Skip to contentSkip to conversation

Add support for Firefox's gecko profile format

PATCH Superseded Anup Sharma 3 participants 14 messages Jul 18, 2023 – Jul 20, 2023 perf events

v4 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v4 cover Add support for Firefox's gecko profile format

This patch series adds support for Firefox's gecko profile format. The format is documented here [1].

I have incorporated several changes based on feedback from the previous version of the patch.

Changes in v3:
- Implemented object-oriented programming (OOP) approach for Thread and Sample to improve code organization and maintainability.
- Enhanced user experience by introducing argparse for changing color settings of user and kernel symbols during execution.
- Ensured proper module encapsulation by adding imports wherever necessary.
- Improved code readability by adding descriptive comments and type annotations.

These changes focus on adopting OOP principles, enhancing user interaction with argparse, and making the code more readable with comments and type information.

TODO:
- use cpu_mode to determine the category instead of finding it from symbol name.
- Write a test.
- add direct execution command for this script under script/python/bin directory.

Committer Testing:
- Tested with a perf.data file generated for single and multiple cpu cores enabled.
- Uploaded the stdout into profiler.firefox.com and verified the output.
- Verified the output with the output generated by the existing script as mentioned here [2].
Method:
- perf record -F 99 -a -g -- sleep 5
- perf script firefox-gecko-converter.py > output.json
- upload output.json to profiler.firefox.com

[1] https://github.com/firefox-devtools/profiler/blob/main/docs-developer/gecko-profile-format.md
[2] https://perf.wiki.kernel.org/index.php/Tutorial#Firefox_Profiler

Anup Sharma (6):
perf scripts python: Add initial script file with usage information perf scripts python: Extact necessary information from process event perf scripts python: Add classes and conversion functions perf scripts python: Add trace end processing and PRODUCT and
CATEGORIES information
perf scripts python: implement internal get or create frame, stack and
string function
perf scripts python: Implement add sample function and thread
processing

 .../scripts/python/firefox-gecko-converter.py | 339 ++++++++++++++++++
 1 file changed, 339 insertions(+)
 create mode 100644 tools/perf/scripts/python/firefox-gecko-converter.py
Signature · 1 line
2.34.1
#1 lore raw

PATCH v4 1/6 perf scripts python: Add initial script file with usage information

Added necessary modules, including the Perf-Trace-Util library, and defines the required functions and variables for using perf script python. The perf_trace_context and Core modules for tracing and processing events has been also imported. Also added usage information.

---
 .../scripts/python/firefox-gecko-converter.py | 31 +++++++++++++++++++
 1 file changed, 31 insertions(+)
 create mode 100644 tools/perf/scripts/python/firefox-gecko-converter.py
tools/perf/scripts/python/firefox-gecko-converter.py new file +31 -0
new file mode 100644
index 000000000000..63ecaf1abee8
@@ -0,0 +1,31 @@
+# firefox-gecko-converter.py - Convert perf record output to Firefox's gecko profile format
+# SPDX-License-Identifier: GPL-2.0
+#
+# The script converts perf.data to Gecko Profile Format,
+# which can be read by https://profiler.firefox.com/.
+#
+# Usage:
+#
+#     perf record -a -g -F 99 sleep 60
+#     perf script firefox-gecko-converter.py > output.json
+
+import os
+import sys
+from typing import Dict
+
+# Add the Perf-Trace-Util library to the Python path
+sys.path.append(os.environ['PERF_EXEC_PATH'] + \
+	'/scripts/python/Perf-Trace-Util/lib/Perf/Trace')
+
+from perf_trace_context import *
+from Core import *
+
+# Uses perf script python interface to parse each
+# event and store the data in the thread builder.
+def process_event(param_dict: Dict) -> None:
+	pass
+
+# Trace_end runs at the end and will be used to aggregate
+# the data into the final json object and print it out to stdout.
+def trace_end() -> None:
+	pass
Signature · 1 line
2.34.1
#2 lore raw

PATCH v4 2/6 perf scripts python: Extact necessary information from process event

In reply to Anup Sharma

The script takes in a sample event dictionary(param_dict) and retrieves relevant data such as time stamp, PID, TID, and comm for each event. Also start time is defined as a global variable as it need to be passed to trace_end for gecko meta information field creation.

