bp #40330
This commit is contained in:
Mingyu Chen
@ -93,4 +93,5 @@ header:
|
||||
- "pytest/sys/data"
|
||||
- "pytest/deploy/*.conf"
|
||||
- "tools/jeprof"
|
||||
- "tools/FlameGraph/*"
|
||||
comment: on-failure
|
||||
|
||||
10
LICENSE.txt
10
LICENSE.txt
@ -725,4 +725,12 @@ Apache 2.0, Copyright 2023 SAP SE or an SAP affiliate company, Johannes Bechberg
|
||||
|
||||
This project is maintained by the SapMachine team at SAP SE
|
||||
|
||||
----------------------------------------------------------------------------------
|
||||
----------------------------------------------------------------------------------
|
||||
|
||||
be/tools/FlameGraph/*.pl: COMMON DEVELOPMENT AND DISTRIBUTION LICENSE Version 1.0
|
||||
|
||||
Unless otherwise noted, all files in this distribution are released
|
||||
under the Common Development and Distribution License (CDDL).
|
||||
Exceptions are noted within the associated source files.
|
||||
|
||||
----------------------------------------------------------------------------------
|
||||
|
||||
@ -17,9 +17,15 @@
|
||||
|
||||
#include "http/action/pprof_actions.h"
|
||||
|
||||
#if !defined(__SANITIZE_ADDRESS__) && !defined(ADDRESS_SANITIZER) && !defined(LEAK_SANITIZER) && \
|
||||
!defined(THREAD_SANITIZER) && !defined(USE_JEMALLOC)
|
||||
#include <gperftools/heap-profiler.h> // IWYU pragma: keep
|
||||
#include <gperftools/malloc_extension.h> // IWYU pragma: keep
|
||||
#include <gperftools/profiler.h> // IWYU pragma: keep
|
||||
#endif
|
||||
#if !defined(__SANITIZE_ADDRESS__) && !defined(ADDRESS_SANITIZER) && !defined(LEAK_SANITIZER) && \
|
||||
!defined(THREAD_SANITIZER)
|
||||
#include <gperftools/profiler.h> // IWYU pragma: keep
|
||||
#endif
|
||||
#include <stdio.h>
|
||||
|
||||
#include <fstream>
|
||||
@ -133,8 +139,7 @@ public:
|
||||
};
|
||||
|
||||
void ProfileAction::handle(HttpRequest* req) {
|
||||
#if defined(ADDRESS_SANITIZER) || defined(LEAK_SANITIZER) || defined(THREAD_SANITIZER) || \
|
||||
defined(USE_JEMALLOC)
|
||||
#if defined(ADDRESS_SANITIZER) || defined(LEAK_SANITIZER) || defined(THREAD_SANITIZER)
|
||||
std::string str = "CPU profiling is not available with address sanitizer or jemalloc builds.";
|
||||
HttpChannel::send_reply(req, str);
|
||||
#else
|
||||
@ -170,6 +175,7 @@ void ProfileAction::handle(HttpRequest* req) {
|
||||
prof_file.close();
|
||||
std::string str = ss.str();
|
||||
HttpChannel::send_reply(req, str);
|
||||
return;
|
||||
}
|
||||
|
||||
// text type. we will return readable content via http response
|
||||
@ -185,7 +191,7 @@ void ProfileAction::handle(HttpRequest* req) {
|
||||
std::string svg_file_content;
|
||||
std::string flamegraph_install_dir =
|
||||
std::string(std::getenv("DORIS_HOME")) + "/tools/FlameGraph/";
|
||||
Status st = PprofUtils::generate_flamegraph(30, flamegraph_install_dir, false,
|
||||
Status st = PprofUtils::generate_flamegraph(seconds, flamegraph_install_dir, false,
|
||||
&svg_file_content);
|
||||
if (!st.ok()) {
|
||||
HttpChannel::send_reply(req, st.to_string());
|
||||
|
||||
@ -281,8 +281,7 @@ void heap_handler(const WebPageHandler::ArgumentMap& args, std::stringstream* ou
|
||||
void cpu_handler(const WebPageHandler::ArgumentMap& args, std::stringstream* output) {
|
||||
(*output) << "<h2>CPU Profile</h2>" << std::endl;
|
||||
|
||||
#if defined(ADDRESS_SANITIZER) || defined(LEAK_SANITIZER) || defined(THREAD_SANITIZER) || \
|
||||
defined(USE_JEMALLOC)
|
||||
#if defined(ADDRESS_SANITIZER) || defined(LEAK_SANITIZER) || defined(THREAD_SANITIZER)
|
||||
(*output) << "<pre>" << std::endl;
|
||||
(*output) << "CPU profiling is not available with address sanitizer builds." << std::endl;
|
||||
(*output) << "</pre>" << std::endl;
|
||||
@ -313,7 +312,8 @@ void cpu_handler(const WebPageHandler::ArgumentMap& args, std::stringstream* out
|
||||
<< std::endl;
|
||||
(*output) << "And you need to download the FlameGraph and place it under 'be/tools/FlameGraph'."
|
||||
<< std::endl;
|
||||
(*output) << "Finally, check if the following files exist" << std::endl;
|
||||
(*output) << "Finally, check if the following files exist. And should be executable."
