Re: [PATCH v2 0/6] reroll ma/plugleaks; more `object_array`-fixes

[Martin Ågren]

On 25 September 2017 at 18:08, Jeff King <p...@peff.net> wrote:
> On Sun, Sep 24, 2017 at 09:59:28PM +0200, Martin Ågren wrote:
>
>> > I'm not sure of the best way to count things.
>>
>
> But at least on the topic of "how many unique leaks are there", I wrote
> the script below to try to give some basic answers. It just finds the
> first non-boring entry in each stack trace and reports that. Where
> "boring" is really "this function is not expected to free, but hands off
> memory ownership to somebody else".

Thanks. I combined your script with this:

-- >8 --
#!/usr/bin/perl -w

# Extract the stacktraces and identify them
# by their SHA hashes (these identifiers are
# not guaranteed to be stable across
# re-compilations of the Git binaries).

use Digest::SHA qw(sha1 sha1_hex);

my $ctx = Digest::SHA->new("SHA-1");
my $stage = 0;

while (<>) {
  my $collect = 0;
  if ($stage == 0 && /irect leak of \d+ byte.*allocated from:$/) {
    $stage++;
    $collect = 1;
  } elsif($stage == 1 && /^\s*\#\d+\s+/) {
    $collect = 1;
  } elsif ($stage == 1 && /^\s*$/) {
    $digest = $ctx->hexdigest;
    printf "Stacktrace-hash: %s\n", $digest;
    $ctx = Digest::SHA->new("SHA-1");
    $stage = 0;
  } elsif ($stage == 1) {
    print "warning: unidentified string '$_'\n";
  }
  if ($collect) {
    $ctx->add_bits($_);
    print;
  }
}
-- >8 --

Then I report various ad-hoc metrics:

-- >8 --
#!/bin/bash

for d in "$@"
do
  echo $d

  echo -n "  direct leaks: "
  grep "Direct leak" "$d"/* | wc -l

  echo -n "  indirect leaks: "
  grep "Indirect leak" "$d"/* | wc -l

  echo -n "  allocating places: "
  perl leaks.pl "$d"/* | sort -u | wc -l

  echo -n "  most common allocating place: "
  perl leaks.pl "$d"/* | sort \
       | uniq -c | sort -nr | head -1 | awk '{print $1;}'

  echo -n "  size of leak-reports: "
  cat "$d"/* | wc -l

  echo -n "  unique leaking stacktraces: "
  perl extract-traces.pl "$d"/* | grep "Stacktrace-hash" | sort -u | wc -l

  echo -n "  most common stacktrace: "
  perl extract-traces.pl "$d"/* | grep "Stacktrace-hash" | sort \
       | uniq -c | sort -nr | head -1 | awk '{print $1;}'
done
-- >8 --

If PIDs of leaking processes collide, reports are lost. Something like
this as root helps: `echo 4194303 > /proc/sys/kernel/pid_max`

Still, the numbers vary for back-to-back runs. Here are two runs on
master and two runs on master plus the lockfile-patches I just sent.
(I don't run all tests.)

lsan_ea220ee4
  direct leaks: 127165
  indirect leaks: 83897
  allocating places: 504
  most common allocating place: 10212
  size of leak-reports: 3662204
  unique leaking stacktraces: 83265
  most common stacktrace: 55
lsan_ea220ee4-rerun
  direct leaks: 127172
  indirect leaks: 83903
  allocating places: 504
  most common allocating place: 10212
  size of leak-reports: 3662334
  unique leaking stacktraces: 83644
  most common stacktrace: 57
lsan_ea220ee4+lockfile_fixes
  direct leaks: 118678
  indirect leaks: 83908
  allocating places: 493
  most common allocating place: 10212
  size of leak-reports: 3545563
  unique leaking stacktraces: 99834
  most common stacktrace: 32
lsan_ea220ee4+lockfile_fixes-rerun
  direct leaks: 118678
  indirect leaks: 83902
  allocating places: 491
  most common allocating place: 10212
  size of leak-reports: 3545463
  unique leaking stacktraces: 82171
  most common stacktrace: 40

> So I don't know how useful any of that will be, but it at least should
> give _some_ metric that should be diminishing as we fix leaks.

Indeed.

Martin