++ed by:
KES

1 PAUSE user

Binary.com
and 1 contributors

NAME

Test::NoLeaks - Memory and file descriptor leak detector

SYNOPSYS

  use Test::NoLeaks;

  test_noleaks (
      code          => sub{
        # code that might leak
      },
      track_memory  => 1,
      track_fds     => 1,
      passes        => 2,
      warmup_passes => 1,
      tolerate_hits => 0,
  );

  Sample output:
  # pass 1, leaked: 225280 bytes 0 file descriptors
  # pass 36, leaked: 135168 bytes 0 file descriptors
  # pass 52, leaked: 319488 bytes 0 file descriptors
  # pass 84, leaked: 135168 bytes 0 file descriptors
  # pass 98, leaked: 155648 bytes 0 file descriptors
  not ok 1214 - Leaked 970752 bytes (5 hits) 0 file descriptors

  test_noleaks (
      code          => sub { ... },
      track_memory  => 1,
      passes        => 2,
  );

  # old-school way
  use Test::More;
  use Test::NoLeaks qw/noleaks/;
  ok noleaks(
      code          => sub { ... },
      track_memory  => ...,
      track_fds     => ...,
      passes        => ...,
      warmup_passes => ...,
    ), "non-leaked code description";

DESCRIPTION

It is hard to track memory leaks. There are a lot of perl modules (e.g. Test::LeakTrace), that try to detect and point leaks. Unfortunately, they do not always work, and they are rather limited because they are not able to detect leaks in XS-code or external libraries.

Instead of examining perl internals, this module offers a bit naive empirical approach: let the suspicious code to be launched in infinite loop some time and watch (via tools like top)if the memory consumption by perl process increses over time. If it does, while it is expected to be constant (stabilized), then, surely, there are leaks.

This approach is able only to detect and not able to point them. The module Test::NoLeaks implements the general idea of the approach, which might be enough in many cases.

INTERFACE

test_noleaks

noleaks

The mandatory hash has the following members

  • code

    Suspicious for leaks subroutine, that will be executed multiple times.

  • track_memory

  • track_fds

    Track memory or file descriptor leaks. At leas one of them should be specified.

    In Unices, every socket is file descriptor too, so, track_fds will be able to track unclosed sockets, i.e. network connections.

  • passes

    How many times code should be executed. If memory leak is too small, number of passes should be large enough to trigger additional pages allocation for perl process, and the leak will be detected.

    Page size is 4kb on linux, so, if code leaks 4 bytes on every pass, then 1024 passes should be specified.

    In general, the more passes are specified, the more chance to detect possible leaks.

    It is good idea to initally define passes to some large number, e.g. 10_000 to be sure, that the suspicious code leaks, but then decrease to some smaller number, enough to produce test fail report, i.e. enough to produces 3-5 memory hits (additional pages allocations). This will speed up tests execution and will save CO2 atmospheric emissions a little bit.

    Default value is 100. Minimal value is 2.

  • warmup_passes

    How many times the code should be executed before module starts tracking resources consumption on executing the code passes times.

    If you have caches, memoizes etc., then warmup_passes is your friend.

    Default value is 0.

  • tolerate_hits

    How many passes, which considered leaked, should be ingnored, i.e. maximal number of possible false leak reports.

    Even if your code has no leaks, it might cause perl interpreter allocate additional memory pages, e.g. due to memory fragmentation. Those allocations are legal, and should not be treated as leaks.

    Use this only when memory leaks are already fixed, but there are still false leak reports from test_leak. This value expected to be small enough, i.e. 1 or 2. For additional assurance, please, increase passes value, if tolarate_hits is non-zero.

    Default value is 0.

MEMORY LEAKS TESTING TECHNIQUES

Test::NoLeaks can be used to test web applications for memory leaks.

Let's consider you have the following suspicious code

  sub might_leak {
    my $t = Test::Mojo->new('MyApp');
    $t->post_ok('/search.json' => form => {q => 'Perl'})
        ->status_is(200);
    ...;
  }

  test_noleaks (
      code          => \&might_leak,
      track_memory  => 1,
      track_fds     => 1,
      passes        => 1000,
  );

The might_leak subroutine isn't optimal for leak detection, because it mixes infrastructure-related code (application) with request code. Let's consider, that there is a leak: every request creates some data and puts it into application, but forgets to do clean up. As soon as the application is re-created on every pass, the leaked data might be destroyed together with the application, and leak might remain undetected.

So, the code under test should look much more production like, i.e.

  my $t = Test::Mojo->new('MyApp');
  ok($t);
  sub might_leak {
    $t->post_ok('/search.json' => form => {q => 'Perl'})
        ->status_is(200);
    ...;
  }

That way web-application is created only once, and leaks will be tracked on request-related code.

Anyway, might_leak still wrong, because it unintentionally leaks due to use of direct or indirect Test::More functions, like ok or post_ok. They should not be used; if you still need to assert, that might_leak works propertly, you can use BAIL_OUT subroutine, to cancel any further testing, e.g.

  sub might_leak {
    my $got = some_function_might_leak;
    my $expected = "some_value";
    BAIL_OUT('some_function_might_leak does not work propertly!')
      unless $got eq $expected;
  }

Please, do not use test_noleaks more then once per test file. Consider the following example:

  # (A)
  test_noleaks(
    code => &does_not_leak_but_consumes_large_amount_of_memory,
    ...,
  )

  # (B)
  test_noleaks(
    code => &leaks_but_consumes_small_amount_of_memory,
    ...
  )

In A-case OS already allocated large amount of memory for Perl interpreter. In case-B perl might just re-use them, without allocating new ones, and this will be false negative, i.e. memory leak might not be reported.

LIMITATIONS

  • Currently it works propertly only on Linux

    Patches or pull requests to support other OSes are welcome.

  • The module will not work propertly in forked child

    It seems a little bit strange to use test_noleaks or noleaks in forked child, but if you really need that, please, send PR.

SEE ALSO

Test::MemoryGrowth

SOURCE CODE

GitHub

AUTHOR

binary.com, <perl at binary.com>

BUGS

Please report any bugs or feature requests to https://github.com/binary-com/perl-Test-NoLeaks/issues.