Check samples are grouped, untyped isn't used, and for invalid/missing le/quantile values.

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

Author: brian-brazil

prometheus_client/openmetrics/parser.py

@@ -228,7 +228,26 @@ def _parse_sample(text):
                 _parse_timestamp(exemplar_timestamp))
 
     return core.Sample(''.join(name), labels, val, ts, exemplar)
-    
+
+
+def _group_for_sample(sample, name, typ):
+    if typ == 'info':
+        # We can't distinguish between groups for info metrics.
+        return {}
+    if typ == 'summary' and sample.name == name:
+        d = sample.labels.copy()
+        del d['quantile']
+        return d
+    if typ == 'stateset':
+        d = sample.labels.copy()
+        del d[name]
+        return d
+    if typ in ['histogram', 'gaugehistogram'] and sample.name == name + '_bucket':
+        d = sample.labels.copy()
+        del d['le']
+        return d
+    return sample.labels
+
 
 def text_fd_to_metric_families(fd):
     """Parse Prometheus text format from a file descriptor.
@@ -240,9 +259,11 @@ def text_fd_to_metric_families(fd):
     Yields core.Metric's.
     """
     name = ''
-    documentation = ''
-    typ = 'untyped'
-    unit = ''
+    documentation = None
+    typ = None
+    unit = None
+    group = None
+    seen_groups = set()
     samples = []
     allowed_names = []
     eof = False
@@ -253,7 +274,7 @@ def build_metric(name, documentation, typ, unit, samples):
             raise ValueError("Duplicate metric: " + name)
         seen_metrics.add(name)
         if typ is None:
-            typ = 'untyped'
+            typ = 'unknown'
         if documentation is None:
             documentation = ''
         if unit is None:
@@ -264,7 +285,7 @@ def build_metric(name, documentation, typ, unit, samples):
             raise ValueError("Units not allowed for this metric type: " + name)
         metric = core.Metric(name, documentation, typ, unit)
         # TODO: check labelvalues are valid utf8
-        # TODO: check samples are appropriately grouped and ordered
+        # TODO: check samples are appropriately ordered
         # TODO: Check histogram bucket rules being followed
         # TODO: Check for dupliate samples
         # TODO: Check for decresing timestamps
@@ -294,6 +315,8 @@ def build_metric(name, documentation, typ, unit, samples):
                 unit = None
                 typ = None
                 documentation = None
+                group = None
+                seen_groups = set()
                 samples = []
                 allowed_names = [parts[2]]
 
@@ -308,6 +331,8 @@ def build_metric(name, documentation, typ, unit, samples):
                 if typ is not None:
                     raise ValueError("More than one TYPE for metric: " + line)
                 typ = parts[3]
+                if typ == 'untyped':
+                    raise ValueError("Invalid TYPE for metric: " + line)
                 allowed_names = {
                     'counter': ['_total', '_created'],
                     'summary': ['_count', '_sum', '', '_created'],
@@ -327,16 +352,33 @@ def build_metric(name, documentation, typ, unit, samples):
             if sample.name not in allowed_names:
                 if name != '':
                     yield build_metric(name, documentation, typ, unit, samples)
-                # Start an untyped metric.
+                # Start an unknown metric.
                 name = sample.name
-                documentation = ''
-                unit = ''
-                typ = 'untyped'
+                documentation = None
+                unit = None
+                typ = 'unknown'
                 samples = [sample]
+                group = None
+                seen_groups = set()
                 allowed_names = [sample.name]
             else:
                 samples.append(sample)
 