---
 .../perf/scripts/python/firefox-gecko-converter.py | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)
tools/perf/scripts/python/firefox-gecko-converter.py +13 -1
index 63ecaf1abee8..bfc8c2b026bf 100644
@@ -20,10 +20,22 @@ sys.path.append(os.environ['PERF_EXEC_PATH'] + \
 from perf_trace_context import *
 from Core import *
 
+# start_time is intialiazed only once for the all event traces.
+start_time = None
+
 # Uses perf script python interface to parse each
 # event and store the data in the thread builder.
 def process_event(param_dict: Dict) -> None:
-	pass
+	global start_time
+	global tid_to_thread
+	time_stamp = (param_dict['sample']['time'] // 1000) / 1000
+	pid = param_dict['sample']['pid']
+	tid = param_dict['sample']['tid']
+	comm = param_dict['comm']
+
+	# Start time is the time of the first sample
+	if not start_time:
+		start_time = time_stamp
 
 # Trace_end runs at the end and will be used to aggregate
 # the data into the final json object and print it out to stdout.
Signature · 1 line
2.34.1
#3 lore raw

PATCH v4 3/6 perf scripts python: Add classes and conversion functions

In reply to Anup Sharma

This commit introduces new classes and conversion functions to facilitate the representation of Gecko profile information. The new classes Frame, Stack, Sample, and Thread are added to handle specific components of the profile data, also link to the origin docs has been commented out.

Additionally, Inside the Thread class _to_json_dict() method has been created that converts the current thread data into the corresponding format expected by the GeckoThread JSON schema, as per the Gecko profile format specification.

---
 .../scripts/python/firefox-gecko-converter.py | 134 +++++++++++++++++-
 1 file changed, 133 insertions(+), 1 deletion(-)
tools/perf/scripts/python/firefox-gecko-converter.py +133 -1
index bfc8c2b026bf..d9b1ec18997a 100644
@@ -11,7 +11,8 @@
 
 import os
 import sys
-from typing import Dict
+from dataclasses import dataclass, field
+from typing import List, Dict, Optional, NamedTuple, Set, Tuple, Any
 
 # Add the Perf-Trace-Util library to the Python path
 sys.path.append(os.environ['PERF_EXEC_PATH'] + \
@@ -20,9 +21,140 @@ sys.path.append(os.environ['PERF_EXEC_PATH'] + \
 from perf_trace_context import *
 from Core import *
 
+StringID = int
+StackID = int
+FrameID = int
+CategoryID = int
+Milliseconds = float
+
 # start_time is intialiazed only once for the all event traces.
 start_time = None
 
