

Thanks.  Will queue.

Show 16 quoted lines

>
> diff --git a/fetch-pack.c b/fetch-pack.c
> index 31a72d43de..dba6d79944 100644
> --- a/fetch-pack.c
> +++ b/fetch-pack.c
> @@ -2218,7 +2218,7 @@ void negotiate_using_fetch(const struct oid_array *negotiation_tips,
>  					   the_repository, "%d",
>  					   negotiation_round);
>  	}
> -	trace2_region_enter("fetch-pack", "negotiate_using_fetch", the_repository);
> +	trace2_region_leave("fetch-pack", "negotiate_using_fetch", the_repository);
>  	trace2_data_intmax("negotiate_using_fetch", the_repository,
>  			   "total_rounds", negotiation_round);
>  	clear_common_flag(acked_commits);
>
> base-commit: a26002b62827b89a19b1084bd75d9371d565d03c

Sam Delmerico· Jan 5, 2024, 01:12 UTC · re: Junio C Hamano · lore

Re: [PATCH] push: region_leave trace for negotiate_using_fetch

* I don't exactly remember how I noticed it. I was doing some
debugging around the push negotiation code and either 1) saw this
region get entered twice in the trace output, or 2) I was just reading
the code around here and saw two enters.

* Perhaps there could be a check before the last git process ends that
checks that all opened trace regions have been closed? I'm not sure
how much work this would involve. It's probably also not a very
proactive way to catch these bugs since it would only get triggered
when a *user* hits a code path with a trace region that never exits.

* There could also be a test that checks that every region_enter trace
log has a corresponding region_leave. But I'm not sure how to ensure
that every code path is checked.

Overall, I'm not sure how much benefit there is from checking for this. I'm not sure that it would have a large impact if it were to happen again. For example, I think that it could be noticed relatively quickly by a person/system looking at metrics like I was (e.g. if the time spent in a region is infinite or zero).

FWIW I didn't see any other examples of this when going through logs.

Sam <div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Wed, Jan 3, 2024 at 3:37 PM Junio C Hamano <gitster@pobox.com> wrote: </div><blockquote class="gmail_quote" style="margin: 0px 0px 0px 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;">Sam Delmerico <<a href="mailto:delmerico@google.com" target="_blank">delmerico@google.com</a>> writes: > There were two region_enter events for negotiate_using_fetch instead of > one enter and one leave. This commit replaces the second region_enter > event with a region_leave. > > Signed-off-by: Sam Delmerico <<a href="mailto:delmerico@google.com" target="_blank">delmerico@google.com</a>> > --- >  fetch-pack.c | 2 +- >  1 file changed, 1 insertion(+), 1 deletion(-) Looks right, after skimming a29263cf (fetch-pack: add tracing for negotiation rounds, 2022-08-02).  Two questions and a half.  * How did you find it?  Code inspection?  While writing a script to    parse the output from around this area, your script noticed the    ever-increasing nesting level?  Something else?  * Would it be feasible to write some tests or tools that find    similar problems (semi-)automatically?  * Is the breakage (before this patch) something easily demonstrated    in a new test in t/ somewhere?  And if so, would it be worth    doing? Thanks.  Will queue. > > diff --git a/fetch-pack.c b/fetch-pack.c > index 31a72d43de..dba6d79944 100644 > --- a/fetch-pack.c > +++ b/fetch-pack.c > @@ -2218,7 +2218,7 @@ void negotiate_using_fetch(const struct oid_array *negotiation_tips, >                                          the_repository, "%d", >                                          negotiation_round); >       } > -     trace2_region_enter("fetch-pack", "negotiate_using_fetch", the_repository); > +     trace2_region_leave("fetch-pack", "negotiate_using_fetch", the_repository); >       trace2_data_intmax("negotiate_using_fetch", the_repository, >                          "total_rounds", negotiation_round); >       clear_common_flag(acked_commits); > > base-commit: a26002b62827b89a19b1084bd75d9371d565d03c </blockquote></div>

Junio C Hamano· Jan 5, 2024, 16:18 UTC · re: Sam Delmerico · lore

Re: [PATCH] push: region_leave trace for negotiate_using_fetch

Sam Delmerico <delmerico@google.com> writes:

Show 22 quoted lines

> * I don't exactly remember how I noticed it. I was doing some
> debugging around the push negotiation code and either 1) saw this
> region get entered twice in the trace output, or 2) I was just reading
> the code around here and saw two enters.
>
> * Perhaps there could be a check before the last git process ends that
> checks that all opened trace regions have been closed? I'm not sure
> how much work this would involve. It's probably also not a very
> proactive way to catch these bugs since it would only get triggered
> when a *user* hits a code path with a trace region that never exits.
>
> * There could also be a test that checks that every region_enter trace
> log has a corresponding region_leave. But I'm not sure how to ensure
> that every code path is checked.
>
> Overall, I'm not sure how much benefit there is from checking for
> this. I'm not sure that it would have a large impact if it were to
> happen again. For example, I think that it could be noticed relatively
> quickly by a person/system looking at metrics like I was (e.g. if the
> time spent in a region is infinite or zero).
>
> FWIW I didn't see any other examples of this when going through logs.

The above matches my intuition. A test that covers this specific case would likely be of low value as it is unlikely for us to regress this specific one. A CI job that runs all the tests under tracing and inspects the log may catch some but its finding would be limited to the code paths that are covered (but increasing the test coverage would help here, not just for finding unbalanced region markers, but for finding bugs in general).

Thanks.