|
||||
<< std::endl;
|
||||
(*output) << std::endl;
|
||||
(*output) << " be/tools/FlameGraph/stackcollapse-perf.pl" << std::endl;
|
||||
(*output) << " be/tools/FlameGraph/flamegraph.pl" << std::endl;
|
||||
@ -333,9 +333,6 @@ void cpu_handler(const WebPageHandler::ArgumentMap& args, std::stringstream* out
|
||||
<< std::endl;
|
||||
(*output) << " <br/>" << std::endl;
|
||||
(*output) << " <div id=\"cpuResult\"><pre id=\"cpuContent\"></pre></div>" << std::endl;
|
||||
(*output) << " <br/>" << std::endl;
|
||||
(*output) << " <div id=\"cpuResultGraph\"><pre id=\"cpuContentGraph\"></pre></div>"
|
||||
<< std::endl;
|
||||
(*output) << "</div>" << std::endl;
|
||||
|
||||
// for text profile
|
||||
@ -348,14 +345,14 @@ void cpu_handler(const WebPageHandler::ArgumentMap& args, std::stringstream* out
|
||||
(*output) << " type: \"GET\"," << std::endl;
|
||||
(*output) << " dataType: \"text\"," << std::endl;
|
||||
(*output) << " url: \"pprof/profile?type=text\"," << std::endl;
|
||||
(*output) << " timeout: 60000," << std::endl;
|
||||
(*output) << " timeout: 120000," << std::endl;
|
||||
(*output) << " success: function (result) {" << std::endl;
|
||||
(*output) << " document.getElementById(\"cpuContent\").innerText = result;"
|
||||
<< std::endl;
|
||||
(*output) << " }" << std::endl;
|
||||
(*output) << " ," << std::endl;
|
||||
(*output) << " error: function (result) {" << std::endl;
|
||||
(*output) << " alert(result);" << std::endl;
|
||||
(*output) << " alert(JSON.stringify(result));" << std::endl;
|
||||
(*output) << " }" << std::endl;
|
||||
(*output) << " ," << std::endl;
|
||||
(*output) << " });" << std::endl;
|
||||
@ -363,21 +360,21 @@ void cpu_handler(const WebPageHandler::ArgumentMap& args, std::stringstream* out
|
||||
|
||||
// for graph profile
|
||||
(*output) << "$('#getCpuGraph').click(function () {" << std::endl;
|
||||
(*output) << " document.getElementById(\"cpuContentGraph\").innerText = \"Sampling... (30 "
|
||||
(*output) << " document.getElementById(\"cpuContent\").innerText = \"Sampling... (30 "
|
||||
"seconds)\";"
|
||||
<< std::endl;
|
||||
(*output) << " $.ajax({" << std::endl;
|
||||
(*output) << " type: \"GET\"," << std::endl;
|
||||
(*output) << " dataType: \"text\"," << std::endl;
|
||||
(*output) << " url: \"pprof/profile?type=flamegraph\"," << std::endl;
|
||||
(*output) << " timeout: 60000," << std::endl;
|
||||
(*output) << " timeout: 120000," << std::endl;
|
||||
(*output) << " success: function (result) {" << std::endl;
|
||||
(*output) << " document.getElementById(\"cpuResultGraph\").innerHTML = result;"
|
||||
(*output) << " document.getElementById(\"cpuContent\").innerHTML = result;"
|
||||
<< std::endl;
|
||||
(*output) << " }" << std::endl;
|
||||
(*output) << " ," << std::endl;
|
||||
(*output) << " error: function (result) {" << std::endl;
|
||||
(*output) << " alert(result);" << std::endl;
|
||||
(*output) << " alert(JSON.stringify(result));" << std::endl;
|
||||
(*output) << " }" << std::endl;
|
||||
(*output) << " ," << std::endl;
|
||||
(*output) << " });" << std::endl;
|
||||
|
||||
@ -114,6 +114,7 @@ Status PprofUtils::get_readable_profile(const std::string& file_or_content, bool
|
||||
std::string final_cmd =
|
||||
pprof_cmd + strings::Substitute(" --text $0 $1", self_cmdline, final_file);
|
||||
AgentUtils util;
|
||||
LOG(INFO) << "begin to run command: " << final_cmd;
|
||||
bool rc = util.exec_cmd(final_cmd, &cmd_output, false);
|
||||
|
||||
// delete raw file
|
||||
@ -158,6 +159,7 @@ Status PprofUtils::generate_flamegraph(int32_t sample_seconds,
|
||||
|
||||
AgentUtils util;
|
||||
std::string cmd_output;
|
||||
