stats: use RE2 and a better pattern to accelerate a single stats tag-extraction RE

source/common/common/regex.h

@@ -9,6 +9,8 @@
 namespace Envoy {
 namespace Regex {
 
+enum class Type { Re2, StdRegex };
+
 /**
  * Utilities for constructing regular expressions.
  */

source/common/config/BUILD

@@ -450,6 +450,7 @@ envoy_cc_library(
     hdrs = ["well_known_names.h"],
     deps = [
         "//source/common/common:assert_lib",
+        "//source/common/common:regex_lib",
         "//source/common/singleton:const_singleton",
     ],
 )

source/common/config/well_known_names.cc

@@ -92,7 +92,8 @@ TagNameValues::TagNameValues() {
   addRegex(RATELIMIT_PREFIX, R"(^ratelimit\.((.*?)\.)\w+?$)");
 
   // cluster.(<cluster_name>.)*
-  addRegex(CLUSTER_NAME, "^cluster\\.((.*?)\\.)");
+  // addRegex(CLUSTER_NAME, "^cluster\\.((.*?)\\.)");
+  addRe2(CLUSTER_NAME, "^cluster\\.(([^\\.]+)\\.).*");
 
   // listener.[<address>.]http.(<stat_prefix>.)*
   addRegex(HTTP_CONN_MANAGER_PREFIX, R"(^listener(?=\.).*?\.http\.((.*?)\.))", ".http.");
@@ -119,7 +120,12 @@ TagNameValues::TagNameValues() {
 
 void TagNameValues::addRegex(const std::string& name, const std::string& regex,
                              const std::string& substr) {
-  descriptor_vec_.emplace_back(Descriptor(name, regex, substr));
+  descriptor_vec_.emplace_back(Descriptor{name, regex, substr, Regex::Type::StdRegex});
+}
+
+void TagNameValues::addRe2(const std::string& name, const std::string& regex,
+                           const std::string& substr) {
+  descriptor_vec_.emplace_back(Descriptor{name, regex, substr, Regex::Type::Re2});
 }
 
 } // namespace Config

source/common/config/well_known_names.h

@@ -6,6 +6,7 @@
 #include "envoy/common/exception.h"
 
 #include "common/common/assert.h"
+#include "common/common/regex.h"
 #include "common/singleton/const_singleton.h"
 
 namespace Envoy {
@@ -62,11 +63,10 @@ class TagNameValues {
    * tags, such as "_rq_(\\d)xx$", will probably stay as regexes.
    */
   struct Descriptor {
-    Descriptor(const std::string& name, const std::string& regex, const std::string& substr = "")
-        : name_(name), regex_(regex), substr_(substr) {}
     const std::string name_;
     const std::string regex_;
     const std::string substr_;
+    Regex::Type re_type_;
   };
 
   // Cluster name tag
@@ -130,6 +130,7 @@ class TagNameValues {
 
 private:
   void addRegex(const std::string& name, const std::string& regex, const std::string& substr = "");
+  void addRe2(const std::string& name, const std::string& regex, const std::string& substr = "");
 
   // Collection of tag descriptors.
   std::vector<Descriptor> descriptor_vec_;

source/common/stats/BUILD

@@ -195,6 +195,7 @@ envoy_cc_library(
     hdrs = ["tag_extractor_impl.h"],
     deps = [
         "//include/envoy/stats:stats_interface",
+        "//source/common/common:assert_lib",
         "//source/common/common:perf_annotation_lib",
         "//source/common/common:regex_lib",
     ],

source/common/stats/tag_extractor_impl.cc

@@ -5,6 +5,7 @@
 
 #include "envoy/common/exception.h"
 
+#include "common/common/assert.h"
 #include "common/common/fmt.h"
 #include "common/common/perf_annotation.h"
 #include "common/common/regex.h"
@@ -23,12 +24,11 @@ bool regexStartsWithDot(absl::string_view regex) {
 
 } // namespace
 
-TagExtractorImpl::TagExtractorImpl(const std::string& name, const std::string& regex,
-                                   const std::string& substr)
-    : name_(name), prefix_(std::string(extractRegexPrefix(regex))), substr_(substr),
-      regex_(Regex::Utility::parseStdRegex(regex)) {}
+TagExtractorImplBase::TagExtractorImplBase(const std::string& name, const std::string& regex,
+                                           const std::string& substr)
+    : name_(name), prefix_(std::string(extractRegexPrefix(regex))), substr_(substr) {}
 
-std::string TagExtractorImpl::extractRegexPrefix(absl::string_view regex) {
+std::string TagExtractorImplBase::extractRegexPrefix(absl::string_view regex) {
   std::string prefix;
   if (absl::StartsWith(regex, "^")) {
     for (absl::string_view::size_type i = 1; i < regex.size(); ++i) {
@@ -47,10 +47,10 @@ std::string TagExtractorImpl::extractRegexPrefix(absl::string_view regex) {
   return prefix;
 }
 
