Add mostrecent aggregation to Gauge (#967)

In the multiprocess mode, the process that exposes the metrics needs to
aggregate the samples from other processes. Gauge metric allows users to
choose the aggregation mode. This implements 'mostrecent' (and
'livemostrecent') mode where the last observed value is exposed.

In order to support this, the file format is expanded to store the
timestamps in addition to the values. The stored timestamps are read by
the reader process and it's used to find the latest value.

Closes #847

Consideration on the atomicity:

Previously, mmap_dict.py had a comment saying "We assume that reading
from an 8 byte aligned value is atomic". With this change, the value
write becomes a 16 bytes 8-byte aligned write. The code author tried to
find a basis on the original assumption, but couldn't find any.
According to write(2), **if a file descriptor is shared**, the write
becomes atomic. However, we do not share the file descriptors in the
current architecture.

Considering that Ruby implementation also does the same and hadn't seen
an issue with it, this write atomicity problem might be practically not
an issue.

See also:

* prometheus/client_ruby#172

  The approach and naming are taken from client_ruby.

* https://github.com/prometheus/client_golang/blob/v1.17.0/prometheus/metric.go#L149-L161

  client_golang has an API for setting timestamp already. It explains
  the use case for the timestamp beyond the client-local aggregation. In
  order to support the same use case in Python, further changes are
  needed.

Signed-off-by: Masaya Suzuki <[email protected]>

Author: draftcode

README.md

@@ -711,9 +711,10 @@ Gauges have several modes they can run in, which can be selected with the `multi
 - 'min': Return a single timeseries that is the minimum of the values of all processes (alive or dead).
 - 'max': Return a single timeseries that is the maximum of the values of all processes (alive or dead).
 - 'sum': Return a single timeseries that is the sum of the values of all processes (alive or dead).
+- 'mostrecent': Return a single timeseries that is the most recent value among all processes (alive or dead).
 