+# https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L156
+class Frame(NamedTuple):
+	string_id: StringID
+	relevantForJS: bool
+	innerWindowID: int
+	implementation: None
+	optimizations: None
+	line: None
+	column: None
+	category: CategoryID
+	subcategory: int
+
+# https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L216
+class Stack(NamedTuple):
+	prefix_id: Optional[StackID]
+	frame_id: FrameID
+
+# https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L90
+class Sample(NamedTuple):
+	stack_id: Optional[StackID]
+	time_ms: Milliseconds
+	responsiveness: int
+
+@dataclass
+class Thread:
+	"""A builder for a profile of the thread.
+
+	Attributes:
+		comm: Thread command-line (name).
+		pid: process ID of containing process.
+		tid: thread ID.
+		samples: Timeline of profile samples.
+		frameTable: interned stack frame ID -> stack frame.
+		stringTable: interned string ID -> string.
+		stringMap: interned string -> string ID.
+		stackTable: interned stack ID -> stack.
+		stackMap: (stack prefix ID, leaf stack frame ID) -> interned Stack ID.
+		frameMap: Stack Frame string -> interned Frame ID.
+		comm: str
+		pid: int
+		tid: int
+		samples: List[Sample] = field(default_factory=list)
+		frameTable: List[Frame] = field(default_factory=list)
+		stringTable: List[str] = field(default_factory=list)
+		stringMap: Dict[str, int] = field(default_factory=dict)
+		stackTable: List[Stack] = field(default_factory=list)
+		stackMap: Dict[Tuple[Optional[int], int], int] = field(default_factory=dict)
+		frameMap: Dict[str, int] = field(default_factory=dict)
+	"""
+	comm: str
+	pid: int
+	tid: int
+	samples: List[Sample] = field(default_factory=list)
+	frameTable: List[Frame] = field(default_factory=list)
+	stringTable: List[str] = field(default_factory=list)
+	stringMap: Dict[str, int] = field(default_factory=dict)
+	stackTable: List[Stack] = field(default_factory=list)
+	stackMap: Dict[Tuple[Optional[int], int], int] = field(default_factory=dict)
+	frameMap: Dict[str, int] = field(default_factory=dict)
+
+	def _to_json_dict(self) -> Dict:
+		"""Converts current Thread to GeckoThread JSON format."""
+		# Gecko profile format is row-oriented data as List[List],
+		# And a schema for interpreting each index.
+		# Schema:
+		# https://github.com/firefox-devtools/profiler/blob/main/docs-developer/gecko-profile-format.md
+		# https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L230
+		return {
+			"tid": self.tid,
+			"pid": self.pid,
+			"name": self.comm,
+			# https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L51
+			"markers": {
+				"schema": {
+					"name": 0,
+					"startTime": 1,
+					"endTime": 2,
+					"phase": 3,
+					"category": 4,
+					"data": 5,
+				},
+				"data": [],
+			},
+
+			# https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L90
+			"samples": {
+				"schema": {
+					"stack": 0,
+					"time": 1,
+					"responsiveness": 2,
+				},
+				"data": self.samples
+			},
+
+			# https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L156
+			"frameTable": {
+				"schema": {
+					"location": 0,
+					"relevantForJS": 1,
+					"innerWindowID": 2,
+					"implementation": 3,
+					"optimizations": 4,
+					"line": 5,
+					"column": 6,
+					"category": 7,
+					"subcategory": 8,
+				},
+				"data": self.frameTable,
+			},
+
+			# https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L216
+			"stackTable": {
+				"schema": {
+					"prefix": 0,
+					"frame": 1,
+				},
+				"data": self.stackTable,
+			},
+			"stringTable": self.stringTable,
+			"registerTime": 0,
+			"unregisterTime": None,
+			"processType": "default",
+		}
+
+
 # Uses perf script python interface to parse each
 # event and store the data in the thread builder.
 def process_event(param_dict: Dict) -> None:
Signature · 1 line
2.34.1
#4 lore raw

PATCH v4 4/6 perf scripts python: Add trace end processing and PRODUCT and CATEGORIES information

In reply to Anup Sharma

The final output will now be presented in JSON format following the Gecko profile structure. Additionally, the inclusion of PRODUCT allows easy retrieval of header information for UI.

Furthermore, CATEGORIES have been introduced to enable customization of kernel and user colors using input arguments. To facilitate this functionality, an argparse-based parser has been implemented.

Note that the implementation of threads will be addressed in subsequent commits."

---
 .../scripts/python/firefox-gecko-converter.py | 64 ++++++++++++++++++-
 1 file changed, 63 insertions(+), 1 deletion(-)
tools/perf/scripts/python/firefox-gecko-converter.py +63 -1
index d9b1ec18997a..a0218e2245f2 100644
@@ -11,6 +11,8 @@
 
 import os
 import sys
+import json
+import argparse
 from dataclasses import dataclass, field
 from typing import List, Dict, Optional, NamedTuple, Set, Tuple, Any
 
@@ -30,6 +32,13 @@ Milliseconds = float
 # start_time is intialiazed only once for the all event traces.
 start_time = None
 
