patchtrace2: don't include "fsync" events in all trace2 logs

Fix the overly verbose trace2 logging added in 9a4987677d3 (trace2: add stats for fsync operations, 2022-03-30) (first released with v2.36.0).

Since that change every single "git" command invocation has included these "data" events, even though we'll only make use of these with core.fsyncMethod=batch, and even then only have non-zero values if we're writing object data to disk. See c0f4752ed2f (core.fsyncmethod: batched disk flushes for loose-objects, 2022-04-04) for that feature.

The result was that e.g. "git version" would previously emit a total
of 6 trace2 events with the GIT_TRACE2_EVENT target (version, start,
cmd_ancestry, cmd_name, exit, atexit), but afterwards would emit
8. We'd emit 2 "data" events before the "exit" event.

The reason we didn't catch this was that the trace2 unit tests added in a15860dca3f (trace2: t/helper/test-trace2, t0210.sh, t0211.sh, t0212.sh, 2019-02-22) would omit any "data" events that weren't the ones it cared about. Before this change to the C code 6/7 of our "t/t0212-trace2-event.sh" tests would fail if this change was applied to "t/t0212/parse_events.perl".

Let's make the trace2 testing more strict, and further append any new events types we don't know about in "t/t0212/parse_events.perl". Since we only invoke the "test-tool trace2" there's no guarantee that we'll catch other overly verbose events in the future, but we'll at least notice if we start emitting new events that are issues every time we log anything with trace2's JSON target.

We exclude the "data_json" event type, we'd otherwise would fail on both "win test" and "win+VS test" CI due to the logging added in 353d3d77f4f (trace2: collect Windows-specific process information, 2019-02-22). It looks like that logging should really be using trace2_cmd_ancestry() instead, which was introduced later in 2f732bf15e6 (tr2: log parent process name, 2021-07-21), but let's leave it for now.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
---

This is a fix for a v2.36.0 regression/change, so perhaps it's categorically out for v2.37.0, but given how isolated the fix is & how verbose the new trace2 data is perhaps it's worth including anyway.

Passing CI at: https://github.com/avar/git/actions/runs/2548732003

 t/t0212/parse_events.perl | 17 +++++++++++------
 wrapper.c                 |  3 +++
 2 files changed, 14 insertions(+), 6 deletions(-)

This patch looks good to me. 

Thanks, Neeraj

Ævar Arnfjörð Bjarmason  <avarab@gmail.com> writes:

The special casing is unfortunate, but if it is "let's leave it for now", probably we should leave a NEEDSWORK comment here to remind us, instead of giving a false impresion that abusing data_json is OK?

OK, both counters are unconditionally incremented whenever the corresponding action triggers, so this implementation is probably OK for now, but I suspect that it would become a maintenance burden if we were to add more fsync actions later.

If each of these is guarded with "show only when non-zero", that would become less problematic, though.

A bigger issue is when somebody is debugging the callers of git_fsync() wondering why they do not trigger. Getting log entries that say "no git_fsync() call with these actions have been made" may be useful (i.e. explicitly negative answer).

I am not sure if it is a good solution in the longer term, but this implementation may be a good-enough way to squelch the immediate loudness issue.

Thanks.

Fix the overly verbose trace2 logging added in 9a4987677d3 (trace2: add stats for fsync operations, 2022-03-30) (first released with v2.36.0).

Since that change every single "git" command invocation has included these "data" events, even though we'll only make use of these with core.fsyncMethod=batch, and even then only have non-zero values if we're writing object data to disk. See c0f4752ed2f (core.fsyncmethod: batched disk flushes for loose-objects, 2022-04-04) for that feature.

As we're needing to indent the trace2_data_intmax() lines let's introduce helper variables to ensure that our resulting lines (which were already too) don't exceed the recommendations of the CodingGuidelines. Doing that requires either wrapping them twice, or introducing short throwaway variable names, let's do the latter.

The result was that e.g. "git version" would previously emit a total
of 6 trace2 events with the GIT_TRACE2_EVENT target (version, start,
cmd_ancestry, cmd_name, exit, atexit), but afterwards would emit
8. We'd emit 2 "data" events before the "exit" event.

The reason we didn't catch this was that the trace2 unit tests added in a15860dca3f (trace2: t/helper/test-trace2, t0210.sh, t0211.sh, t0212.sh, 2019-02-22) would omit any "data" events that weren't the ones it cared about. Before this change to the C code 6/7 of our "t/t0212-trace2-event.sh" tests would fail if this change was applied to "t/t0212/parse_events.perl".

Let's make the trace2 testing more strict, and further append any new events types we don't know about in "t/t0212/parse_events.perl". Since we only invoke the "test-tool trace2" there's no guarantee that we'll catch other overly verbose events in the future, but we'll at least notice if we start emitting new events that are issues every time we log anything with trace2's JSON target.