 Prepend 'live' to the beginning of the mode to return the same result but only considering living processes
-(e.g., 'liveall, 'livesum', 'livemax', 'livemin').
+(e.g., 'liveall, 'livesum', 'livemax', 'livemin', 'livemostrecent').
 
 ```python
 from prometheus_client import Gauge

prometheus_client/metrics.py

@@ -346,7 +346,8 @@ def f():
         d.set_function(lambda: len(my_dict))
     """
     _type = 'gauge'
-    _MULTIPROC_MODES = frozenset(('all', 'liveall', 'min', 'livemin', 'max', 'livemax', 'sum', 'livesum'))
+    _MULTIPROC_MODES = frozenset(('all', 'liveall', 'min', 'livemin', 'max', 'livemax', 'sum', 'livesum', 'mostrecent', 'livemostrecent'))
+    _MOST_RECENT_MODES = frozenset(('mostrecent', 'livemostrecent'))
 
     def __init__(self,
                  name: str,
@@ -357,7 +358,7 @@ def __init__(self,
                  unit: str = '',
                  registry: Optional[CollectorRegistry] = REGISTRY,
                  _labelvalues: Optional[Sequence[str]] = None,
-                 multiprocess_mode: Literal['all', 'liveall', 'min', 'livemin', 'max', 'livemax', 'sum', 'livesum'] = 'all',
+                 multiprocess_mode: Literal['all', 'liveall', 'min', 'livemin', 'max', 'livemax', 'sum', 'livesum', 'mostrecent', 'livemostrecent'] = 'all',
                  ):
         self._multiprocess_mode = multiprocess_mode
         if multiprocess_mode not in self._MULTIPROC_MODES:
@@ -373,6 +374,7 @@ def __init__(self,
             _labelvalues=_labelvalues,
         )
         self._kwargs['multiprocess_mode'] = self._multiprocess_mode
+        self._is_most_recent = self._multiprocess_mode in self._MOST_RECENT_MODES
 
     def _metric_init(self) -> None:
         self._value = values.ValueClass(
@@ -382,18 +384,25 @@ def _metric_init(self) -> None:
 
     def inc(self, amount: float = 1) -> None:
         """Increment gauge by the given amount."""
+        if self._is_most_recent:
+            raise RuntimeError("inc must not be used with the mostrecent mode")
         self._raise_if_not_observable()
         self._value.inc(amount)
 
     def dec(self, amount: float = 1) -> None:
         """Decrement gauge by the given amount."""
+        if self._is_most_recent:
+            raise RuntimeError("dec must not be used with the mostrecent mode")
         self._raise_if_not_observable()
         self._value.inc(-amount)
 
     def set(self, value: float) -> None:
         """Set gauge to the given value."""
         self._raise_if_not_observable()
-        self._value.set(float(value))
+        if self._is_most_recent:
+            self._value.set(float(value), timestamp=time.time())
+        else:
+            self._value.set(float(value))
 
     def set_to_current_time(self) -> None:
         """Set gauge to the current unixtime."""

prometheus_client/mmap_dict.py

@@ -6,25 +6,26 @@
 
 _INITIAL_MMAP_SIZE = 1 << 16
 _pack_integer_func = struct.Struct(b'i').pack
-_pack_double_func = struct.Struct(b'd').pack
+_pack_two_doubles_func = struct.Struct(b'dd').pack
 _unpack_integer = struct.Struct(b'i').unpack_from
-_unpack_double = struct.Struct(b'd').unpack_from
+_unpack_two_doubles = struct.Struct(b'dd').unpack_from
 
 
 # struct.pack_into has atomicity issues because it will temporarily write 0 into
 # the mmap, resulting in false reads to 0 when experiencing a lot of writes.
 # Using direct assignment solves this issue.
 
-def _pack_double(data, pos, value):
-    data[pos:pos + 8] = _pack_double_func(value)
+
+def _pack_two_doubles(data, pos, value, timestamp):
+    data[pos:pos + 16] = _pack_two_doubles_func(value, timestamp)
 
 
 def _pack_integer(data, pos, value):
     data[pos:pos + 4] = _pack_integer_func(value)
 
 
 def _read_all_values(data, used=0):
-    """Yield (key, value, pos). No locking is performed."""
+    """Yield (key, value, timestamp, pos). No locking is performed."""
 
     if used <= 0:
         # If not valid `used` value is passed in, read it from the file.
@@ -41,9 +42,9 @@ def _read_all_values(data, used=0):
         encoded_key = data[pos:pos + encoded_len]
         padded_len = encoded_len + (8 - (encoded_len + 4) % 8)
         pos += padded_len
-        value = _unpack_double(data, pos)[0]
-        yield encoded_key.decode('utf-8'), value, pos
-        pos += 8
+        value, timestamp = _unpack_two_doubles(data, pos)
+        yield encoded_key.decode('utf-8'), value, timestamp, pos
+        pos += 16
 
 
 class MmapedDict:
@@ -53,7 +54,8 @@ class MmapedDict:
     Then 4 bytes of padding.
     There's then a number of entries, consisting of a 4 byte int which is the
     size of the next field, a utf-8 encoded string key, padding to a 8 byte