+# https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/profile.js#L425
+# Follow Brendan Gregg's Flamegraph convention: orange for kernel and yellow for user space by default.
+CATEGORIES = None
+
+# The product name is used by the profiler UI to show the Operating system and Processor.
+PRODUCT = os.popen('uname -op').read().strip()
+
 # https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L156
 class Frame(NamedTuple):
 	string_id: StringID
@@ -172,4 +181,57 @@ def process_event(param_dict: Dict) -> None:
 # Trace_end runs at the end and will be used to aggregate
 # the data into the final json object and print it out to stdout.
 def trace_end() -> None:
-	pass
+	# Schema: https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L305
+	gecko_profile_with_meta = {
+		"meta": {
+			"interval": 1,
+			"processType": 0,
+			"product": PRODUCT,
+			"stackwalk": 1,
+			"debug": 0,
+			"gcpoison": 0,
+			"asyncstack": 1,
+			"startTime": start_time,
+			"shutdownTime": None,
+			"version": 24,
+			"presymbolicated": True,
+			"categories": CATEGORIES,
+			"markerSchema": [],
+			},
+		"libs": [],
+		"threads": threads,
+		"processes": [],
+		"pausedRanges": [],
+	}
+	json.dump(gecko_profile_with_meta, sys.stdout, indent=2)
+
+def main() -> None:
+	global CATEGORIES
+	parser = argparse.ArgumentParser(description="Convert perf.data to Firefox\'s Gecko Profile format")
+
+	# Add the command-line options
+	# Colors must be defined according to this:
+	# https://github.com/firefox-devtools/profiler/blob/50124adbfa488adba6e2674a8f2618cf34b59cd2/res/css/categories.css
+	parser.add_argument('--user-color', default='yellow', help='Color for the User category')
+	parser.add_argument('--kernel-color', default='orange', help='Color for the Kernel category')
+	# Parse the command-line arguments
+	args = parser.parse_args()
+	# Access the values provided by the user
+	user_color = args.user_color
+	kernel_color = args.kernel_color
+
+	CATEGORIES = [
+		{
+			"name": 'User',
+			"color": user_color,
+			"subcategories": ['Other']
+		},
+		{
+			"name": 'Kernel',
+			"color": kernel_color,
+			"subcategories": ['Other']
+		},
+	]
+
+if __name__ == '__main__':
+    main()
Signature · 1 line
2.34.1
#5 lore raw

Re: [PATCH v4 4/6] perf scripts python: Add trace end processing and PRODUCT and CATEGORIES information

Em Wed, Jul 19, 2023 at 04:20:08AM +0530, Anup Sharma escreveu:

Quoted 9 lines

The final output will now be presented in JSON format following the Gecko profile structure. Additionally, the inclusion of PRODUCT allows easy retrieval of header information for UI.

Furthermore, CATEGORIES have been introduced to enable customization of kernel and user colors using input arguments. To facilitate this functionality, an argparse-based parser has been implemented.

Note that the implementation of threads will be addressed in subsequent commits."

It is failing here:

[root@five ~]# perf script firefox-gecko-converter.py
Traceback (most recent call last):
File "/var/home/acme/libexec/perf-core/scripts/python/firefox-gecko-converter.py", line 202, in trace_end
"threads": threads,
NameError: name 'threads' is not defined. Did you mean: 'Thread'?
Fatal Python error: handler_call_die: problem in Python trace event handler
Python runtime state: initialized

Current thread 0x00007fde43f98f40 (most recent call first):
<no Python frame>

Extension modules: perf_trace_context (total: 1) Aborted (core dumped)
[root@five ~]#

We need to be able to test it at each changeset, not just at the end of the patchkit, so please reorganize this so that running:

perf script firefox-gecko-converter.py

After each patch works.

We need this to bisect problems, etc.

- Arnaldo

Quoted 94 lines
---
 .../scripts/python/firefox-gecko-converter.py | 64 ++++++++++++++++++-
 1 file changed, 63 insertions(+), 1 deletion(-)
tools/perf/scripts/python/firefox-gecko-converter.py +63 -1
index d9b1ec18997a..a0218e2245f2 100644
@@ -11,6 +11,8 @@
 
 import os
 import sys
+import json
+import argparse
 from dataclasses import dataclass, field
 from typing import List, Dict, Optional, NamedTuple, Set, Tuple, Any
 
@@ -30,6 +32,13 @@ Milliseconds = float
 # start_time is intialiazed only once for the all event traces.
 start_time = None
 