We exclude the "data_json" event type, we'd otherwise would fail on both "win test" and "win+VS test" CI due to the logging added in 353d3d77f4f (trace2: collect Windows-specific process information, 2019-02-22). It looks like that logging should really be using trace2_cmd_ancestry() instead, which was introduced later in 2f732bf15e6 (tr2: log parent process name, 2021-07-21), but let's leave it for now.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
---

I think this re-roll addresses the concerns Junio brought up in: https://lore.kernel.org/git/xmqqh74byy19.fsf@gitster.g/; As to the larger question of:

On Thu, Jun 23 2022, Junio C Hamano wrote:

I think a better long-term solution in this case, per the initial feedback on the patch would be to have a generic facility to trace2 log how often we call a given syscall:

	https://lore.kernel.org/git/220330.86zgl7o8l8.gmgdl@evledraar.gmail.com/

I.e. in most (all?) other cases our trace2 logging is logging git's "business logic", but in this case the logging is functionally the same as a more dumb interception of an arbitrary system call, and then having some generic "syscall stats" trace2 event.

In this case such a generic logger would need to be a bit smarter than just a dumb wrapper, i.e. we'd need:

    fsync() != EINTR /* due to fsync_loop() */
    sync_file_range()

That still wouldn't be the same as the current logging if implemented correctly, but I think in a good way. E.g. now we'll log a failed fsync() the same as a successful one, as we increment the counter before seeing what the return value was.

But anyway, as noted in the original thread Neeraj didn't have time to look into that, neither do I, and the patch is now a part of git. So let's solve the immediate issue of verbose logging, and leave any other improvements for the future.

Range-diff against v1:
1:  df87e515efd ! 1:  a1fc37de947 trace2: don't include "fsync" events in all trace2 logs
    @@ Commit message
         we're writing object data to disk. See c0f4752ed2f (core.fsyncmethod:
         batched disk flushes for loose-objects, 2022-04-04) for that feature.
     
    +    As we're needing to indent the trace2_data_intmax() lines let's
    +    introduce helper variables to ensure that our resulting lines (which
    +    were already too) don't exceed the recommendations of the
    +    CodingGuidelines. Doing that requires either wrapping them twice, or
    +    introducing short throwaway variable names, let's do the latter.
    +
         The result was that e.g. "git version" would previously emit a total
         of 6 trace2 events with the GIT_TRACE2_EVENT target (version, start,
         cmd_ancestry, cmd_name, exit, atexit), but afterwards would emit
    @@ t/t0212/parse_events.perl
     +    }
     +
     +    elsif ($event eq 'data_json') {
    -+	# Ignore due to compat/win32/trace2_win32_process_info.c
    ++	# NEEDSWORK: Ignore due to
    ++	# compat/win32/trace2_win32_process_info.c, which should log a
    ++	# "cmd_ancestry" event instead.
     +    }
     +
     +    else {
    @@ wrapper.c: int git_fsync(int fd, enum fsync_action action)
      
      void trace_git_fsync_stats(void)
      {
    -+	if (!count_fsync_writeout_only && !count_fsync_hardware_flush)
    -+		return;
    +-	trace2_data_intmax("fsync", the_repository, "fsync/writeout-only", count_fsync_writeout_only);
    +-	trace2_data_intmax("fsync", the_repository, "fsync/hardware-flush", count_fsync_hardware_flush);
    ++	const struct repository *r = the_repository;
    ++	const intmax_t cfwo = count_fsync_writeout_only;
    ++	const intmax_t cfhf = count_fsync_hardware_flush;
     +
    - 	trace2_data_intmax("fsync", the_repository, "fsync/writeout-only", count_fsync_writeout_only);
    - 	trace2_data_intmax("fsync", the_repository, "fsync/hardware-flush", count_fsync_hardware_flush);
    ++	if (cfwo)
    ++		trace2_data_intmax("fsync", r, "fsync/writeout-only", cfwo);
    ++	if (cfhf)
    ++		trace2_data_intmax("fsync", r, "fsync/hardware-flush", cfhf);
      }
    + 
    + static int warn_if_unremovable(const char *op, const char *file, int rc)

 t/t0212/parse_events.perl | 19 +++++++++++++------
 wrapper.c                 | 10 ++++++++--
 2 files changed, 21 insertions(+), 8 deletions(-)

Ævar Arnfjörð Bjarmason  <avarab@gmail.com> writes:

too -> too something

This does read better.

I would have wrapped the lines instead of introducing these extra variables; the key observation is that it would make a much easier pattern to follow for a future developer who needs to add the third kind of "fsync" on top of the existing wo and hf.

