Commit 8ae778e3 authored by Jun LI's avatar Jun LI Committed by dcsommer

Defer query string parsing from ctor to getQueryParams()

Summary:
Query string parsing uses a lot of CPU, it happens in ctor killing CPU
even in users that are not interested in query string. Move it to
getQueryParams().

Test Plan:
fbconfig folly/test
fbmake runtests_dbg

Reviewed By: ldemailly@fb.com

Subscribers: trunkagent, njormrod, zellux

FB internal diff: D1604973

Tasks: 5304484

Blame Revision: https://phabricator.fb.com/D1455158
parent 4a9aef96
...@@ -91,26 +91,6 @@ Uri::Uri(StringPiece str) : port_(0) { ...@@ -91,26 +91,6 @@ Uri::Uri(StringPiece str) : port_(0) {
} }
query_ = submatch(match, 3); query_ = submatch(match, 3);
if (!query_.empty()) {
// Parse query string
static const boost::regex queryParamRegex(
"(^|&)([^=&]*)=?([^=&]*)(?=(&|$))");
boost::cregex_iterator paramBeginItr(
match[3].first,
match[3].second,
queryParamRegex);
boost::cregex_iterator paramEndItr;
for(auto itr = paramBeginItr; itr != paramEndItr; itr++) {
if (itr->length(2) == 0) {
// key is empty, ignore it
continue;
}
queryParams_.emplace_back(
fbstring((*itr)[2].first, (*itr)[2].second), // parameter name
fbstring((*itr)[3].first, (*itr)[3].second) // parameter value
);
}
}
fragment_ = submatch(match, 4); fragment_ = submatch(match, 4);
} }
...@@ -150,4 +130,30 @@ fbstring Uri::hostname() const { ...@@ -150,4 +130,30 @@ fbstring Uri::hostname() const {
return host_; return host_;
} }
const std::vector<std::pair<fbstring, fbstring>>& Uri::getQueryParams() {
if (!query_.empty() && queryParams_.empty()) {
// Parse query string
static const boost::regex queryParamRegex(
"(^|&)" /*start of query or start of parameter "&"*/
"([^=&]*)=?" /*parameter name and "=" if value is expected*/
"([^=&]*)" /*parameter value*/
"(?=(&|$))" /*forward reference, next should be end of query or
start of next parameter*/);
boost::cregex_iterator paramBeginItr(
query_.data(), query_.data() + query_.size(), queryParamRegex);
boost::cregex_iterator paramEndItr;
for (auto itr = paramBeginItr; itr != paramEndItr; itr++) {
if (itr->length(2) == 0) {
// key is empty, ignore it
continue;
}
queryParams_.emplace_back(
fbstring((*itr)[2].first, (*itr)[2].second), // parameter name
fbstring((*itr)[3].first, (*itr)[3].second) // parameter value
);
}
}
return queryParams_;
}
} // namespace folly } // namespace folly
...@@ -90,13 +90,15 @@ class Uri { ...@@ -90,13 +90,15 @@ class Uri {
* one equal signs, we don't know which one is the delimiter for key and * one equal signs, we don't know which one is the delimiter for key and
* value. * value.
* *
* Note, this method is not thread safe, it might update internal state, but
* only the first call to this method update the state. After the first call
* is finished, subsequent calls to this method are thread safe.
*
* @return query parameter key-value pairs in a vector, each element is a * @return query parameter key-value pairs in a vector, each element is a
* pair of which the first element is parameter name and the second * pair of which the first element is parameter name and the second
* one is parameter value * one is parameter value
*/ */