+# https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/profile.js#L425
+# Follow Brendan Gregg's Flamegraph convention: orange for kernel and yellow for user space by default.
+CATEGORIES = None
+
+# The product name is used by the profiler UI to show the Operating system and Processor.
+PRODUCT = os.popen('uname -op').read().strip()
+
 # https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L156
 class Frame(NamedTuple):
 	string_id: StringID
@@ -172,4 +181,57 @@ def process_event(param_dict: Dict) -> None:
 # Trace_end runs at the end and will be used to aggregate
 # the data into the final json object and print it out to stdout.
 def trace_end() -> None:
-	pass
+	# Schema: https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L305
+	gecko_profile_with_meta = {
+		"meta": {
+			"interval": 1,
+			"processType": 0,
+			"product": PRODUCT,
+			"stackwalk": 1,
+			"debug": 0,
+			"gcpoison": 0,
+			"asyncstack": 1,
+			"startTime": start_time,
+			"shutdownTime": None,
+			"version": 24,
+			"presymbolicated": True,
+			"categories": CATEGORIES,
+			"markerSchema": [],
+			},
+		"libs": [],
+		"threads": threads,
+		"processes": [],
+		"pausedRanges": [],
+	}
+	json.dump(gecko_profile_with_meta, sys.stdout, indent=2)
+
+def main() -> None:
+	global CATEGORIES
+	parser = argparse.ArgumentParser(description="Convert perf.data to Firefox\'s Gecko Profile format")
+
+	# Add the command-line options
+	# Colors must be defined according to this:
+	# https://github.com/firefox-devtools/profiler/blob/50124adbfa488adba6e2674a8f2618cf34b59cd2/res/css/categories.css
+	parser.add_argument('--user-color', default='yellow', help='Color for the User category')
+	parser.add_argument('--kernel-color', default='orange', help='Color for the Kernel category')
+	# Parse the command-line arguments
+	args = parser.parse_args()
+	# Access the values provided by the user
+	user_color = args.user_color
+	kernel_color = args.kernel_color
+
+	CATEGORIES = [
+		{
+			"name": 'User',
+			"color": user_color,
+			"subcategories": ['Other']
+		},
+		{
+			"name": 'Kernel',
+			"color": kernel_color,
+			"subcategories": ['Other']
+		},
+	]
+
+if __name__ == '__main__':
+    main()
Signature · 1 line
2.34.1
Signature · 1 line
- Arnaldo
#6 lore raw

PATCH v4 5/6 perf scripts python: implement internal get or create frame, stack and string function

In reply to Anup Sharma

The intern_stack function is responsible for retrieving or creating a stack_id based on the provided frame_id and prefix_id. It first generates a key using the frame_id and prefix_id values. If the stack corresponding to the key is found in the stackMap, it is returned. Otherwise, a new stack is created by appending the prefix_id and frame_id to the stackTable. The key
and the index of the newly created stack are added to the stackMap for future reference.

The _intern_frame function is responsible for retrieving or creating a frame_id based on the provided frame string. If the frame_id corresponding to the frameString is found in the frameMap, it is returned. Otherwise, a new frame is created by appending relevant information to the frameTable and adding the frameString to the string_id through _intern_string.

The _intern_string function will gets a matching string, or saves the new string and returns a String ID.

---
 .../scripts/python/firefox-gecko-converter.py | 58 ++++++++++++++++++-
 1 file changed, 57 insertions(+), 1 deletion(-)
tools/perf/scripts/python/firefox-gecko-converter.py +57 -1
index a0218e2245f2..ae69c0a4af13 100644
@@ -13,6 +13,7 @@ import os
 import sys
 import json
 import argparse
+from functools import reduce
 from dataclasses import dataclass, field
 from typing import List, Dict, Optional, NamedTuple, Set, Tuple, Any
 
@@ -39,6 +40,13 @@ CATEGORIES = None
 # The product name is used by the profiler UI to show the Operating system and Processor.
 PRODUCT = os.popen('uname -op').read().strip()
 
+# Here key = tid, value = Thread
+tid_to_thread = dict()
+
+# The category index is used by the profiler UI to show the color of the flame graph.
+USER_CATEGORY_INDEX = 0
+KERNEL_CATEGORY_INDEX = 1
+
 # https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L156
 class Frame(NamedTuple):
 	string_id: StringID
@@ -99,6 +107,55 @@ class Thread:
 	stackMap: Dict[Tuple[Optional[int], int], int] = field(default_factory=dict)
 	frameMap: Dict[str, int] = field(default_factory=dict)
 