But we can address that when somebody actually adds the third one, so let's take the patch as-is.

Thanks.

Fix the overly verbose trace2 logging added in 9a4987677d3 (trace2: add stats for fsync operations, 2022-03-30) (first released with v2.36.0).

Since that change every single "git" command invocation has included these "data" events, even though we'll only make use of these with core.fsyncMethod=batch, and even then only have non-zero values if we're writing object data to disk. See c0f4752ed2f (core.fsyncmethod: batched disk flushes for loose-objects, 2022-04-04) for that feature.

As we're needing to indent the trace2_data_intmax() lines let's introduce helper variables to ensure that our resulting lines (which were already too) don't exceed the recommendations of the CodingGuidelines. Doing that requires either wrapping them twice, or introducing short throwaway variable names, let's do the latter.

The result was that e.g. "git version" would previously emit a total
of 6 trace2 events with the GIT_TRACE2_EVENT target (version, start,
cmd_ancestry, cmd_name, exit, atexit), but afterwards would emit
8. We'd emit 2 "data" events before the "exit" event.

The reason we didn't catch this was that the trace2 unit tests added in a15860dca3f (trace2: t/helper/test-trace2, t0210.sh, t0211.sh, t0212.sh, 2019-02-22) would omit any "data" events that weren't the ones it cared about. Before this change to the C code 6/7 of our "t/t0212-trace2-event.sh" tests would fail if this change was applied to "t/t0212/parse_events.perl".

Let's make the trace2 testing more strict, and further append any new events types we don't know about in "t/t0212/parse_events.perl". Since we only invoke the "test-tool trace2" there's no guarantee that we'll catch other overly verbose events in the future, but we'll at least notice if we start emitting new events that are issues every time we log anything with trace2's JSON target.

We exclude the "data_json" event type, we'd otherwise would fail on both "win test" and "win+VS test" CI due to the logging added in 353d3d77f4f (trace2: collect Windows-specific process information, 2019-02-22). It looks like that logging should really be using trace2_cmd_ancestry() instead, which was introduced later in 2f732bf15e6 (tr2: log parent process name, 2021-07-21), but let's leave it for now.

The fix-up to aaf81223f48 (unpack-objects: use stream_loose_object() to unpack large objects, 2022-06-11) is needed because we're changing the behavior of these events as discussed above. Since we'd always emit a "hardware-flush" event the test added in aaf81223f48 wasn't testing anything except that this trace2 data was unconditionally logged. Even if "core.fsyncMethod" wasn't set to "batch" we'd pass the test.

Now we'll check the expected number of "writeout" v.s. "flush" calls under "core.fsyncMethod=batch", but note that this doesn't actually test if we carried out the sync using that method, on a platform where we'd have to fall back to fsync() each of those "writeout" would really be a "flush" (i.e. a full fsync()).

But in this case what we're testing is that the logic in "unpack-objects" behaves as expected, not the OS-specific question of whether we actually were able to use the "bulk" method.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
---

This v3:

 * Fixes the failing tests, due to a merge with
   hx/unpack-streaming. See details at:
   https://lore.kernel.org/git/220715.86bktqzdb8.gmgdl@evledraar.gmail.com/

   We now test how many "writeout" and "flush" sync events we emit,
   rather than the (meaningless) previous behavior (see above)>

 * Creates a log_trace_fsync_if() helper in wrapper.c, to avoid both
   the duplication and line-wrapping issues previous rounds tried to
   address more verbosely.

Range-diff against v2:
1:  a1fc37de947 ! 1:  979dea5956a trace2: don't include "fsync" events in all trace2 logs
    @@ Metadata
     Author: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
     
      ## Commit message ##
    -    trace2: don't include "fsync" events in all trace2 logs
    +    trace2: only include "fsync" events if we git_fsync()
     
         Fix the overly verbose trace2 logging added in 9a4987677d3 (trace2:
         add stats for fsync operations, 2022-03-30) (first released with
    @@ Commit message
         2f732bf15e6 (tr2: log parent process name, 2021-07-21), but let's
         leave it for now.
     