-TagExtractorPtr TagExtractorImpl::createTagExtractor(const std::string& name,
-                                                     const std::string& regex,
-                                                     const std::string& substr) {
-
+TagExtractorPtr TagExtractorImplBase::createTagExtractor(const std::string& name,
+                                                         const std::string& regex,
+                                                         const std::string& substr,
+                                                         Regex::Type re_type) {
   if (name.empty()) {
     throw EnvoyException("tag_name cannot be empty");
   }
@@ -59,19 +59,29 @@ TagExtractorPtr TagExtractorImpl::createTagExtractor(const std::string& name,
     throw EnvoyException(fmt::format(
         "No regex specified for tag specifier and no default regex for name: '{}'", name));
   }
-  return TagExtractorPtr{new TagExtractorImpl(name, regex, substr)};
+  switch (re_type) {
+  case Regex::Type::Re2:
+    return std::make_unique<TagExtractorRe2Impl>(name, regex, substr);
+  case Regex::Type::StdRegex:
+    return std::make_unique<TagExtractorStdRegexImpl>(name, regex, substr);
+  }
+  NOT_REACHED_GCOVR_EXCL_LINE;
 }
 
-bool TagExtractorImpl::substrMismatch(absl::string_view stat_name) const {
+bool TagExtractorImplBase::substrMismatch(absl::string_view stat_name) const {
   return !substr_.empty() && stat_name.find(substr_) == absl::string_view::npos;
 }
 
-bool TagExtractorImpl::extractTag(absl::string_view stat_name, TagVector& tags,
-                                  IntervalSet<size_t>& remove_characters) const {
+TagExtractorStdRegexImpl::TagExtractorStdRegexImpl(const std::string& name, const std::string regex,
+                                                   const std::string& substr)
+    : TagExtractorImplBase(name, regex, substr), regex_(Regex::Utility::parseStdRegex(regex)) {}
+
+bool TagExtractorStdRegexImpl::extractTag(absl::string_view stat_name, std::vector<Tag>& tags,
+                                          IntervalSet<size_t>& remove_characters) const {
   PERF_OPERATION(perf);
 
   if (substrMismatch(stat_name)) {
-    PERF_RECORD(perf, "re-skip-substr", name_);
+    PERF_RECORD(perf, "re-skip", name_);
     return false;
   }
 
@@ -105,5 +115,50 @@ bool TagExtractorImpl::extractTag(absl::string_view stat_name, TagVector& tags,
   return false;
 }
 
+TagExtractorRe2Impl::TagExtractorRe2Impl(const std::string& name, const std::string regex,
+                                         const std::string& substr)
+    : TagExtractorImplBase(name, regex, substr), regex_(regex) {}
+
+bool TagExtractorRe2Impl::extractTag(absl::string_view stat_name, std::vector<Tag>& tags,
+                                     IntervalSet<size_t>& remove_characters) const {
+  PERF_OPERATION(perf);
+
+  if (substrMismatch(stat_name)) {
+    PERF_RECORD(perf, "re2-skip", name_);
+    return false;
+  }
+
+  // remove_subexpr is the first submatch. It represents the portion of the string to be removed.
+  re2::StringPiece remove_subexpr, value_subexpr;
+
+  // The regex must match and contain one or more subexpressions (all after the first are ignored).
+  if (re2::RE2::FullMatch(re2::StringPiece(stat_name.data(), stat_name.size()), regex_,
+                          &remove_subexpr, &value_subexpr) &&
+      !remove_subexpr.empty()) {
+
+    // value_subexpr is the optional second submatch. It is usually inside the first submatch
+    // (remove_subexpr) to allow the expression to strip off extra characters that should be removed
+    // from the string but also not necessary in the tag value ("." for example). If there is no
+    // second submatch, then the value_subexpr is the same as the remove_subexpr.
+    if (value_subexpr.empty()) {
+      value_subexpr = remove_subexpr;
+    }
+
+    tags.emplace_back();
+    Tag& tag = tags.back();
+    tag.name_ = name_;
+    tag.value_ = std::string(value_subexpr);
+
+    // Determines which characters to remove from stat_name to elide remove_subexpr.
+    std::string::size_type start = remove_subexpr.data() - stat_name.data();
+    std::string::size_type end = remove_subexpr.data() + remove_subexpr.size() - stat_name.data();
+    remove_characters.insert(start, end);
+    PERF_RECORD(perf, "re2-match", name_);
+    return true;
+  }
+  PERF_RECORD(perf, "re2-miss", name_);
+  return false;
+}
+
 } // namespace Stats
 } // namespace Envoy

source/common/stats/tag_extractor_impl.h

@@ -6,12 +6,15 @@
 
 #include "envoy/stats/tag_extractor.h"
 