LOG(INFO) << "begin to run command: " << cmd.str();
|
||||
bool rc = util.exec_cmd(cmd.str(), &cmd_output);
|
||||
if (!rc) {
|
||||
static_cast<void>(io::global_local_filesystem()->delete_file(tmp_file.str()));
|
||||
@ -174,6 +176,7 @@ Status PprofUtils::generate_flamegraph(int32_t sample_seconds,
|
||||
std::stringstream gen_cmd;
|
||||
gen_cmd << perf_cmd << " script -i " << tmp_file.str() << " | " << stackcollapse_perf_pl
|
||||
<< " | " << flamegraph_pl << " > " << graph_file.str();
|
||||
LOG(INFO) << "begin to run command: " << gen_cmd.str();
|
||||
rc = util.exec_cmd(gen_cmd.str(), &res_content);
|
||||
if (!rc) {
|
||||
static_cast<void>(io::global_local_filesystem()->delete_file(tmp_file.str()));
|
||||
@ -185,6 +188,7 @@ Status PprofUtils::generate_flamegraph(int32_t sample_seconds,
|
||||
std::stringstream gen_cmd;
|
||||
gen_cmd << perf_cmd << " script -i " << tmp_file.str() << " | " << stackcollapse_perf_pl
|
||||
<< " | " << flamegraph_pl;
|
||||
LOG(INFO) << "begin to run command: " << gen_cmd.str();
|
||||
rc = util.exec_cmd(gen_cmd.str(), &res_content, false);
|
||||
if (!rc) {
|
||||
static_cast<void>(io::global_local_filesystem()->delete_file(tmp_file.str()));
|
||||
|
||||
@ -826,7 +826,7 @@ Status VFileScanner::_get_next_reader() {
|
||||
std::unique_ptr<ParquetReader> parquet_reader = ParquetReader::create_unique(
|
||||
_profile, *_params, range, _state->query_options().batch_size,
|
||||
const_cast<cctz::time_zone*>(&_state->timezone_obj()), _io_ctx.get(), _state,
|
||||
_shoudl_enable_file_meta_cache() ? ExecEnv::GetInstance()->file_meta_cache()
|
||||
_should_enable_file_meta_cache() ? ExecEnv::GetInstance()->file_meta_cache()
|
||||
: nullptr,
|
||||
_state->query_options().enable_parquet_lazy_mat);
|
||||
{
|
||||
|
||||
@ -243,7 +243,7 @@ private:
|
||||
// 1. max_external_file_meta_cache_num is > 0
|
||||
// 2. the file number is less than 1/3 of cache's capacibility
|
||||
// Otherwise, the cache miss rate will be high
|
||||
bool _shoudl_enable_file_meta_cache() {
|
||||
bool _should_enable_file_meta_cache() {
|
||||
return config::max_external_file_meta_cache_num > 0 &&
|
||||
_split_source->num_scan_ranges() < config::max_external_file_meta_cache_num / 3;
|
||||
}
|
||||
|
||||
4
build.sh
4
build.sh
@ -666,7 +666,8 @@ if [[ "${OUTPUT_BE_BINARY}" -eq 1 ]]; then
|
||||
install -d "${DORIS_OUTPUT}/be/bin" \
|
||||
"${DORIS_OUTPUT}/be/conf" \
|
||||
"${DORIS_OUTPUT}/be/lib" \
|
||||
"${DORIS_OUTPUT}/be/www"
|
||||
"${DORIS_OUTPUT}/be/www" \
|
||||
"${DORIS_OUTPUT}/be/tools/FlameGraph"
|
||||
|
||||
cp -r -p "${DORIS_HOME}/be/output/bin"/* "${DORIS_OUTPUT}/be/bin"/
|
||||
cp -r -p "${DORIS_HOME}/be/output/conf"/* "${DORIS_OUTPUT}/be/conf"/
|
||||
@ -712,6 +713,7 @@ EOF
|
||||
fi
|
||||
|
||||
cp -r -p "${DORIS_HOME}/webroot/be"/* "${DORIS_OUTPUT}/be/www"/
|
||||
cp -r -p "${DORIS_HOME}/tools/FlameGraph"/* "${DORIS_OUTPUT}/be/tools/FlameGraph"/
|
||||
if [[ "${STRIP_DEBUG_INFO}" = "ON" ]]; then
|
||||
cp -r -p "${DORIS_HOME}/be/output/lib/debug_info" "${DORIS_OUTPUT}/be/lib"/
|
||||
fi
|
||||
|
||||
65
dist/LICENSE-dist.txt
vendored
65
dist/LICENSE-dist.txt
vendored
@ -1503,35 +1503,36 @@ LGPL -- licenes/LICENSE-LGPL.txt
|
||||
* gsasl: 1.10.0/1.8.0
|
||||
|
||||
Other dependencies:
|
||||
* libevent: 2.1.12 -- license/LICENSE-libevent.txt
|
||||
* openssl: 1.1.1s -- license/LICENSE-openssl.txt
|
||||
* gflag: 2.2.2 -- license/LICENSE-gflag.txt
|
||||
* glog: 0.4.0 -- license/LICENSE-glog.txt
|
||||
* gtest: 1.11.0 -- license/LICENSE-gtest.txt
|
||||
* snappy: 1.1.8 -- license/LICENSE-snappy.txt
|
||||
* gperftools: 2.9.1 -- license/LICENSE-gperftools.txt
|
||||