+	def _intern_stack(self, frame_id: int, prefix_id: Optional[int]) -> int:
+		"""Gets a matching stack, or saves the new stack. Returns a Stack ID."""
+		key = f"{frame_id}" if prefix_id is None else f"{frame_id},{prefix_id}"
+		# key = (prefix_id, frame_id)
+		stack_id = self.stackMap.get(key)
+		if stack_id is None:
+			# return stack_id
+			stack_id = len(self.stackTable)
+			self.stackTable.append(Stack(prefix_id=prefix_id, frame_id=frame_id))
+			self.stackMap[key] = stack_id
+		return stack_id
+
+	def _intern_string(self, string: str) -> int:
+		"""Gets a matching string, or saves the new string. Returns a String ID."""
+		string_id = self.stringMap.get(string)
+		if string_id is not None:
+			return string_id
+		string_id = len(self.stringTable)
+		self.stringTable.append(string)
+		self.stringMap[string] = string_id
+		return string_id
+
+	def _intern_frame(self, frame_str: str) -> int:
+		"""Gets a matching stack frame, or saves the new frame. Returns a Frame ID."""
+		frame_id = self.frameMap.get(frame_str)
+		if frame_id is not None:
+			return frame_id
+		frame_id = len(self.frameTable)
+		self.frameMap[frame_str] = frame_id
+		string_id = self._intern_string(frame_str)
+
+		symbol_name_to_category = KERNEL_CATEGORY_INDEX if frame_str.find('kallsyms') != -1 \
+		or frame_str.find('/vmlinux') != -1 \
+		or frame_str.endswith('.ko)') \
+		else USER_CATEGORY_INDEX
+
+		self.frameTable.append(Frame(
+			string_id=string_id,
+			relevantForJS=False,
+			innerWindowID=0,
+			implementation=None,
+			optimizations=None,
+			line=None,
+			column=None,
+			category=symbol_name_to_category,
+			subcategory=None,
+		))
+		return frame_id
+
 	def _to_json_dict(self) -> Dict:
 		"""Converts current Thread to GeckoThread JSON format."""
 		# Gecko profile format is row-oriented data as List[List],
@@ -163,7 +220,6 @@ class Thread:
 			"processType": "default",
 		}
 
-
 # Uses perf script python interface to parse each
 # event and store the data in the thread builder.
 def process_event(param_dict: Dict) -> None:
Signature · 1 line
2.34.1
#7 lore raw

PATCH v4 6/6 perf scripts python: Implement add sample function and thread processing

In reply to Anup Sharma

The stack has been created for storing func and dso from the callchain. The sample has been added to a specific thread. It first checks if the thread exists in the Thread class. Then it call _add_sample function which is responsible for appending a new entry to the samples list.

---
 .../scripts/python/firefox-gecko-converter.py | 46 +++++++++++++++++++
 1 file changed, 46 insertions(+)
tools/perf/scripts/python/firefox-gecko-converter.py +46 -0
index ae69c0a4af13..9c4393787daa 100644
@@ -156,6 +156,25 @@ class Thread:
 		))
 		return frame_id
 
+	def _add_sample(self, comm: str, stack: List[str], time_ms: Milliseconds) -> None:
+		"""Add a timestamped stack trace sample to the thread builder.
+		Args:
+			comm: command-line (name) of the thread at this sample
+			stack: sampled stack frames. Root first, leaf last.
+			time_ms: timestamp of sample in milliseconds.
+		"""
+		# Ihreads may not set their names right after they are created.
+		# Instead, they might do it later. In such situations, to use the latest name they have set.
+		if self.comm != comm:
+			self.comm = comm
+
+		prefix_stack_id = reduce(lambda prefix_id, frame: self._intern_stack
+						(self._intern_frame(frame), prefix_id), stack, None)
+		if prefix_stack_id is not None:
+			self.samples.append(Sample(stack_id=prefix_stack_id,
+									time_ms=time_ms,
+									responsiveness=0))
+
 	def _to_json_dict(self) -> Dict:
 		"""Converts current Thread to GeckoThread JSON format."""
 		# Gecko profile format is row-oriented data as List[List],
@@ -234,9 +253,36 @@ def process_event(param_dict: Dict) -> None:
 	if not start_time:
 		start_time = time_stamp
 
