test

Author: sarah-witt

datadog_checks_base/datadog_checks/base/checks/libs/prometheus.py

@@ -7,6 +7,8 @@
 from prometheus_client.metrics_core import Metric
 from prometheus_client.parser import _parse_sample, _replace_help_escaping
 
+import datadog_checks.base.checks.openmetrics.parser_optimizations  # noqa: F401
+
 
 def text_fd_to_metric_families(fd):
     raw_lines, input_lines = tee(fd, 2)

datadog_checks_base/datadog_checks/base/checks/openmetrics/parser_optimizations.py

@@ -0,0 +1,50 @@
+# (C) Datadog, Inc. 2025-present
+# All rights reserved
+# Licensed under a 3-clause BSD style license (see LICENSE)
+"""Optimized replacements for prometheus_client parser hot-path functions.
+
+The prometheus_client v0.22.0+ parser introduced character-by-character scanning
+via _next_unquoted_char() for UTF-8 support, causing a ~3-5x performance regression
+(see https://github.com/prometheus/client_python/issues/1114).
+
+This module replaces _next_unquoted_char with a version that uses str.find() to
+jump directly to candidate characters instead of iterating character-by-character,
+restoring near-original performance. Quote-aware scanning is omitted because
+structural characters inside quoted label values do not occur in practice.
+"""
+
+import string
+
+import prometheus_client.parser as _prom_parser
+
+
+def _next_unquoted_char(text, chs, startidx=0):
+    """Find the next occurrence of any character in chs."""
+    if chs is None:
+        chs = string.whitespace
+
+    best = -1
+    for ch in chs:
+        p = text.find(ch, startidx)
+        if p != -1 and (best == -1 or p < best):
+            best = p
+    return best
+
+
+def apply():
+    """Monkey-patch prometheus_client parser modules with optimized functions."""
+    if getattr(_prom_parser, '_dd_optimized', False):
+        return
+
+    _prom_parser._next_unquoted_char = _next_unquoted_char
+    _prom_parser._dd_optimized = True
+
+    try:
+        import prometheus_client.openmetrics.parser as _om_parser
+
+        _om_parser._next_unquoted_char = _next_unquoted_char
+    except (ImportError, AttributeError):
+        pass
+
+
+apply()

datadog_checks_base/datadog_checks/base/checks/openmetrics/v2/scraper/base_scraper.py

@@ -13,6 +13,8 @@
 from prometheus_client import Metric
 from prometheus_client.openmetrics.parser import text_fd_to_metric_families as parse_openmetrics
 from prometheus_client.parser import text_fd_to_metric_families as parse_prometheus
+
+import datadog_checks.base.checks.openmetrics.parser_optimizations  # noqa: F401
 from requests.exceptions import ConnectionError
 