* zlib: 1.2.11 -- license/LICENSE-zlib.txt
|
||||
* lz4: 1.9.3 -- license/LICENSE-lz4.txt
|
||||
* bzip2: 1.0.8 -- license/LICENSE-bzip2.txt
|
||||
* rapidjson@1a803826 -- license/LICENSE-rapidjson.txt
|
||||
* curl: 7.79.0 -- license/LICENSE-curl.txt
|
||||
* re2: 2021-02-02 -- license/LICENSE-re2.txt
|
||||
* hyperscan: 5.4.0 -- license/LICENSE-hyperscan.txt
|
||||
* vectorscan: 5.4.7 -- license/LICENSE-vectorscan.txt
|
||||
* boost: 1.73.0 -- license/LICENSE-boost.txt
|
||||
* unixodbc: 2.3.7 -- license/LICENSE-unixodbc.txt
|
||||
* leveldb: 1.23 -- license/LICENSE-leveldb.txt
|
||||
* cyrus-sasl: 2.1.27 -- license/LICENSE-cyrus-sasl.txt
|
||||
* librdkafka: 1.8.2 -- license/LICENSE-librdkafka.txt
|
||||
* zstd: 1.5.2 -- license/LICENSE-zstd.txt
|
||||
* brotli: 1.0.9 -- license/LICENSE-brotli.txt
|
||||
* bitshuffle: 0.5.1 -- license/LICENSE-bigshuffle.txt
|
||||
* fmt: 7.1.3 -- license/LICENSE-fmt.txt
|
||||
* jemalloc: 5.3.0 -- license/LICENSE-jemolloc.txt
|
||||
* lzma@master -- license/LICENSE-lzma.txt
|
||||
* libdivide: 5.0 -- license/LICENSE-libdivide.txt
|
||||
* pdqsort: 0.0.0+git20180419 -- license/LICENSE-pdqsort.txt
|
||||
* breakpad@38ee0be -- license/LICENSE-breakpod.txt
|
||||
* xsimd: xmid@e9234cd6 -- license/LICENSE-xsimd.txt
|
||||
* xxhash: 0.8.1 -- license/LICENSE-xxhash.txt
|
||||
* concurrentqueue: 1.0.3 -- license/LICENSE-concurrentqueue.txt
|
||||
* libevent: 2.1.12 -- licenses/LICENSE-libevent.txt
|
||||
* openssl: 1.1.1s -- licenses/LICENSE-openssl.txt
|
||||
* gflag: 2.2.2 -- licenses/LICENSE-gflag.txt
|
||||
* glog: 0.4.0 -- licenses/LICENSE-glog.txt
|
||||
* gtest: 1.11.0 -- licenses/LICENSE-gtest.txt
|
||||
* snappy: 1.1.8 -- licenses/LICENSE-snappy.txt
|
||||
* gperftools: 2.9.1 -- licenses/LICENSE-gperftools.txt
|
||||
* zlib: 1.2.11 -- licenses/LICENSE-zlib.txt
|
||||
* lz4: 1.9.3 -- licenses/LICENSE-lz4.txt
|
||||
* bzip2: 1.0.8 -- licenses/LICENSE-bzip2.txt
|
||||
* rapidjson@1a803826 -- licenses/LICENSE-rapidjson.txt
|
||||
* curl: 7.79.0 -- licenses/LICENSE-curl.txt
|
||||
* re2: 2021-02-02 -- licenses/LICENSE-re2.txt
|
||||
* hyperscan: 5.4.0 -- licenses/LICENSE-hyperscan.txt
|
||||
* vectorscan: 5.4.7 -- licenses/LICENSE-vectorscan.txt
|
||||
* boost: 1.73.0 -- licenses/LICENSE-boost.txt
|
||||
* unixodbc: 2.3.7 -- licenses/LICENSE-unixodbc.txt
|
||||
* leveldb: 1.23 -- licenses/LICENSE-leveldb.txt
|
||||
* cyrus-sasl: 2.1.27 -- licenses/LICENSE-cyrus-sasl.txt
|
||||
* librdkafka: 1.8.2 -- licenses/LICENSE-librdkafka.txt
|
||||
* zstd: 1.5.2 -- licenses/LICENSE-zstd.txt
|
||||
* brotli: 1.0.9 -- licenses/LICENSE-brotli.txt
|
||||
* bitshuffle: 0.5.1 -- licenses/LICENSE-bigshuffle.txt
|
||||
* fmt: 7.1.3 -- licenses/LICENSE-fmt.txt
|
||||
* jemalloc: 5.3.0 -- licenses/LICENSE-jemolloc.txt
|
||||
* lzma@master -- licenses/LICENSE-lzma.txt
|
||||
* libdivide: 5.0 -- licenses/LICENSE-libdivide.txt
|
||||
* pdqsort: 0.0.0+git20180419 -- licenses/LICENSE-pdqsort.txt
|
||||
* breakpad@38ee0be -- licenses/LICENSE-breakpod.txt
|
||||
* xsimd: xmid@e9234cd6 -- licenses/LICENSE-xsimd.txt
|
||||
* xxhash: 0.8.1 -- licenses/LICENSE-xxhash.txt
|
||||
* concurrentqueue: 1.0.3 -- licenses/LICENSE-concurrentqueue.txt
|
||||
* FlameGraph -- licenses/LICENSE-CDDL-1.0.txt
|
||||
|
||||
13
tools/FlameGraph/README
Normal file
13
tools/FlameGraph/README
Normal file
@ -0,0 +1,13 @@
|
||||
These 2 files:
|
||||
|
||||
- flamegraph.pl
|
||||
- stackcollapse-perf.pl
|
||||
|
||||
are copied from:
|
||||
|
||||
https://github.com/brendangregg/FlameGraph/blob/master/flamegraph.pl
|
||||
https://github.com/brendangregg/FlameGraph/blob/master/stackcollapse-perf.pl
|
||||
|
||||
which are under license:
