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

Re: [PATCH] trace2: intercept all common signals

From
Junio C Hamano <gitster@pobox.com>
Date
May 10, 2024, 18:47 UTC
Message-ID
<xmqqv83l4i86.fsf@gitster.g>
In-Reply-To
<20240510172243.3529851-1-emilyshaffer@google.com>
Emily Shaffer <emilyshaffer@google.com> writes:
Show 75 quoted lines
> From: Emily Shaffer <nasamuffin@google.com>
>
> We already use trace2 to find out about unexpected pipe breakages, which
> is nice for detecting bugs or system problems, by adding a handler for
> SIGPIPE which simply writes a trace2 line. However, there are a handful
> of other common signals which we might want to snoop on:
>
>  - SIGINT, SIGTERM, or SIGQUIT, when a user manually cancels a command in
>    frustration or mistake (via Ctrl-C, Ctrl-D, or `kill`)
>  - SIGHUP, when the network closes unexpectedly (indicating there may be
>    a problem to solve)
>
> There are lots more signals which we might find useful later, but at
> least let's teach trace2 to report these egregious ones. Conveniently,
> they're also already covered by the `_common` variants in sigchain.[ch].
>
> Sigchain itself is already tested via helper/test-sigchain.c, and trace2
> is tested in a number of places - let's also add tests demonstrating
> that sigchain + trace2 works correctly.
>
> Signed-off-by: Emily Shaffer <nasamuffin@google.com>
> ---
>  t/helper/test-trace2.c   | 17 +++++++++++++++++
>  t/t0210-trace2-normal.sh | 22 ++++++++++++++++++++++
>  trace2.c                 |  2 +-
>  3 files changed, 40 insertions(+), 1 deletion(-)
>
> diff --git a/t/helper/test-trace2.c b/t/helper/test-trace2.c
> index 1adac29a57..8970956ea8 100644
> --- a/t/helper/test-trace2.c
> +++ b/t/helper/test-trace2.c
> @@ -231,6 +231,22 @@ static int ut_010bug_BUG(int argc UNUSED, const char **argv UNUSED)
>  	BUG("a %s message", "BUG");
>  }
>  
> +static int ut_011signal(int argc, const char **argv)
> +{
> +	const char *usage_error = "expect <bool common>";
> +	int common = 0;
> +
> +	if (argc != 1 || get_i(&common, argv[0]))
> +		die("%s", usage_error);
> +
> +	/*
> +	 * There is no strong reason SIGSEGV is ignored by trace2 - it's just
> +	 * not included by sigchain_push_common().
> +	 */
> +	raise(common ? SIGINT : SIGSEGV);
> +	return 0; /*unreachable*/
> +}
> +
>  /*
>   * Single-threaded timer test.  Create several intervals using the
>   * TEST1 timer.  The test script can verify that an aggregate Trace2
> @@ -482,6 +498,7 @@ static struct unit_test ut_table[] = {
>  	{ ut_008bug,      "008bug",    "" },
>  	{ ut_009bug_BUG,  "009bug_BUG","" },
>  	{ ut_010bug_BUG,  "010bug_BUG","" },
> +	{ ut_011signal,   "011signal","" },
>  
>  	{ ut_100timer,    "100timer",  "<count> <ms_delay>" },
>  	{ ut_101timer,    "101timer",  "<count> <ms_delay> <threads>" },
> diff --git a/t/t0210-trace2-normal.sh b/t/t0210-trace2-normal.sh
> index c312657a12..c34ccc518c 100755
> --- a/t/t0210-trace2-normal.sh
> +++ b/t/t0210-trace2-normal.sh
> @@ -244,6 +244,28 @@ test_expect_success 'bug messages followed by BUG() are written to trace2' '
>  	test_cmp expect actual
>  '
>  
> +test_expect_success 'trace2 reports common signals' '
> +	test_when_finished "rm trace.normal actual" &&
> +
> +	# signals are fatal, so expect this to fail
> +	! env GIT_TRACE2="$(pwd)/trace.normal" test-tool trace2 011signal 1 &&

Is it deliberate that this does not use test_must_fail or is it an oversight? The same comment applies to all other uses of "! env".

We often see the use of "env" in conjunction with a test that is expected to fail because

    test_must_fail VAR=VAL cmd
simply does not work.  If you are not using test_expect_fail, then
    ! VAR=VAL cmd

should be sufficient, but it would mean that you will be happy even if the way "cmd" dies is not in a controlled way (e.g. due to receiving a signal).

Ah, perhaps that is it? Is "test-tool trace2 011signal 1" raise a signal to kill itself and after showing the event in the trace stream it is expected to die the narual death of receiving the same signal by re-raising it?

If that is what is happening here, not using test_must_fail is absolutely the right thing to do, but then I doubt you need "env" there. Also, if we know what signal is raised, then we should also know the exit status from this (i.e. signal number plus 128 or something) that we want to validate? I dunno.

Show 13 quoted lines
> diff --git a/trace2.c b/trace2.c
> index f894532d05..3692010f5d 100644
> --- a/trace2.c
> +++ b/trace2.c
> @@ -231,7 +231,7 @@ void trace2_initialize_fl(const char *file, int line)
>  	tr2_sid_get();
>  
>  	atexit(tr2main_atexit_handler);
> -	sigchain_push(SIGPIPE, tr2main_signal_handler);
> +	sigchain_push_common(tr2main_signal_handler);
>  	tr2tls_init();
>  
>  	/*
Previous: Emily ShafferNext: Emily Shaffer
Message 3 of 14 in “trace2: intercept all common signals”
  1. trace2: intercept all common signalsEmily Shaffer, May 10, 2024
  2. Emily ShafferMay 10, 2024
  3. Junio C HamanoMay 10, 2024
  4. Emily ShafferMay 10, 2024
  5. Jeff KingMay 10, 2024
  6. Emily ShafferMay 10, 2024
  7. Jeff KingMay 10, 2024
  8. Junio C HamanoMay 10, 2024
  9. Jeff KingMay 10, 2024
  10. Jeff KingMay 10, 2024
  11. Emily ShafferMay 13, 2024
  12. Jeff KingMay 16, 2024
  13. Junio C HamanoMay 16, 2024
  14. Jeff KingMay 23, 2024

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.