+            if typ == 'stateset' and name not in sample.labels:
+                raise ValueError("Stateset missing label: " + line)
+            if (typ in ['histogram', 'gaugehistogram'] and name + '_bucket' == sample.name
+                    and float(sample.labels.get('le', -1)) < 0):
+                raise ValueError("Invalid le label: " + line)
+            if (typ == 'summary' and name == sample.name
+                    and not (0 <= float(sample.labels.get('quantile', -1)) <= 1)):
+                raise ValueError("Invalid quantile label: " + line)
+
+            g = tuple(sorted(_group_for_sample(sample, name, typ).items()))
+            if group is not None and g != group and g in seen_groups:
+                raise ValueError("Invalid metric group ordering: " + line)
+            group = g
+            seen_groups.add(g)
+
             if typ == 'stateset' and sample.value not in [0, 1]:
                 raise ValueError("Stateset samples can only have values zero and one: " + line)
             if typ == 'info' and sample.value != 1:

tests/openmetrics/test_parser.py

@@ -192,7 +192,7 @@ def test_empty_metadata(self):
     def test_untyped(self):
         # https://github.com/prometheus/client_python/issues/79
         families = text_string_to_metric_families("""# HELP redis_connected_clients Redis connected clients
-# TYPE redis_connected_clients untyped
+# TYPE redis_connected_clients unknown
 redis_connected_clients{instance="rough-snowflake-web",port="6380"} 10.0
 redis_connected_clients{instance="rough-snowflake-web",port="6381"} 12.0
 # EOF
@@ -446,6 +446,7 @@ def test_invalid_input(self):
                 ('# TYPE a meh\n# EOF\n'),
                 ('# TYPE a meh \n# EOF\n'),
                 ('# TYPE a gauge \n# EOF\n'),
+                ('# TYPE a untyped\n# EOF\n'),
                 # Bad UNIT.
                 ('# UNIT\n# EOF\n'),
                 ('# UNIT \n# EOF\n'),
@@ -498,14 +499,28 @@ def test_invalid_input(self):
                 ('# TYPE a info\na 2\n# EOF\n'),
                 ('# TYPE a stateset\na 2.0\n# EOF\n'),
                 ('# TYPE a info\na 2.0\n# EOF\n'),
+                # Missing or invalid labels for a type.
+                ('# TYPE a summary\na 0\n# EOF\n'),
+                ('# TYPE a summary\na{quantile="-1"} 0\n# EOF\n'),
+                ('# TYPE a summary\na{quantile="foo"} 0\n# EOF\n'),
+                ('# TYPE a summary\na{quantile="1.01"} 0\n# EOF\n'),
+                ('# TYPE a summary\na{quantile="NaN"} 0\n# EOF\n'),
+                ('# TYPE a histogram\na_bucket 0\n# EOF\n'),
+                ('# TYPE a gaugehistogram\na_bucket 0\n# EOF\n'),
+                ('# TYPE a stateset\na 0\n# EOF\n'),
                 # Bad counter values.
                 ('# TYPE a counter\na_total NaN\n# EOF\n'),
                 ('# TYPE a histogram\na_sum NaN\n# EOF\n'),
                 ('# TYPE a histogram\na_count NaN\n# EOF\n'),
-                ('# TYPE a histogram\na_bucket NaN\n# EOF\n'),
-                ('# TYPE a gaugehistogram\na_bucket NaN\n# EOF\n'),
+                ('# TYPE a histogram\na_bucket{le="+Inf"} NaN\n# EOF\n'),
+                ('# TYPE a gaugehistogram\na_bucket{le="+Inf"} NaN\n# EOF\n'),
                 ('# TYPE a summary\na_sum NaN\n# EOF\n'),
                 ('# TYPE a summary\na_count NaN\n# EOF\n'),
+                # Bad grouping.
+                ('# TYPE a histogram\na_sum{a="1"} 0\na_sum{a="2"} 0\na_count{a="1"} 0\n# EOF\n'),
+                ('# TYPE a histogram\na_bucket{a="1",le="1"} 0\na_bucket{a="2",le="+Inf""} 0\na_bucket{a="1",le="+Inf"} 0\n# EOF\n'),
+                ('# TYPE a gaugehistogram\na_gsum{a="1"} 0\na_gsum{a="2"} 0\na_gcount{a="1"} 0\n# EOF\n'),
+                ('# TYPE a summary\nquantile{quantile="0"} 0\na_sum{a="1"} 0\nquantile{quantile="1"} 0\n# EOF\n'),
                 ]:
             with self.assertRaises(ValueError):
                 list(text_string_to_metric_families(case))