+	# Parse and append the callchain of the current sample into a stack.
+	stack = []
+	if param_dict['callchain']:
+		for call in param_dict['callchain']:
+			if 'sym' not in call:
+				continue
+			stack.append(f'{call["sym"]["name"]} (in {call["dso"]})')
+		if len(stack) != 0:
+			# Reverse the stack, as root come first and the leaf at the end.
+			stack = stack[::-1]
+
+	# During perf record if -g is not used, the callchain is not available.
+	# In that case, the symbol and dso are available in the event parameters.
+	else:
+		func = param_dict['symbol'] if 'symbol' in param_dict else '[unknown]'
+		dso = param_dict['dso'] if 'dso' in param_dict else '[unknown]'
+		stack.append(f'{func} (in {dso})')
+
+	# Add sample to the specific thread.
+	thread = tid_to_thread.get(tid)
+	if thread is None:
+		thread = Thread(comm=comm, pid=pid, tid=tid)
+		tid_to_thread[tid] = thread
+	thread._add_sample(comm=comm, stack=stack, time_ms=time_stamp)
+
 # Trace_end runs at the end and will be used to aggregate
 # the data into the final json object and print it out to stdout.
 def trace_end() -> None:
+	threads = [thread._to_json_dict() for thread in tid_to_thread.values()]
+
 	# Schema: https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L305
 	gecko_profile_with_meta = {
 		"meta": {
Signature · 1 line
2.34.1
#8 lore raw

Re: [PATCH v4 6/6] perf scripts python: Implement add sample function and thread processing

Em Wed, Jul 19, 2023 at 04:22:13AM +0530, Anup Sharma escreveu:

Quoted 4 lines

The stack has been created for storing func and dso from the callchain. The sample has been added to a specific thread. It first checks if the thread exists in the Thread class. Then it call _add_sample function which is responsible for appending a new entry to the samples list.

At the end of the series it produces JSON output, I was expecting that at this point you would provide a way for us to use this output and see it rendered somewhere, can you add instructions for that in this cset comment?

Also it isn't working when I try to see just the start of the output:

[root@five ~]# perf script firefox-gecko-converter.py | head {

  "meta": {
    "interval": 1,
    "processType": 0,
    "product": "x86_64 GNU/Linux",
    "stackwalk": 1,
    "debug": 0,
    "gcpoison": 0,
    "asyncstack": 1,
    "startTime": 6796435.569,

Traceback (most recent call last):
File "/var/home/acme/libexec/perf-core/scripts/python/firefox-gecko-converter.py", line 308, in trace_end
json.dump(gecko_profile_with_meta, sys.stdout, indent=2)
File "/usr/lib64/python3.10/json/__init__.py", line 180, in dump
fp.write(chunk)
BrokenPipeError: [Errno 32] Broken pipe
Fatal Python error: handler_call_die: problem in Python trace event handler
Python runtime state: initialized

Current thread 0x00007fc21c0ebf40 (most recent call first):
<no Python frame>

Extension modules: perf_trace_context (total: 1)
[root@five ~]#

Quoted 75 lines
---
 .../scripts/python/firefox-gecko-converter.py | 46 +++++++++++++++++++
 1 file changed, 46 insertions(+)
tools/perf/scripts/python/firefox-gecko-converter.py +46 -0
index ae69c0a4af13..9c4393787daa 100644
@@ -156,6 +156,25 @@ class Thread:
 		))
 		return frame_id
 
+	def _add_sample(self, comm: str, stack: List[str], time_ms: Milliseconds) -> None:
+		"""Add a timestamped stack trace sample to the thread builder.
+		Args:
+			comm: command-line (name) of the thread at this sample
+			stack: sampled stack frames. Root first, leaf last.
+			time_ms: timestamp of sample in milliseconds.
+		"""
+		# Ihreads may not set their names right after they are created.
+		# Instead, they might do it later. In such situations, to use the latest name they have set.
+		if self.comm != comm:
+			self.comm = comm
+
+		prefix_stack_id = reduce(lambda prefix_id, frame: self._intern_stack
+						(self._intern_frame(frame), prefix_id), stack, None)
+		if prefix_stack_id is not None:
+			self.samples.append(Sample(stack_id=prefix_stack_id,
+									time_ms=time_ms,
+									responsiveness=0))
+
 	def _to_json_dict(self) -> Dict:
 		"""Converts current Thread to GeckoThread JSON format."""
 		# Gecko profile format is row-oriented data as List[List],
@@ -234,9 +253,36 @@ def process_event(param_dict: Dict) -> None:
 	if not start_time:
 		start_time = time_stamp
 