|
||||
|
||||
https://github.com/brendangregg/FlameGraph/blob/master/docs/cddl1.txt
|
||||
1302
tools/FlameGraph/flamegraph.pl
Executable file
1302
tools/FlameGraph/flamegraph.pl
Executable file
File diff suppressed because it is too large
Load Diff
435
tools/FlameGraph/stackcollapse-perf.pl
Executable file
435
tools/FlameGraph/stackcollapse-perf.pl
Executable file
@ -0,0 +1,435 @@
|
||||
#!/usr/bin/perl -w
|
||||
#
|
||||
# stackcollapse-perf.pl collapse perf samples into single lines.
|
||||
#
|
||||
# Parses a list of multiline stacks generated by "perf script", and
|
||||
# outputs a semicolon separated stack followed by a space and a count.
|
||||
# If memory addresses (+0xd) are present, they are stripped, and resulting
|
||||
# identical stacks are colased with their counts summed.
|
||||
#
|
||||
# USAGE: ./stackcollapse-perf.pl [options] infile > outfile
|
||||
#
|
||||
# Run "./stackcollapse-perf.pl -h" to list options.
|
||||
#
|
||||
# Example input:
|
||||
#
|
||||
# swapper 0 [000] 158665.570607: cpu-clock:
|
||||
# ffffffff8103ce3b native_safe_halt ([kernel.kallsyms])
|
||||
# ffffffff8101c6a3 default_idle ([kernel.kallsyms])
|
||||
# ffffffff81013236 cpu_idle ([kernel.kallsyms])
|
||||
# ffffffff815bf03e rest_init ([kernel.kallsyms])
|
||||
# ffffffff81aebbfe start_kernel ([kernel.kallsyms].init.text)
|
||||
# [...]
|
||||
#
|
||||
# Example output:
|
||||
#
|
||||
# swapper;start_kernel;rest_init;cpu_idle;default_idle;native_safe_halt 1
|
||||
#
|
||||
# Input may be created and processed using:
|
||||
#
|
||||
# perf record -a -g -F 997 sleep 60
|
||||
# perf script | ./stackcollapse-perf.pl > out.stacks-folded
|
||||
#
|
||||
# The output of "perf script" should include stack traces. If these are missing
|
||||
# for you, try manually selecting the perf script output; eg:
|
||||
#
|
||||
# perf script -f comm,pid,tid,cpu,time,event,ip,sym,dso,trace | ...
|
||||
#
|
||||
# This is also required for the --pid or --tid options, so that the output has
|
||||
# both the PID and TID.
|
||||
#
|
||||
# Copyright 2012 Joyent, Inc. All rights reserved.
|
||||
# Copyright 2012 Brendan Gregg. All rights reserved.
|
||||
#
|
||||
# CDDL HEADER START
|
||||
#
|
||||
# The contents of this file are subject to the terms of the
|
||||
# Common Development and Distribution License (the "License").
|
||||
# You may not use this file except in compliance with the License.
|
||||
#
|
||||
# You can obtain a copy of the license at docs/cddl1.txt or
|
||||
# http://opensource.org/licenses/CDDL-1.0.
|
||||
# See the License for the specific language governing permissions
|
||||
# and limitations under the License.
|
||||
#
|
||||
# When distributing Covered Code, include this CDDL HEADER in each
|
||||
# file and include the License file at docs/cddl1.txt.
|
||||
# If applicable, add the following below this CDDL HEADER, with the
|
||||
# fields enclosed by brackets "[]" replaced with your own identifying
|
||||
# information: Portions Copyright [yyyy] [name of copyright owner]
|
||||
#
|
||||
# CDDL HEADER END
|
||||
#
|
||||
# 02-Mar-2012 Brendan Gregg Created this.
|
||||
# 02-Jul-2014 " " Added process name to stacks.
|
||||
|
||||
use strict;
|
||||
use Getopt::Long;
|
||||
|
||||
my %collapsed;
|
||||
|
||||
sub remember_stack {
|
||||
my ($stack, $count) = @_;
|
||||
$collapsed{$stack} += $count;
|
||||
}
|
||||
my $annotate_kernel = 0; # put an annotation on kernel function
|
||||
my $annotate_jit = 0; # put an annotation on jit symbols
|
||||
my $annotate_all = 0; # enale all annotations
|
||||
my $include_pname = 1; # include process names in stacks
|
||||
my $include_pid = 0; # include process ID with process name
|
||||
my $include_tid = 0; # include process & thread ID with process name
|
||||
my $include_addrs = 0; # include raw address where a symbol can't be found
|
||||
my $tidy_java = 1; # condense Java signatures
|
||||
my $tidy_generic = 1; # clean up function names a little
|
||||
my $target_pname; # target process name from perf invocation
|
||||
my $event_filter = ""; # event type filter, defaults to first encountered event
|
||||
my $event_defaulted = 0; # whether we defaulted to an event (none provided)