 from datadog_checks.base.agent import datadog_agent

datadog_checks_base/tests/base/checks/openmetrics/test_parser_optimizations.py

@@ -0,0 +1,207 @@
+# (C) Datadog, Inc. 2025-present
+# All rights reserved
+# Licensed under a 3-clause BSD style license (see LICENSE)
+"""Tests for the optimized _next_unquoted_char replacement.
+
+Verifies that the optimized version produces the same results as the
+original prometheus_client implementation across representative inputs.
+"""
+
+import pytest
+
+from datadog_checks.base.checks.openmetrics.parser_optimizations import (
+    _next_unquoted_char,
+)
+
+
+class TestNextUnquotedChar:
+    """Tests for the optimized _next_unquoted_char function."""
+
+    def test_find_single_char(self):
+        assert _next_unquoted_char('foo{bar="baz"} 1', '{') == 3
+
+    def test_find_closing_brace(self):
+        assert _next_unquoted_char('bar="baz"} 1', '}') == 9
+
+    def test_find_equals(self):
+        assert _next_unquoted_char('label="value"', '=') == 5
+
+    def test_find_comma(self):
+        assert _next_unquoted_char('a="1",b="2"', ',') == 5
+
+    def test_find_space(self):
+        assert _next_unquoted_char('metric{l="v"} 42', ' ') == 13
+
+    def test_find_multiple_targets(self):
+        assert _next_unquoted_char('label=value,next', '=,}') == 5
+
+    def test_find_multiple_targets_comma_first(self):
+        assert _next_unquoted_char('value,next=foo', '=,}') == 5
+
+    def test_find_multiple_targets_brace(self):
+        assert _next_unquoted_char('value}', '=,}') == 5
+
+    def test_not_found(self):
+        assert _next_unquoted_char('no_special_chars', '{') == -1
+
+    def test_empty_string(self):
+        assert _next_unquoted_char('', '{') == -1
+
+    def test_startidx(self):
+        assert _next_unquoted_char('a{b{c', '{', 2) == 3
+
+    def test_startidx_at_target(self):
+        assert _next_unquoted_char('a{b', '{', 1) == 1
+
+    def test_startidx_past_end(self):
+        assert _next_unquoted_char('abc', '{', 10) == -1
+
+    def test_whitespace_default(self):
+        assert _next_unquoted_char('foo bar', None) == 3
+
+    def test_whitespace_tab(self):
+        assert _next_unquoted_char('foo\tbar', None) == 3
+
+    def test_first_char_is_target(self):
+        assert _next_unquoted_char('{foo}', '{') == 0
+
+    def test_last_char_is_target(self):
+        assert _next_unquoted_char('foo}', '}') == 3
+
+    def test_multiple_occurrences_returns_first(self):
+        assert _next_unquoted_char('a{b{c', '{') == 1
+
+
+class TestNextUnquotedCharWithRealMetrics:
+    """Tests using real Prometheus metric line patterns."""
+
+    def test_simple_gauge(self):
+        line = 'envoy_server_live 1'
+        assert _next_unquoted_char(line, '{') == -1
+        assert _next_unquoted_char(line, ' ') == 17
+
+    def test_labeled_metric(self):
+        line = 'envoy_cluster_upstream_cx_active{envoy_cluster_name="service1"} 0'
+        assert _next_unquoted_char(line, '{') == 32
+        assert _next_unquoted_char(line, '}', 33) == 62
+
+    def test_multi_label_metric(self):
+        line = 'http_requests_total{method="GET",code="200"} 1027'
+        assert _next_unquoted_char(line, '{') == 19
+        assert _next_unquoted_char(line, '=', 20) == 26
+        labels_text = 'method="GET",code="200"'
+        assert _next_unquoted_char(labels_text, '=,}') == 6
+        assert _next_unquoted_char(labels_text, ',}', 12) == 12
+
+    def test_histogram_bucket(self):
+        line = 'http_request_duration_seconds_bucket{le="0.5"} 24054'
+        assert _next_unquoted_char(line, '{') == 36
+        assert _next_unquoted_char(line, '}', 37) == 45
+
+    def test_help_line_split(self):
+        line = '# HELP http_requests_total The total number of HTTP requests.'
+        assert _next_unquoted_char(line, None) == 1
+
+    def test_type_line_split(self):
+        line = '# TYPE http_requests_total counter'
+        assert _next_unquoted_char(line, None) == 1
+
+
+class TestParseFullMetricText:
+    """Integration tests that parse complete metric text through the patched parser."""
+
+    def test_parse_simple_metrics(self):
+        from prometheus_client.parser import text_string_to_metric_families
+
+        text = (
+            '# HELP test_gauge A test gauge.\n'
+            '# TYPE test_gauge gauge\n'
+            'test_gauge 42\n'
+        )
+        families = list(text_string_to_metric_families(text))
+        assert len(families) == 1
+        assert families[0].name == 'test_gauge'
+        assert families[0].samples[0].value == 42
+
+    def test_parse_labeled_metrics(self):
+        from prometheus_client.parser import text_string_to_metric_families
+
+        text = (
+            '# HELP http_requests_total Total requests.\n'
+            '# TYPE http_requests_total counter\n'
+            'http_requests_total{method="GET",code="200"} 1027\n'
+            'http_requests_total{method="POST",code="200"} 3\n'
+        )
+        families = list(text_string_to_metric_families(text))
+        assert len(families) == 1
+        assert len(families[0].samples) == 2
+        assert families[0].samples[0].labels == {'method': 'GET', 'code': '200'}
+        assert families[0].samples[0].value == 1027
+        assert families[0].samples[1].labels == {'method': 'POST', 'code': '200'}
+
+    def test_parse_histogram(self):
+        from prometheus_client.parser import text_string_to_metric_families
+
+        text = (
+            '# HELP rpc_duration_seconds RPC duration.\n'
+            '# TYPE rpc_duration_seconds histogram\n'
+            'rpc_duration_seconds_bucket{le="0.5"} 2000\n'
+            'rpc_duration_seconds_bucket{le="1.0"} 2500\n'
+            'rpc_duration_seconds_bucket{le="+Inf"} 3000\n'
+            'rpc_duration_seconds_sum 5000\n'
+            'rpc_duration_seconds_count 3000\n'
+        )
+        families = list(text_string_to_metric_families(text))
+        assert len(families) == 1
+        assert families[0].type == 'histogram'
+        assert len(families[0].samples) == 5
+
+    def test_parse_escaped_label_value(self):
+        from prometheus_client.parser import text_string_to_metric_families
+
+        text = (
+            '# HELP test_metric A test.\n'
+            '# TYPE test_metric gauge\n'
+            'test_metric{label="value with \\"quotes\\""} 1\n'
+        )
+        families = list(text_string_to_metric_families(text))
+        assert len(families) == 1
+        assert families[0].samples[0].labels == {'label': 'value with "quotes"'}
+
+    def test_parse_multiple_families(self):
+        from prometheus_client.parser import text_string_to_metric_families
+
+        text = (
+            '# HELP gauge_one First.\n'
+            '# TYPE gauge_one gauge\n'
+            'gauge_one 1\n'
+            '# HELP gauge_two Second.\n'
+            '# TYPE gauge_two gauge\n'
+            'gauge_two{env="prod"} 2\n'
+        )
+        families = list(text_string_to_metric_families(text))
+        assert len(families) == 2
+        assert families[0].name == 'gauge_one'
+        assert families[1].name == 'gauge_two'
+
+    def test_parse_empty_label_value(self):
+        from prometheus_client.parser import text_string_to_metric_families
+
+        text = (
+            '# HELP test_metric A test.\n'
+            '# TYPE test_metric gauge\n'
+            'test_metric{label=""} 1\n'
+        )
+        families = list(text_string_to_metric_families(text))
+        assert families[0].samples[0].labels == {'label': ''}
+
+    def test_parse_newline_in_label_value(self):
+        from prometheus_client.parser import text_string_to_metric_families
+
+        text = (
+            '# HELP test_metric A test.\n'
+            '# TYPE test_metric gauge\n'
+            'test_metric{label="line1\\nline2"} 1\n'
+        )
+        families = list(text_string_to_metric_families(text))
+        assert families[0].samples[0].labels == {'label': 'line1\nline2'}