-    alignment, and then a 8 byte float which is the value.
+    alignment, and then a 8 byte float which is the value and a 8 byte float
+    which is a UNIX timestamp in seconds.
 
     Not thread safe.
     """
@@ -76,7 +78,7 @@ def __init__(self, filename, read_mode=False):
             _pack_integer(self._m, 0, self._used)
         else:
             if not read_mode:
-                for key, _, pos in self._read_all_values():
+                for key, _, _, pos in self._read_all_values():
                     self._positions[key] = pos
 
     @staticmethod
@@ -95,7 +97,7 @@ def _init_value(self, key):
         encoded = key.encode('utf-8')
         # Pad to be 8-byte aligned.
         padded = encoded + (b' ' * (8 - (len(encoded) + 4) % 8))
-        value = struct.pack(f'i{len(padded)}sd'.encode(), len(encoded), padded, 0.0)
+        value = struct.pack(f'i{len(padded)}sdd'.encode(), len(encoded), padded, 0.0, 0.0)
         while self._used + len(value) > self._capacity:
             self._capacity *= 2
             self._f.truncate(self._capacity)
@@ -105,30 +107,28 @@ def _init_value(self, key):
         # Update how much space we've used.
         self._used += len(value)
         _pack_integer(self._m, 0, self._used)
-        self._positions[key] = self._used - 8
+        self._positions[key] = self._used - 16
 
     def _read_all_values(self):
         """Yield (key, value, pos). No locking is performed."""
         return _read_all_values(data=self._m, used=self._used)
 
     def read_all_values(self):
-        """Yield (key, value). No locking is performed."""
-        for k, v, _ in self._read_all_values():
-            yield k, v
+        """Yield (key, value, timestamp). No locking is performed."""
+        for k, v, ts, _ in self._read_all_values():
+            yield k, v, ts
 
     def read_value(self, key):
         if key not in self._positions:
             self._init_value(key)
         pos = self._positions[key]
-        # We assume that reading from an 8 byte aligned value is atomic
-        return _unpack_double(self._m, pos)[0]
+        return _unpack_two_doubles(self._m, pos)
 
-    def write_value(self, key, value):
+    def write_value(self, key, value, timestamp):
         if key not in self._positions:
             self._init_value(key)
         pos = self._positions[key]
-        # We assume that writing to an 8 byte aligned value is atomic
-        _pack_double(self._m, pos, value)
+        _pack_two_doubles(self._m, pos, value, timestamp)
 
     def close(self):
         if self._f:

prometheus_client/multiprocess.py

@@ -68,7 +68,7 @@ def _parse_key(key):
                     # the file is missing
                     continue
                 raise
-            for key, value, _ in file_values:
+            for key, value, timestamp, _ in file_values:
                 metric_name, name, labels, labels_key, help_text = _parse_key(key)
 
                 metric = metrics.get(metric_name)
@@ -79,7 +79,7 @@ def _parse_key(key):
                 if typ == 'gauge':
                     pid = parts[2][:-3]
                     metric._multiprocess_mode = parts[1]
-                    metric.add_sample(name, labels_key + (('pid', pid),), value)
+                    metric.add_sample(name, labels_key + (('pid', pid),), value, timestamp)
                 else:
                     # The duplicates and labels are fixed in the next for.
                     metric.add_sample(name, labels_key, value)
@@ -89,6 +89,7 @@ def _parse_key(key):
     def _accumulate_metrics(metrics, accumulate):
         for metric in metrics.values():
             samples = defaultdict(float)
+            sample_timestamps = defaultdict(float)
             buckets = defaultdict(lambda: defaultdict(float))
             samples_setdefault = samples.setdefault
             for s in metric.samples:
@@ -105,6 +106,12 @@ def _accumulate_metrics(metrics, accumulate):
                             samples[without_pid_key] = value
                     elif metric._multiprocess_mode in ('sum', 'livesum'):
                         samples[without_pid_key] += value
+                    elif metric._multiprocess_mode in ('mostrecent', 'livemostrecent'):
+                        current_timestamp = sample_timestamps[without_pid_key]
+                        timestamp = float(timestamp or 0)
+                        if current_timestamp < timestamp:
+                            samples[without_pid_key] = value
+                            sample_timestamps[without_pid_key] = timestamp
                     else:  # all/liveall
                         samples[(name, labels)] = value
 