|
||||
my $event_warning = 0; # if we printed a warning for the event
|
||||
|
||||
my $show_inline = 0;
|
||||
my $show_context = 0;
|
||||
|
||||
my $srcline_in_input = 0; # if there are extra lines with source location (perf script -F+srcline)
|
||||
GetOptions('inline' => \$show_inline,
|
||||
'context' => \$show_context,
|
||||
'srcline' => \$srcline_in_input,
|
||||
'pid' => \$include_pid,
|
||||
'kernel' => \$annotate_kernel,
|
||||
'jit' => \$annotate_jit,
|
||||
'all' => \$annotate_all,
|
||||
'tid' => \$include_tid,
|
||||
'addrs' => \$include_addrs,
|
||||
'event-filter=s' => \$event_filter)
|
||||
or die <<USAGE_END;
|
||||
USAGE: $0 [options] infile > outfile\n
|
||||
--pid # include PID with process names [1]
|
||||
--tid # include TID and PID with process names [1]
|
||||
--inline # un-inline using addr2line
|
||||
--all # all annotations (--kernel --jit)
|
||||
--kernel # annotate kernel functions with a _[k]
|
||||
--jit # annotate jit functions with a _[j]
|
||||
--context # adds source context to --inline
|
||||
--srcline # parses output of 'perf script -F+srcline' and adds source context
|
||||
--addrs # include raw addresses where symbols can't be found
|
||||
--event-filter=EVENT # event name filter\n
|
||||
[1] perf script must emit both PID and TIDs for these to work; eg, Linux < 4.1:
|
||||
perf script -f comm,pid,tid,cpu,time,event,ip,sym,dso,trace
|
||||
for Linux >= 4.1:
|
||||
perf script -F comm,pid,tid,cpu,time,event,ip,sym,dso,trace
|
||||
If you save this output add --header on Linux >= 3.14 to include perf info.
|
||||
USAGE_END
|
||||
|
||||
if ($annotate_all) {
|
||||
$annotate_kernel = $annotate_jit = 1;
|
||||
}
|
||||
|
||||
my %inlineCache;
|
||||
|
||||
my %nmCache;
|
||||
|
||||
sub inlineCacheAdd {
|
||||
my ($pc, $mod, $result) = @_;
|
||||
if (defined($inlineCache{$pc})) {
|
||||
$inlineCache{$pc}{$mod} = $result;
|
||||
} else {
|
||||
$inlineCache{$pc} = {$mod => $result};
|
||||
}
|
||||
}
|
||||
|
||||
# for the --inline option
|
||||
sub inline {
|
||||
my ($pc, $rawfunc, $mod) = @_;
|
||||
|
||||
return $inlineCache{$pc}{$mod} if defined($inlineCache{$pc}{$mod});
|
||||
|
||||
# capture addr2line output
|
||||
my $a2l_output = `addr2line -a $pc -e $mod -i -f -s -C`;
|
||||
|
||||
# remove first line
|
||||
$a2l_output =~ s/^(.*\n){1}//;
|
||||
|
||||
if ($a2l_output =~ /\?\?\n\?\?:0/) {
|
||||
# if addr2line fails and rawfunc is func+offset, then fall back to it
|
||||
if ($rawfunc =~ /^(.+)\+0x([0-9a-f]+)$/) {
|
||||
my $func = $1;
|
||||
my $addr = hex $2;
|
||||
|
||||
$nmCache{$mod}=`nm $mod` unless defined $nmCache{$mod};
|
||||
|
||||
if ($nmCache{$mod} =~ /^([0-9a-f]+) . \Q$func\E$/m) {
|
||||
my $base = hex $1;
|
||||
my $newPc = sprintf "0x%x", $base+$addr;
|
||||
my $result = inline($newPc, '', $mod);
|
||||
inlineCacheAdd($pc, $mod, $result);
|
||||
return $result;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
my @fullfunc;
|
||||
my $one_item = "";
|
||||
for (split /^/, $a2l_output) {
|
||||
chomp $_;
|
||||
|
||||
# remove discriminator info if exists
|
||||
$_ =~ s/ \(discriminator \S+\)//;
|
||||
|
||||
if ($one_item eq "") {
|
||||
$one_item = $_;
|
||||
} else {
|
||||
if ($show_context == 1) {
|
||||
unshift @fullfunc, $one_item . ":$_";
|
||||
} else {
|
||||
unshift @fullfunc, $one_item;
|
||||
}
|
||||
$one_item = "";
|
||||
}
|
||||
}
|
||||
|
||||
my $result = join ";" , @fullfunc;
|
||||
|
||||
inlineCacheAdd($pc, $mod, $result);
|
||||
|
||||
return $result;
|
||||
}
|
||||
|
||||
my @stack;
|
||||
my $pname;
|
||||
my $m_pid;
|
||||
my $m_tid;
|
||||
my $m_period;
|
||||
|
||||
#
|
||||
# Main loop
|
||||
#
|
||||
while (defined($_ = <>)) {
|
||||
|
||||
# find the name of the process launched by perf, by stepping backwards
|
||||
# over the args to find the first non-option (no dash):
|
||||
if (/^# cmdline/) {
|
||||
my @args = split ' ', $_;
|
||||
foreach my $arg (reverse @args) {
|
||||
if ($arg !~ /^-/) {
|
||||
$target_pname = $arg;
|
||||
$target_pname =~ s:.*/::; # strip pathname
|
||||
last;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
# skip remaining comments
|
||||
next if m/^#/;
|
||||
chomp;