    +    The fix-up to aaf81223f48 (unpack-objects: use stream_loose_object()
    +    to unpack large objects, 2022-06-11) is needed because we're changing
    +    the behavior of these events as discussed above. Since we'd always
    +    emit a "hardware-flush" event the test added in aaf81223f48 wasn't
    +    testing anything except that this trace2 data was unconditionally
    +    logged. Even if "core.fsyncMethod" wasn't set to "batch" we'd pass the
    +    test.
    +
    +    Now we'll check the expected number of "writeout" v.s. "flush" calls
    +    under "core.fsyncMethod=batch", but note that this doesn't actually
    +    test if we carried out the sync using that method, on a platform where
    +    we'd have to fall back to fsync() each of those "writeout" would
    +    really be a "flush" (i.e. a full fsync()).
    +
    +    But in this case what we're testing is that the logic in
    +    "unpack-objects" behaves as expected, not the OS-specific question of
    +    whether we actually were able to use the "bulk" method.
    +
         Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
     
      ## t/t0212/parse_events.perl ##
    @@ t/t0212/parse_events.perl
      
          # This trace2 target does not emit 'printf' events.
     
    + ## t/t5351-unpack-large-objects.sh ##
    +@@ t/t5351-unpack-large-objects.sh: test_expect_success 'unpack big object in stream' '
    + 	test_dir_is_empty dest.git/objects/pack
    + '
    + 
    ++check_fsync_events () {
    ++	local trace="$1" &&
    ++	shift &&
    ++
    ++	cat >expect &&
    ++	sed -n \
    ++		-e '/^{"event":"data",.*"category":"fsync",/ {
    ++			s/.*"category":"fsync",//;
    ++			s/}$//;
    ++			p;
    ++		}' \
    ++		<"$trace" >actual &&
    ++	test_cmp expect actual
    ++}
    ++
    + BATCH_CONFIGURATION='-c core.fsync=loose-object -c core.fsyncmethod=batch'
    + 
    + test_expect_success 'unpack big object in stream (core.fsyncmethod=batch)' '
    + 	prepare_dest 1m &&
    + 	GIT_TRACE2_EVENT="$(pwd)/trace2.txt" \
    ++	GIT_TEST_FSYNC=true \
    + 		git -C dest.git $BATCH_CONFIGURATION unpack-objects <pack-$PACK.pack &&
    +-	grep fsync/hardware-flush trace2.txt &&
    ++	check_fsync_events trace2.txt <<-\EOF &&
    ++	"key":"fsync/writeout-only","value":"6"
    ++	"key":"fsync/hardware-flush","value":"1"
    ++	EOF
    ++
    + 	test_dir_is_empty dest.git/objects/pack &&
    + 	git -C dest.git cat-file --batch-check="%(objectname)" <obj-list >current &&
    + 	cmp obj-list current
    +
      ## wrapper.c ##
     @@ wrapper.c: int git_fsync(int fd, enum fsync_action action)
    + 	}
    + }
      
    ++static void log_trace_fsync_if(const char *key, intmax_t value)
    ++{
    ++	if (value)
    ++		trace2_data_intmax("fsync", the_repository, key, value);
    ++}
    ++
      void trace_git_fsync_stats(void)
      {
     -	trace2_data_intmax("fsync", the_repository, "fsync/writeout-only", count_fsync_writeout_only);
     -	trace2_data_intmax("fsync", the_repository, "fsync/hardware-flush", count_fsync_hardware_flush);
    -+	const struct repository *r = the_repository;
    -+	const intmax_t cfwo = count_fsync_writeout_only;
    -+	const intmax_t cfhf = count_fsync_hardware_flush;
    -+
    -+	if (cfwo)
    -+		trace2_data_intmax("fsync", r, "fsync/writeout-only", cfwo);
    -+	if (cfhf)
    -+		trace2_data_intmax("fsync", r, "fsync/hardware-flush", cfhf);
    ++	log_trace_fsync_if("fsync/writeout-only", count_fsync_writeout_only);
    ++	log_trace_fsync_if("fsync/hardware-flush", count_fsync_hardware_flush);
      }
      
      static int warn_if_unremovable(const char *op, const char *file, int rc)

 t/t0212/parse_events.perl       | 19 +++++++++++++------
 t/t5351-unpack-large-objects.sh | 22 +++++++++++++++++++++-
 wrapper.c                       | 10 ++++++++--
 3 files changed, 42 insertions(+), 9 deletions(-)

Ævar Arnfjörð Bjarmason  <avarab@gmail.com> writes:

Or a macro that can be used like

	log_trace_fsync_nonzero(writeout_only);
	log_trace_fsync_nonzero(hardware_flush);

something along the lines of

#define log_trace_fsync_nonzero(var) \
	if (count_fsync_ #var) \
		trace2_data_intmax("fsync", the_repository, \
			"fsync/" ##var, count_fsync_ #var)

using token pasting and token stringification ;-)

The macro itself I am not serious at all. s/_if/_nonzero/ may be worth pondering, though.

Makes sense.

Neeraj, Ack, as the author of 9a498767 (trace2: add stats for fsync operations, 2022-03-30)?

Thanks all.

Yes, ACK.  Looks fine to me.  This is still enough info to diagnose if the configuration is being parsed properly and if we're going down the right codepaths.

Thanks, Neeraj