+#include "common/common/regex.h"
+
 #include "absl/strings/string_view.h"
+#include "re2/re2.h"
 
 namespace Envoy {
 namespace Stats {
 
-class TagExtractorImpl : public TagExtractor {
+class TagExtractorImplBase : public TagExtractor {
 public:
   /**
    * Creates a tag extractor from the regex provided. name and regex must be non-empty.
@@ -23,13 +26,12 @@ class TagExtractorImpl : public TagExtractor {
    * @return TagExtractorPtr newly constructed TagExtractor.
    */
   static TagExtractorPtr createTagExtractor(const std::string& name, const std::string& regex,
-                                            const std::string& substr = "");
+                                            const std::string& substr = "",
+                                            Regex::Type re_type = Regex::Type::StdRegex);
 
-  TagExtractorImpl(const std::string& name, const std::string& regex,
-                   const std::string& substr = "");
+  TagExtractorImplBase(const std::string& name, const std::string& regex,
+                       const std::string& substr = "");
   std::string name() const override { return name_; }
-  bool extractTag(absl::string_view tag_extracted_name, TagVector& tags,
-                  IntervalSet<size_t>& remove_characters) const override;
   absl::string_view prefixToken() const override { return prefix_; }
 
   /**
@@ -39,7 +41,7 @@ class TagExtractorImpl : public TagExtractor {
    */
   bool substrMismatch(absl::string_view stat_name) const;
 
-private:
+protected:
   /**
    * Examines a regex string, looking for the pattern: ^alphanumerics_with_underscores\.
    * Returns "alphanumerics_with_underscores" if that pattern is found, empty-string otherwise.
@@ -50,8 +52,31 @@ class TagExtractorImpl : public TagExtractor {
   const std::string name_;
   const std::string prefix_;
   const std::string substr_;
+};
+
+class TagExtractorStdRegexImpl : public TagExtractorImplBase {
+public:
+  TagExtractorStdRegexImpl(const std::string& name, const std::string regex,
+                           const std::string& substr = "");
+
+  bool extractTag(absl::string_view tag_extracted_name, std::vector<Tag>& tags,
+                  IntervalSet<size_t>& remove_characters) const override;
+
+private:
   const std::regex regex_;
 };
 
+class TagExtractorRe2Impl : public TagExtractorImplBase {
+public:
+  TagExtractorRe2Impl(const std::string& name, const std::string regex,
+                      const std::string& substr = "");
+
+  bool extractTag(absl::string_view tag_extracted_name, std::vector<Tag>& tags,
+                  IntervalSet<size_t>& remove_characters) const override;
+
+private:
+  re2::RE2 regex_;
+};
+
 } // namespace Stats
 } // namespace Envoy

source/common/stats/tag_producer_impl.cc

@@ -34,7 +34,7 @@ TagProducerImpl::TagProducerImpl(const envoy::config::metrics::v3::StatsConfig&
               "No regex specified for tag specifier and no default regex for name: '{}'", name));
         }
       } else {
-        addExtractor(Stats::TagExtractorImpl::createTagExtractor(name, tag_specifier.regex()));
+        addExtractor(Stats::TagExtractorImplBase::createTagExtractor(name, tag_specifier.regex()));
       }
     } else if (tag_specifier.tag_value_case() ==
                envoy::config::metrics::v3::TagSpecifier::TagValueCase::kFixedValue) {
@@ -47,8 +47,8 @@ int TagProducerImpl::addExtractorsMatching(absl::string_view name) {
   int num_found = 0;
   for (const auto& desc : Config::TagNames::get().descriptorVec()) {
     if (desc.name_ == name) {
-      addExtractor(
-          Stats::TagExtractorImpl::createTagExtractor(desc.name_, desc.regex_, desc.substr_));
+      addExtractor(Stats::TagExtractorImplBase::createTagExtractor(desc.name_, desc.regex_,
+                                                                   desc.substr_, desc.re_type_));
       ++num_found;
     }
   }
@@ -103,8 +103,8 @@ TagProducerImpl::addDefaultExtractors(const envoy::config::metrics::v3::StatsCon
   if (!config.has_use_all_default_tags() || config.use_all_default_tags().value()) {
     for (const auto& desc : Config::TagNames::get().descriptorVec()) {
       names.emplace(desc.name_);
-      addExtractor(
-          Stats::TagExtractorImpl::createTagExtractor(desc.name_, desc.regex_, desc.substr_));
+      addExtractor(Stats::TagExtractorImplBase::createTagExtractor(desc.name_, desc.regex_,
+                                                                   desc.substr_, desc.re_type_));
     }
   }
   return names;

test/common/common/BUILD

@@ -269,6 +269,17 @@ envoy_cc_test(
     deps = ["//source/common/common:callback_impl_lib"],
 )
 
+envoy_cc_benchmark_binary(
+    name = "re_speed_test",
+    srcs = ["re_speed_test.cc"],
+    external_deps = ["benchmark"],
+    deps = [
+        "//source/common/common:assert_lib",
+        "//source/common/common:utility_lib",
+        "@com_googlesource_code_re2//:re2",
+    ],
+)
+
 envoy_cc_benchmark_binary(
     name = "utility_speed_test",
     srcs = ["utility_speed_test.cc"],