|
||||
|
||||
# end of stack. save cached data.
|
||||
if (m/^$/) {
|
||||
# ignore filtered samples
|
||||
next if not $pname;
|
||||
|
||||
if ($include_pname) {
|
||||
if (defined $pname) {
|
||||
unshift @stack, $pname;
|
||||
} else {
|
||||
unshift @stack, "";
|
||||
}
|
||||
}
|
||||
remember_stack(join(";", @stack), $m_period) if @stack;
|
||||
undef @stack;
|
||||
undef $pname;
|
||||
next;
|
||||
}
|
||||
|
||||
#
|
||||
# event record start
|
||||
#
|
||||
if (/^(\S.+?)\s+(\d+)\/*(\d+)*\s+/) {
|
||||
# default "perf script" output has TID but not PID
|
||||
# eg, "java 25607 4794564.109216: 1 cycles:"
|
||||
# eg, "java 12688 [002] 6544038.708352: 235 cpu-clock:"
|
||||
# eg, "V8 WorkerThread 25607 4794564.109216: 104345 cycles:"
|
||||
# eg, "java 24636/25607 [000] 4794564.109216: 1 cycles:"
|
||||
# eg, "java 12688/12764 6544038.708352: 10309278 cpu-clock:"
|
||||
# eg, "V8 WorkerThread 24636/25607 [000] 94564.109216: 100 cycles:"
|
||||
# other combinations possible
|
||||
my ($comm, $pid, $tid, $period) = ($1, $2, $3, "");
|
||||
if (not $tid) {
|
||||
$tid = $pid;
|
||||
$pid = "?";
|
||||
}
|
||||
|
||||
if (/:\s*(\d+)*\s+(\S+):\s*$/) {
|
||||
$period = $1;
|
||||
my $event = $2;
|
||||
|
||||
if ($event_filter eq "") {
|
||||
# By default only show events of the first encountered
|
||||
# event type. Merging together different types, such as
|
||||
# instructions and cycles, produces misleading results.
|
||||
$event_filter = $event;
|
||||
$event_defaulted = 1;
|
||||
} elsif ($event ne $event_filter) {
|
||||
if ($event_defaulted and $event_warning == 0) {
|
||||
# only print this warning if necessary:
|
||||
# when we defaulted and there was
|
||||
# multiple event types.
|
||||
print STDERR "Filtering for events of type: $event\n";
|
||||
$event_warning = 1;
|
||||
}
|
||||
next;
|
||||
}
|
||||
}
|
||||
|
||||
if (not $period) {
|
||||
$period = 1
|
||||
}
|
||||
($m_pid, $m_tid, $m_period) = ($pid, $tid, $period);
|
||||
|
||||
if ($include_tid) {
|
||||
$pname = "$comm-$m_pid/$m_tid";
|
||||
} elsif ($include_pid) {
|
||||
$pname = "$comm-$m_pid";
|
||||
} else {
|
||||
$pname = "$comm";
|
||||
}
|
||||
$pname =~ tr/ /_/;
|
||||
|
||||
#
|
||||
# stack line
|
||||
#
|
||||
} elsif (/^\s*(\w+)\s*(.+) \((.*)\)/) {
|
||||
# ignore filtered samples
|
||||
next if not $pname;
|
||||
|
||||
my ($pc, $rawfunc, $mod) = ($1, $2, $3);
|
||||
|
||||
if ($show_inline == 1 && $mod !~ m/(perf-\d+.map|kernel\.|\[[^\]]+\])/) {
|
||||
my $inlineRes = inline($pc, $rawfunc, $mod);
|
||||
# - empty result this happens e.g., when $mod does not exist or is a path to a compressed kernel module
|
||||
# if this happens, the user will see error message from addr2line written to stderr
|
||||
# - if addr2line results in "??" , then it's much more sane to fall back than produce a '??' in graph
|
||||
if($inlineRes ne "" and $inlineRes ne "??" and $inlineRes ne "??:??:0" ) {
|
||||
unshift @stack, $inlineRes;
|
||||
next;
|
||||
}
|
||||
}
|
||||
|
||||
# Linux 4.8 included symbol offsets in perf script output by default, eg:
|
||||
# 7fffb84c9afc cpu_startup_entry+0x800047c022ec ([kernel.kallsyms])
|
||||
# strip these off:
|
||||
$rawfunc =~ s/\+0x[\da-f]+$//;
|
||||
|
||||
next if $rawfunc =~ /^\(/; # skip process names
|
||||
|
||||
my $is_unknown=0;
|
||||
my @inline;
|
||||
for (split /\->/, $rawfunc) {
|
||||
my $func = $_;
|
||||
|
||||
if ($func eq "[unknown]") {
|
||||
if ($mod ne "[unknown]") { # use module name instead, if known
|
||||
$func = $mod;
|
||||
$func =~ s/.*\///;
|
||||
} else {
|
||||
$func = "unknown";
|
||||
$is_unknown=1;
|
||||
}
|
||||
|
||||
if ($include_addrs) {
|
||||
$func = "\[$func \<$pc\>\]";
|
||||
} else {
|
||||
$func = "\[$func\]";
|
||||
}
|
||||
}
|
||||
|
||||
if ($tidy_generic) {
|
||||
$func =~ s/;/:/g;