prometheus_client/values.py

@@ -19,7 +19,7 @@ def inc(self, amount):
         with self._lock:
             self._value += amount
 
-    def set(self, value):
+    def set(self, value, timestamp=None):
         with self._lock:
             self._value = value
 
@@ -82,7 +82,7 @@ def __reset(self):
                 files[file_prefix] = MmapedDict(filename)
             self._file = files[file_prefix]
             self._key = mmap_key(metric_name, name, labelnames, labelvalues, help_text)
-            self._value = self._file.read_value(self._key)
+            self._value, self._timestamp = self._file.read_value(self._key)
 
         def __check_for_pid_change(self):
             actual_pid = process_identifier()
@@ -99,13 +99,15 @@ def inc(self, amount):
             with lock:
                 self.__check_for_pid_change()
                 self._value += amount
-                self._file.write_value(self._key, self._value)
+                self._timestamp = 0.0
+                self._file.write_value(self._key, self._value, self._timestamp)
 
-        def set(self, value):
+        def set(self, value, timestamp=None):
             with lock:
                 self.__check_for_pid_change()
                 self._value = value
-                self._file.write_value(self._key, self._value)
+                self._timestamp = timestamp or 0.0
+                self._file.write_value(self._key, self._value, self._timestamp)
 
         def set_exemplar(self, exemplar):
             # TODO: Implement exemplars for multiprocess mode.

tests/test_multiprocess.py

@@ -185,6 +185,26 @@ def test_gauge_livesum(self):
         mark_process_dead(123, os.environ['PROMETHEUS_MULTIPROC_DIR'])
         self.assertEqual(2, self.registry.get_sample_value('g'))
 
+    def test_gauge_mostrecent(self):
+        g1 = Gauge('g', 'help', registry=None, multiprocess_mode='mostrecent')
+        values.ValueClass = MultiProcessValue(lambda: 456)
+        g2 = Gauge('g', 'help', registry=None, multiprocess_mode='mostrecent')
+        g2.set(2)
+        g1.set(1)
+        self.assertEqual(1, self.registry.get_sample_value('g'))
+        mark_process_dead(123, os.environ['PROMETHEUS_MULTIPROC_DIR'])
+        self.assertEqual(1, self.registry.get_sample_value('g'))
+
+    def test_gauge_livemostrecent(self):
+        g1 = Gauge('g', 'help', registry=None, multiprocess_mode='livemostrecent')
+        values.ValueClass = MultiProcessValue(lambda: 456)
+        g2 = Gauge('g', 'help', registry=None, multiprocess_mode='livemostrecent')
+        g2.set(2)
+        g1.set(1)
+        self.assertEqual(1, self.registry.get_sample_value('g'))
+        mark_process_dead(123, os.environ['PROMETHEUS_MULTIPROC_DIR'])
+        self.assertEqual(2, self.registry.get_sample_value('g'))
+
     def test_namespace_subsystem(self):
         c1 = Counter('c', 'help', registry=None, namespace='ns', subsystem='ss')
         c1.inc(1)
@@ -369,28 +389,28 @@ def setUp(self):
         self.d = mmap_dict.MmapedDict(self.tempfile)
 
     def test_process_restart(self):
-        self.d.write_value('abc', 123.0)
+        self.d.write_value('abc', 123.0, 987.0)
         self.d.close()
         self.d = mmap_dict.MmapedDict(self.tempfile)
-        self.assertEqual(123, self.d.read_value('abc'))
-        self.assertEqual([('abc', 123.0)], list(self.d.read_all_values()))
+        self.assertEqual((123, 987.0), self.d.read_value('abc'))
+        self.assertEqual([('abc', 123.0, 987.0)], list(self.d.read_all_values()))
 
     def test_expansion(self):
         key = 'a' * mmap_dict._INITIAL_MMAP_SIZE
-        self.d.write_value(key, 123.0)
-        self.assertEqual([(key, 123.0)], list(self.d.read_all_values()))
+        self.d.write_value(key, 123.0, 987.0)
+        self.assertEqual([(key, 123.0, 987.0)], list(self.d.read_all_values()))
 
     def test_multi_expansion(self):
         key = 'a' * mmap_dict._INITIAL_MMAP_SIZE * 4
-        self.d.write_value('abc', 42.0)
-        self.d.write_value(key, 123.0)
-        self.d.write_value('def', 17.0)
+        self.d.write_value('abc', 42.0, 987.0)
+        self.d.write_value(key, 123.0, 876.0)
+        self.d.write_value('def', 17.0, 765.0)
         self.assertEqual(
-            [('abc', 42.0), (key, 123.0), ('def', 17.0)],
+            [('abc', 42.0, 987.0), (key, 123.0, 876.0), ('def', 17.0, 765.0)],
             list(self.d.read_all_values()))
 
     def test_corruption_detected(self):
-        self.d.write_value('abc', 42.0)
+        self.d.write_value('abc', 42.0, 987.0)
         # corrupt the written data
         self.d._m[8:16] = b'somejunk'
         with self.assertRaises(RuntimeError):