Memory leak test: take fluctuations into account (#1757)

Preamble
=======

We have a [memory leak test suite](https://github.com/giampaolo/psutil/blob/e1ea2bccf8aea404dca0f79398f36f37217c45f6/psutil/tests/__init__.py#L897), which calls a function many times and fails if the process memory increased. We do this in order to detect missing `free()` or `Py_DECREF` calls in the C modules. When we do, then we have a memory leak.

The problem
==========

A problem we've been having for probably over 10 years, is the false positives. That's because the memory fluctuates. Sometimes it may increase (or even decrease!) due to how the OS handles memory, the Python's garbage collector, the fact that RSS is an approximation and who knows what else. So thus far we tried to compensate that by using the following logic:
- warmup (call fun 10 times)
- call the function many times (1000)
- if memory increased before/after calling function 1000 times, then keep calling it for another 3 secs
- if it still increased at all (> 0) then fail

This logic didn't really solve the problem, as we still had occasional false positives, especially lately on FreeBSD. 

The solution
=========

This PR changes the internal algorithm so that in case of failure (mem > 0 after calling fun() N times) we retry the test for up to 5 times, increasing N (repetitions) each time, so we consider it a failure only if the memory **keeps increasing** between runs. So for instance, here's a legitimate failure:

```
psutil.tests.test_memory_leaks.TestModuleFunctionsLeaks.test_disk_partitions ... 
Run #1: extra-mem=696.0K, per-call=3.5K, calls=200
Run #2: extra-mem=1.4M, per-call=3.5K, calls=400
Run #3: extra-mem=2.1M, per-call=3.5K, calls=600
Run #4: extra-mem=2.7M, per-call=3.5K, calls=800
Run #5: extra-mem=3.4M, per-call=3.5K, calls=1000
FAIL
```

If, on the other hand, the memory increased on one run (say 200 calls) but decreased on the next run (say 400 calls), then it clearly means it's a false positive, because memory consumption may be > 0 on second run, but if it's lower than the previous run with less repetitions, then it cannot possibly represent a leak (just a fluctuation):

```
psutil.tests.test_memory_leaks.TestModuleFunctionsLeaks.test_net_connections ... 
Run #1: extra-mem=568.0K, per-call=2.8K, calls=200
Run #2: extra-mem=24.0K, per-call=61.4B, calls=400
OK
```

Note about mallinfo()
================

Aka #1275. `mallinfo()` on Linux is supposed to provide memory metrics about how many bytes gets allocated on the heap by `malloc()`, so it's supposed to be way more precise than RSS and also [USS](http://grodola.blogspot.com/2016/02/psutil-4-real-process-memory-and-environ.html). In another branch were I exposed it, I verified that fluctuations still occur even when using `mallinfo()` though, despite less often. So that means even `mallinfo()` would not grant 100% stability.

1 files changed, 59 insertions, 76 deletions

diff --git a/psutil/tests/__init__.py b/psutil/tests/__init__.py
index 41a791c1..b452775f 100644
--- a/psutil/tests/__init__.py
+++ b/psutil/tests/__init__.py
@@ -476,12 +476,16 @@ def terminate(proc_or_pid, sig=signal.SIGTERM, wait_timeout=GLOBAL_TIMEOUT):
         from psutil._psposix import wait_pid
 
     def wait(proc, timeout):
-        try:
-            return psutil.Process(proc.pid).wait(timeout)
-        except psutil.NoSuchProcess:
-            # Needed to kill zombies.
-            if POSIX:
-                return wait_pid(proc.pid, timeout)
+        if isinstance(proc, subprocess.Popen) and not PY3:
+            proc.wait()
+        else:
+            proc.wait(timeout)
+        if WINDOWS and isinstance(proc, subprocess.Popen):
+            # Otherwise PID may still hang around.
+            try:
+                return psutil.Process(proc.pid).wait(timeout)
+            except psutil.NoSuchProcess:
+                pass
 
     def sendsig(proc, sig):
         # If the process received SIGSTOP, SIGCONT is necessary first,
@@ -899,22 +903,15 @@ class TestMemoryLeak(PsutilTestCase):
     typically functions implemented in C which forgot to free() memory
     from the heap. It does so by checking whether the process memory
     usage increased before and after calling the function many times.
-    The logic:
-
-        call_fun_n_times()
-        if mem_diff > tolerance:
-            call_fun_for_3_secs()
-            if mem_diff > 0:
-                return 1  # failure
-        return 0  # success
 
     Note that this is hard (probably impossible) to do reliably, due
     to how the OS handles memory, the GC and so on (memory can even
-    decrease!). In order to avoid false positives you should adjust the
-    tolerance of each individual test case, but most of the times you
-    won't have to.
+    decrease!). In order to avoid false positives, in case of failure
+    (mem > 0) we retry the test for up to 5 times, increasing call
+    repetitions each time. If the memory keeps increasing then it's a
+    failure.
 