|
||||
if ($func !~ m/\.\(.*\)\./) {
|
||||
# This doesn't look like a Go method name (such as
|
||||
# "net/http.(*Client).Do"), so everything after the first open
|
||||
# paren (that is not part of an "(anonymous namespace)") is
|
||||
# just noise.
|
||||
$func =~ s/\((?!anonymous namespace\)).*//;
|
||||
}
|
||||
# now tidy this horrible thing:
|
||||
# 13a80b608e0a RegExp:[&<>\"\'] (/tmp/perf-7539.map)
|
||||
$func =~ tr/"\'//d;
|
||||
# fall through to $tidy_java
|
||||
}
|
||||
|
||||
if ($tidy_java and $pname =~ m/^java/) {
|
||||
# along with $tidy_generic, converts the following:
|
||||
# Lorg/mozilla/javascript/ContextFactory;.call(Lorg/mozilla/javascript/ContextAction;)Ljava/lang/Object;
|
||||
# Lorg/mozilla/javascript/ContextFactory;.call(Lorg/mozilla/javascript/C
|
||||
# Lorg/mozilla/javascript/MemberBox;.<init>(Ljava/lang/reflect/Method;)V
|
||||
# into:
|
||||
# org/mozilla/javascript/ContextFactory:.call
|
||||
# org/mozilla/javascript/ContextFactory:.call
|
||||
# org/mozilla/javascript/MemberBox:.init
|
||||
$func =~ s/^L// if $func =~ m:/:;
|
||||
}
|
||||
|
||||
#
|
||||
# Annotations
|
||||
#
|
||||
# detect inlined from the @inline array
|
||||
# detect kernel from the module name; eg, frames to parse include:
|
||||
# ffffffff8103ce3b native_safe_halt ([kernel.kallsyms])
|
||||
# 8c3453 tcp_sendmsg (/lib/modules/4.3.0-rc1-virtual/build/vmlinux)
|
||||
# 7d8 ipv4_conntrack_local+0x7f8f80b8 ([nf_conntrack_ipv4])
|
||||
# detect jit from the module name; eg:
|
||||
# 7f722d142778 Ljava/io/PrintStream;::print (/tmp/perf-19982.map)
|
||||
if (scalar(@inline) > 0) {
|
||||
$func .= "_[i]" unless $func =~ m/\_\[i\]/; # inlined
|
||||
} elsif ($annotate_kernel == 1 && $mod =~ m/(^\[|vmlinux$)/ && $mod !~ /unknown/) {
|
||||
$func .= "_[k]"; # kernel
|
||||
} elsif ($annotate_jit == 1 && $mod =~ m:/tmp/perf-\d+\.map:) {
|
||||
$func .= "_[j]" unless $func =~ m/\_\[j\]/; # jitted
|
||||
}
|
||||
|
||||
#
|
||||
# Source lines
|
||||
#
|
||||
#
|
||||
# Sample outputs:
|
||||
# | a.out 35081 252436.005167: 667783 cycles:
|
||||
# | 408ebb some_method_name+0x8b (/full/path/to/a.out)
|
||||
# | uniform_int_dist.h:300
|
||||
# | 4069f5 main+0x935 (/full/path/to/a.out)
|
||||
# | file.cpp:137
|
||||
# | 7f6d2148eb25 __libc_start_main+0xd5 (/lib64/libc-2.33.so)
|
||||
# | libc-2.33.so[27b25]
|
||||
#
|
||||
# | a.out 35081 252435.738165: 306459 cycles:
|
||||
# | 7f6d213c2750 [unknown] (/usr/lib64/libkmod.so.2.3.6)
|
||||
# | libkmod.so.2.3.6[6750]
|
||||
#
|
||||
# | a.out 35081 252435.738373: 315813 cycles:
|
||||
# | 7f6d215ca51b __strlen_avx2+0x4b (/lib64/libc-2.33.so)
|
||||
# | libc-2.33.so[16351b]
|
||||
# | 7ffc71ee9580 [unknown] ([unknown])
|
||||
# |
|
||||
#
|
||||
# | a.out 35081 252435.718940: 247984 cycles:
|
||||
# | ffffffff814f9302 up_write+0x32 ([kernel.kallsyms])
|
||||
# | [kernel.kallsyms][ffffffff814f9302]
|
||||
if($srcline_in_input and not $is_unknown){
|
||||
$_ = <>;
|
||||
chomp;
|
||||
s/\[.*?\]//g;
|
||||
s/^\s*//g;
|
||||
s/\s*$//g;
|
||||
$func.=':'.$_ unless $_ eq "";
|
||||
}
|
||||
|
||||
push @inline, $func;
|
||||
}
|
||||
|
||||
unshift @stack, @inline;
|
||||
} else {
|
||||
warn "Unrecognized line: $_";
|
||||
}
|
||||
}
|
||||
|
||||
foreach my $k (sort { $a cmp $b } keys %collapsed) {
|
||||
print "$k $collapsed{$k}\n";
|
||||
}
|
||||
Reference in New Issue
Block a user