Allow underscores in label keys (#448)

Sanitize label keys in the stackdriver exporter by replacing sequences of
invalid chars with underscores.

Author: colincadams

opencensus/stats/exporters/stackdriver_exporter.py

@@ -17,6 +17,7 @@
 import os
 import platform
 import re
+import string
 
 from datetime import datetime
 from google.api_core.gapic_v1 import client_info
@@ -440,13 +441,13 @@ def new_label_descriptors(defaults, keys):
     label_descriptors = []
     for key, lbl in defaults.items():
         label = {}
-        label["key"] = remove_non_alphanumeric(key)
+        label["key"] = sanitize_label(key)
         label["description"] = lbl
         label_descriptors.append(label)
 
     for tag_key in keys:
         label = {}
-        label["key"] = remove_non_alphanumeric(tag_key)
+        label["key"] = sanitize_label(tag_key)
         label_descriptors.append(label)
     label_descriptors.append({"key": OPENCENSUS_TASK,
                               "description": OPENCENSUS_TASK_DESCRIPTION})
@@ -461,11 +462,22 @@ def set_metric_labels(series, view, tag_values):
 
     for key, value in zip(view.columns, tag_values):
         if value is not None:
-            series.metric.labels[remove_non_alphanumeric(key)] = value
+            series.metric.labels[sanitize_label(key)] = value
     series.metric.labels[OPENCENSUS_TASK] = get_task_value()
 
 
-def remove_non_alphanumeric(text):
-    """ Remove characters not accepted in labels key
+def sanitize_label(text):
+    """Remove characters not accepted in labels key
+
+    This replaces any non-word characters (alphanumeric or underscore), with
+    an underscore. It also ensures that the first character is a letter by
+    prepending with 'key' if necessary, and trims the text to 100 characters.
     """
-    return str(re.sub('[^0-9a-zA-Z ]+', '', text)).replace(" ", "")
+    if not text:
+        return text
+    text = re.sub('\\W+', '_', text)
+    if text[0] in string.digits:
+        text = "key_" + text
+    elif text[0] == '_':
+        text = "key" + text
+    return text[:100]

tests/unit/stats/exporter/test_stackdriver_stats.py

@@ -37,10 +37,10 @@
 FRONTEND_KEY_INT = tag_key_module.TagKey("my.org/keys/frontend-INT")
 FRONTEND_KEY_STR = tag_key_module.TagKey("my.org/keys/frontend-STR")
 
-FRONTEND_KEY_CLEAN = "myorgkeysfrontend"
-FRONTEND_KEY_FLOAT_CLEAN = "myorgkeysfrontendFLOAT"
-FRONTEND_KEY_INT_CLEAN = "myorgkeysfrontendINT"
-FRONTEND_KEY_STR_CLEAN = "myorgkeysfrontendSTR"
+FRONTEND_KEY_CLEAN = "my_org_keys_frontend"
+FRONTEND_KEY_FLOAT_CLEAN = "my_org_keys_frontend_FLOAT"
+FRONTEND_KEY_INT_CLEAN = "my_org_keys_frontend_INT"
+FRONTEND_KEY_STR_CLEAN = "my_org_keys_frontend_STR"
 
 VIDEO_SIZE_MEASURE = measure_module.MeasureInt(
     "my.org/measure/video_size_test2", "size of processed videos", "By")
@@ -142,16 +142,36 @@ def test_client_info_user_agent(self):
         self.assertIn(stackdriver.get_user_agent_slug(),
                       exporter.client.client_info.to_user_agent())
 
-    def test_remove_invalid_chars(self):
-        invalid_chars = "@#$"
-        valid_chars = "abc"
-
-        result = stackdriver.remove_non_alphanumeric(invalid_chars)
+    def test_sanitize(self):
+        # empty
+        result = stackdriver.sanitize_label("")
         self.assertEqual(result, "")
 
-        result = stackdriver.remove_non_alphanumeric(valid_chars)
+        # all invalid
+        result = stackdriver.sanitize_label("/*^#$")
+        self.assertEqual(result, "key_")
+
+        # all valid
+        result = stackdriver.sanitize_label("abc")
         self.assertEqual(result, "abc")
 
+        # mixed
+        result = stackdriver.sanitize_label("a.b/c")
+        self.assertEqual(result, "a_b_c")
+
+        # starts with '_'
+        result = stackdriver.sanitize_label("_abc")
+        self.assertEqual(result, "key_abc")
+
+        # starts with digit
+        result = stackdriver.sanitize_label("0abc")
+        self.assertEqual(result, "key_0abc")
+
+        # too long
+        result = stackdriver.sanitize_label("0123456789" * 10)
+        self.assertEqual(len(result), 100)
+        self.assertEqual(result, "key_" + "0123456789" * 9 + "012345")
+
     def test_singleton_with_params(self):
         default_labels = {'key1': 'value1'}
         patch_client = mock.patch(
@@ -884,7 +904,7 @@ def test_create_timeseries_from_distribution(self):
         [time_series] = time_series_list
 
         self.assertCorrectLabels(time_series.metric.labels,
-                                 {'tagkey': 'tag_value'},
+                                 {'tag_key': 'tag_value'},
                                  include_opencensus=True)
         self.assertEqual(len(time_series.points), 1)
         [point] = time_series.points