const std::vector<std::pair<fbstring, fbstring>>& getQueryParams() const { const std::vector<std::pair<fbstring, fbstring>>& getQueryParams();
return queryParams_;
};
private: private:
fbstring scheme_; fbstring scheme_;
......
...@@ -15,6 +15,7 @@ ...@@ -15,6 +15,7 @@
*/ */
#include <folly/Uri.h> #include <folly/Uri.h>
#include <folly/Benchmark.h>
#include <boost/algorithm/string.hpp> #include <boost/algorithm/string.hpp>
#include <glog/logging.h> #include <glog/logging.h>
...@@ -389,3 +390,74 @@ TEST(Uri, Simple) { ...@@ -389,3 +390,74 @@ TEST(Uri, Simple) {
} }
} }
} }
/**
* Result of benchmark varies by the complexity of query.
* ============================================================================
* folly/test/UriTest.cpp relative time/iter iters/s
* ============================================================================
* init_uri_simple 4.88us 204.80K
* init_uri_simple_with_query_parsing 22.46us 44.52K
* init_uri_complex 5.92us 168.85K
* init_uri_complex_with_query_parsing 48.70us 20.53K
* ============================================================================
*/
BENCHMARK(init_uri_simple, iters) {
const fbstring s("http://localhost?&key1=foo&key2=&key3&=bar&=bar=&");
for (int i = 0; i < iters; ++i) {
Uri u(s);
}
}
BENCHMARK(init_uri_simple_with_query_parsing, iters) {
const fbstring s("http://localhost?&key1=foo&key2=&key3&=bar&=bar=&");
for (int i = 0; i < iters; ++i) {
Uri u(s);
u.getQueryParams();
}
}
BENCHMARK(init_uri_complex, iters) {
const fbstring s(
"https://mock.example.com/farm/track.php?TmOxQUDF=uSmTS_VwhjKnh_JME&DI"
"h=fbbN&GRsoIm=bGshjaUqavZxQai&UMT=36k18N4dn21&3U=CD8o4A4497W152j6m0V%14"
"%57&Hy=t%05mpr.80JUZ7ne_%23zS8DcA%0qc_%291ymamz096%11Zfb3r%09ZqPD%311ZX"
"tqJd600ot&5U96U-Rh-VZ=-D_6-9xKYj%1gW6b43s1B9-j21P0oUW5-t46G4kgt&ezgj=mcW"
"TTQ.c&Oh=%2PblUfuC%7C997048884827569%03xnyJ%2L1pi7irBioQ6D4r7nNHNdo6v7Y%"
"84aurnSJ%2wCFePHMlGZmIHGfCe7392_lImWsSvN&sBeNN=Nf%80yOE%6X10M64F4gG197aX"
"R2B4g2533x235A0i4e%57%58uWB%04Erw.60&VMS4=Ek_%02GC0Pkx%6Ov_%207WICUz007%"
"04nYX8N%46zzpv%999h&KGmBt988y=q4P57C-Dh-Nz-x_7-5oPxz%1gz3N03t6c7-R67N4DT"
"Y6-f98W1&Lts&%02dOty%8eEYEnLz4yexQQLnL4MGU2JFn3OcmXcatBcabZgBdDdy67hdgW"
"tYn4");
for (int i = 0; i < iters; ++i) {
Uri u(s);
}
}
BENCHMARK(init_uri_complex_with_query_parsing, iters) {
const fbstring s(
"https://mock.example.com/farm/track.php?TmOxQUDF=uSmTS_VwhjKnh_JME&DI"
"h=fbbN&GRsoIm=bGshjaUqavZxQai&UMT=36k18N4dn21&3U=CD8o4A4497W152j6m0V%14"
"%57&Hy=t%05mpr.80JUZ7ne_%23zS8DcA%0qc_%291ymamz096%11Zfb3r%09ZqPD%311ZX"
"tqJd600ot&5U96U-Rh-VZ=-D_6-9xKYj%1gW6b43s1B9-j21P0oUW5-t46G4kgt&ezgj=mcW"
"TTQ.c&Oh=%2PblUfuC%7C997048884827569%03xnyJ%2L1pi7irBioQ6D4r7nNHNdo6v7Y%"
"84aurnSJ%2wCFePHMlGZmIHGfCe7392_lImWsSvN&sBeNN=Nf%80yOE%6X10M64F4gG197aX"
"R2B4g2533x235A0i4e%57%58uWB%04Erw.60&VMS4=Ek_%02GC0Pkx%6Ov_%207WICUz007%"
"04nYX8N%46zzpv%999h&KGmBt988y=q4P57C-Dh-Nz-x_7-5oPxz%1gz3N03t6c7-R67N4DT"
"Y6-f98W1&Lts&%02dOty%8eEYEnLz4yexQQLnL4MGU2JFn3OcmXcatBcabZgBdDdy67hdgW"
"tYn4");
for (int i = 0; i < iters; ++i) {
Uri u(s);
u.getQueryParams();
}
}
int main(int argc, char** argv) {
testing::InitGoogleTest(&argc, argv);
auto r = RUN_ALL_TESTS();
if (r) {
return r;
}
runBenchmarks();
return 0;
}
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment