Add support for parsing exemplars

Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>

Author: brian-brazil

prometheus_client/core.py

@@ -62,7 +62,7 @@ def __float__(self):
         return float(self.sec) + float(self.nsec) / 1e9
 
     def __eq__(self, other):
-        return self.sec == other.sec and self.nsec == other.nsec
+        return type(self) == type(other) and self.sec == other.sec and self.nsec == other.nsec
 
 
 Exemplar = namedtuple('Exemplar', ['labels', 'value', 'timestamp'])

prometheus_client/openmetrics/parser.py

@@ -49,6 +49,8 @@ def _unescape_help(text):
 
 def _parse_value(value):
     value = ''.join(value)
+    if value != value.strip():
+        raise ValueError("Invalid value: {0!r}".format(value))
     try:
         return int(value)
     except ValueError:
@@ -59,6 +61,8 @@ def _parse_timestamp(timestamp):
     timestamp = ''.join(timestamp)
     if not timestamp:
         return None
+    if timestamp != timestamp.strip():
+        raise ValueError("Invalid timestamp: {0!r}".format(timestamp))
     try:
         # Simple int.
         return core.Timestamp(int(timestamp), 0)
@@ -138,11 +142,13 @@ def _parse_sample(text):
     value = []
     timestamp = []
     labels = {}
+    exemplar_value = []
+    exemplar_timestamp = []
+    exemplar_labels = None
 
     state = 'name'
 
     it = iter(text)
-
     for char in it:
         if state == 'name':
             if char == '{':
@@ -159,23 +165,61 @@ def _parse_sample(text):
             else:
                 value.append(char)
         elif state == 'timestamp':
-            if char == ' ':
-                # examplars are not supported, halt
-                break
+            if char == '#' and not timestamp:
+                state = 'exemplarspace'
+            elif char == ' ':
+                state = 'exemplarhash'
             else:
                 timestamp.append(char)
+        elif state == 'exemplarhash':
+            if char == '#':
+                state = 'exemplarspace'
+            else:
+                raise ValueError("Invalid line: " + text)
+        elif state == 'exemplarspace':
+            if char == ' ':
+                state = 'exemplarstartoflabels'
+            else:
+                raise ValueError("Invalid line: " + text)
+        elif state == 'exemplarstartoflabels':
+            if char == '{':
+                exemplar_labels = _parse_labels(it, text)
+                # Space has already been parsed.
+                state = 'exemplarvalue'
+            else:
+                raise ValueError("Invalid line: " + text)
+        elif state == 'exemplarvalue':
+            if char == ' ':
+                state = 'exemplartimestamp'
+            else:
+                exemplar_value.append(char)
+        elif state == 'exemplartimestamp':
+            exemplar_timestamp.append(char)
 
     # Trailing space after value.
     if state == 'timestamp' and not timestamp:
         raise ValueError("Invalid line: " + text)
 
+    # Trailing space after value.
+    if state == 'exemplartimestamp' and not exemplar_timestamp:
+        raise ValueError("Invalid line: " + text)
+
+    # Incomplete exemplar.
+    if state in ['exemplarhash', 'exemplarspace', 'exemplarstartoflabels']:
+        raise ValueError("Invalid line: " + text)
+
     if not value:
         raise ValueError("Invalid line: " + text)
     value = ''.join(value)
     val = _parse_value(value)
     ts = _parse_timestamp(timestamp)
+    exemplar = None
+    if exemplar_labels is not None:
+        exemplar = core.Exemplar(exemplar_labels,
+                _parse_value(exemplar_value),
+                _parse_timestamp(exemplar_timestamp))
 
-    return core.Sample(''.join(name), labels, val, ts)
+    return core.Sample(''.join(name), labels, val, ts, exemplar)
 
 
 def text_fd_to_metric_families(fd):
@@ -206,6 +250,7 @@ def build_metric(name, documentation, typ, unit, samples):
         # TODO: Info and stateset can't have units
         # TODO: check samples are appropriately grouped and ordered
         # TODO: check for metadata in middle of samples
+        # TODO: Check histogram bucket rules being followed
         metric.samples = samples
         return metric
 

tests/openmetrics/test_parser.py

@@ -12,6 +12,7 @@
 from prometheus_client.core import (
     CollectorRegistry,
     CounterMetricFamily,
+    Exemplar,
     GaugeMetricFamily,
     HistogramMetricFamily,
     Metric,
@@ -105,6 +106,20 @@ def test_simple_histogram(self):
 """)
         self.assertEqual([HistogramMetricFamily("a", "help", sum_value=2, buckets=[("1", 0.0), ("+Inf", 3.0)])], list(families))
 
+    def test_histogram_exemplars(self):
+        families = text_string_to_metric_families("""# TYPE a histogram
+# HELP a help
+a_bucket{le="1"} 0 # {a="b"} 0.5
+a_bucket{le="2"} 2 123 # {a="c"} 0.5
+a_bucket{le="+Inf"} 3 # {a="d"} 4 123
+# EOF
+""")
+        hfm = HistogramMetricFamily("a", "help")
+        hfm.add_sample("a_bucket", {"le": "1"}, 0.0, None, Exemplar({"a": "b"}, 0.5))
+        hfm.add_sample("a_bucket", {"le": "2"}, 2.0, Timestamp(123, 0), Exemplar({"a": "c"}, 0.5)), 
+        hfm.add_sample("a_bucket", {"le": "+Inf"}, 3.0, None, Exemplar({"a": "d"}, 4, Timestamp(123, 0)))
+        self.assertEqual([hfm], list(families))
+
     def test_no_metadata(self):
         families = text_string_to_metric_families("""a 1
 # EOF
@@ -376,11 +391,19 @@ def test_invalid_input(self):
                 # Bad value.
                 ('a a\n# EOF\n'),
                 ('a  1\n# EOF\n'),
+                ('a 1\t\n# EOF\n'),
                 ('a 1 \n# EOF\n'),
                 # Bad timestamp.
                 ('a 1 z\n# EOF\n'),
                 ('a 1 1z\n# EOF\n'),
                 ('a 1 1.1.1\n# EOF\n'),
+                # Bad exemplars.
+                ('# TYPE a histogram\na_bucket{le="+Inf"} 1 #\n# EOF\n'),
+                ('# TYPE a histogram\na_bucket{le="+Inf"} 1# {} 1\n# EOF\n'),
+                ('# TYPE a histogram\na_bucket{le="+Inf"} 1 #{} 1\n# EOF\n'),
+                ('# TYPE a histogram\na_bucket{le="+Inf"} 1 # {}1\n# EOF\n'),
+                ('# TYPE a histogram\na_bucket{le="+Inf"} 1 # {} 1 \n# EOF\n'),
+                ('# TYPE a histogram\na_bucket{le="+Inf"} 1 # {} 1 1 \n# EOF\n'),
                 ]:
             with self.assertRaises(ValueError):
                 list(text_string_to_metric_families(case))