-    If available (Linux, OSX, Windows) USS memory is used for comparison,
+    If available (Linux, OSX, Windows), USS memory is used for comparison,
     since it's supposed to be more precise, see:
     http://grodola.blogspot.com/2016/02/psutil-4-real-process-memory-and-environ.html
     If not, RSS memory is used. mallinfo() on Linux and _heapwalk() on
@@ -932,15 +929,15 @@ class TestMemoryLeak(PsutilTestCase):
                 self.execute(some_function)
     """
     # Configurable class attrs.
-    times = 1000
+    times = 200
     warmup_times = 10
-    tolerance = 4096  # memory
-    retry_for = 3.0  # seconds
+    retries = 5
+    tolerance = 0  # memory
     verbose = True
 
-    def setUp(self):
-        self._thisproc = psutil.Process()
-        gc.collect()
+    @classmethod
+    def setUpClass(cls):
+        cls._thisproc = psutil.Process()
 
     def _get_mem(self):
         # USS is the closest thing we have to "real" memory usage and it
@@ -957,77 +954,63 @@ class TestMemoryLeak(PsutilTestCase):
     def _call(self, fun):
         return fun()
 
-    def _itercall(self, fun, iterator):
+    def _call_ntimes(self, fun, times):
         """Get 2 distinct memory samples, before and after having
         called fun repeadetly, and return the memory difference.
         """
-        ncalls = 0
-        gc.collect()
+        gc.collect(generation=1)
         mem1 = self._get_mem()
-        for x in iterator:
+        for x in range(times):
             ret = self._call(fun)
-            ncalls += 1
             del x, ret
-        gc.collect()
+        gc.collect(generation=1)
         mem2 = self._get_mem()
         self.assertEqual(gc.garbage, [])
-        diff = mem2 - mem1
-        if diff < 0:
-            self._log("negative memory diff -%s" % (bytes2human(abs(diff))))
-        return (diff, ncalls)
-
-    def _call_ntimes(self, fun, times):
-        return self._itercall(fun, range(times))[0]
-
-    def _call_for(self, fun, secs):
-        def iterator(secs):
-            stop_at = time.time() + secs
-            while time.time() < stop_at:
-                yield
-        return self._itercall(fun, iterator(secs))
+        diff = mem2 - mem1  # can also be negative
+        return diff
 
     def _log(self, msg):
         if self.verbose:
             print_color(msg, color="yellow", file=sys.stderr)
 
-    def execute(self, fun, times=times, warmup_times=warmup_times,
-                tolerance=tolerance, retry_for=retry_for):
+    def execute(self, fun, times=None, warmup_times=None, retries=None,
+                tolerance=None):
         """Test a callable."""
-        if times <= 0:
-            raise ValueError("times must be > 0")
-        if warmup_times < 0:
-            raise ValueError("warmup_times must be >= 0")
-        if tolerance is not None and tolerance < 0:
-            raise ValueError("tolerance must be >= 0")
-        if retry_for is not None and retry_for < 0:
-            raise ValueError("retry_for must be >= 0")
+        times = times if times is not None else self.times
+        warmup_times = warmup_times if warmup_times is not None \
+            else self.warmup_times
+        retries = retries if retries is not None else self.retries
+        tolerance = tolerance if tolerance is not None else self.tolerance
+        try:
+            assert times >= 1, "times must be >= 1"
+            assert warmup_times >= 0, "warmup_times must be >= 0"
+            assert retries >= 0, "retries must be >= 0"
+            assert tolerance >= 0, "tolerance must be >= 0"
+        except AssertionError as err:
+            raise ValueError(str(err))
 
         # warm up
         self._call_ntimes(fun, warmup_times)
-        mem1 = self._call_ntimes(fun, times)
-
-        if mem1 > tolerance:
-            # This doesn't necessarily mean we have a leak yet.
-            # At this point we assume that after having called the
-            # function so many times the memory usage is stabilized
-            # and if there are no leaks it should not increase
-            # anymore. Let's keep calling fun for N more seconds and
-            # fail if we notice any difference (ignore tolerance).
-            msg = "+%s after %s calls; try calling fun for another %s secs" % (
-                bytes2human(mem1), times, retry_for)
-            if not retry_for:
-                raise self.fail(msg)
+        messages = []
+        prev_mem = 0
+        increase = times
+        for idx in range(1, retries + 1):
+            mem = self._call_ntimes(fun, times)
+            msg = "Run #%s: extra-mem=%s, per-call=%s, calls=%s" % (
+                idx, bytes2human(mem), bytes2human(mem / times), times)
+            messages.append(msg)
+            success = mem <= tolerance or mem < prev_mem
+            if success:
+                if idx > 1:
+                    self._log(msg)
+                return
             else:
+                if idx == 1:
+                    print()  # NOQA
                 self._log(msg)
-
-            mem2, ncalls = self._call_for(fun, retry_for)
-            if mem2 > mem1:
-                # failure
-                msg = "+%s memory increase after %s calls; " % (
-                    bytes2human(mem1), times)
-                msg += "+%s after another %s calls over %s secs" % (
-                    bytes2human(mem2), ncalls, retry_for)
-                raise self.fail(msg)
+                times += increase
+                prev_mem = mem
+        raise self.fail(". ".join(messages))
 
     def execute_w_exc(self, exc, fun, **kwargs):
         """Convenience method to test a callable while making sure it