+	# Parse and append the callchain of the current sample into a stack.
+	stack = []
+	if param_dict['callchain']:
+		for call in param_dict['callchain']:
+			if 'sym' not in call:
+				continue
+			stack.append(f'{call["sym"]["name"]} (in {call["dso"]})')
+		if len(stack) != 0:
+			# Reverse the stack, as root come first and the leaf at the end.
+			stack = stack[::-1]
+
+	# During perf record if -g is not used, the callchain is not available.
+	# In that case, the symbol and dso are available in the event parameters.
+	else:
+		func = param_dict['symbol'] if 'symbol' in param_dict else '[unknown]'
+		dso = param_dict['dso'] if 'dso' in param_dict else '[unknown]'
+		stack.append(f'{func} (in {dso})')
+
+	# Add sample to the specific thread.
+	thread = tid_to_thread.get(tid)
+	if thread is None:
+		thread = Thread(comm=comm, pid=pid, tid=tid)
+		tid_to_thread[tid] = thread
+	thread._add_sample(comm=comm, stack=stack, time_ms=time_stamp)
+
 # Trace_end runs at the end and will be used to aggregate
 # the data into the final json object and print it out to stdout.
 def trace_end() -> None:
+	threads = [thread._to_json_dict() for thread in tid_to_thread.values()]
+
 	# Schema: https://github.com/firefox-devtools/profiler/blob/53970305b51b9b472e26d7457fee1d66cd4e2737/src/types/gecko-profile.js#L305
 	gecko_profile_with_meta = {
 		"meta": {
Signature · 1 line
2.34.1
Signature · 1 line
- Arnaldo
#9 lore raw
In reply to Anup Sharma
Quoted 7 lines · Anup Sharma

This patch series adds support for Firefox's gecko profile format. The format is documented here [1].

I have incorporated several changes based on feedback from the previous version of the patch.

Changes in v3:

a small typo here. It should be v4 instead of v3 (changes in v4).

Quoted 48 lines

- Implemented object-oriented programming (OOP) approach for Thread and Sample to improve code organization and maintainability.
- Enhanced user experience by introducing argparse for changing color settings of user and kernel symbols during execution.
- Ensured proper module encapsulation by adding imports wherever necessary.
- Improved code readability by adding descriptive comments and type annotations.

These changes focus on adopting OOP principles, enhancing user interaction with argparse, and making the code more readable with comments and type information.

TODO:
- use cpu_mode to determine the category instead of finding it from symbol name.
- Write a test.
- add direct execution command for this script under script/python/bin directory.

Committer Testing:
- Tested with a perf.data file generated for single and multiple cpu cores enabled.
- Uploaded the stdout into profiler.firefox.com and verified the output.
- Verified the output with the output generated by the existing script as mentioned here [2].
Method:
- perf record -F 99 -a -g -- sleep 5
- perf script firefox-gecko-converter.py > output.json
- upload output.json to profiler.firefox.com

[1] https://github.com/firefox-devtools/profiler/blob/main/docs-developer/gecko-profile-format.md
[2] https://perf.wiki.kernel.org/index.php/Tutorial#Firefox_Profiler

Anup Sharma (6):
perf scripts python: Add initial script file with usage information perf scripts python: Extact necessary information from process event perf scripts python: Add classes and conversion functions perf scripts python: Add trace end processing and PRODUCT and
CATEGORIES information
perf scripts python: implement internal get or create frame, stack and
string function
perf scripts python: Implement add sample function and thread
processing

 .../scripts/python/firefox-gecko-converter.py | 339 ++++++++++++++++++
 1 file changed, 339 insertions(+)
 create mode 100644 tools/perf/scripts/python/firefox-gecko-converter.py
Signature · 1 line
2.34.1
reviewer Acked-by
#10 lore raw
Quoted 25 lines · Anup Sharma
Quoted 7 lines · Anup Sharma

This patch series adds support for Firefox's gecko profile format. The format is documented here [1].

I have incorporated several changes based on feedback from the previous version of the patch.

Changes in v3:

a small typo here. It should be v4 instead of v3 (changes in v4).

Quoted 15 lines

- Implemented object-oriented programming (OOP) approach for Thread and Sample to improve code organization and maintainability.
- Enhanced user experience by introducing argparse for changing color settings of user and kernel symbols during execution.
- Ensured proper module encapsulation by adding imports wherever necessary.
- Improved code readability by adding descriptive comments and type annotations.

These changes focus on adopting OOP principles, enhancing user interaction with argparse, and making the code more readable with comments and type information.

TODO:
- use cpu_mode to determine the category instead of finding it from symbol name.
- Write a test.
- add direct execution command for this script under script/python/bin directory.

Thanks Anup, is there a minimal TODO so that some code can be merged? I didn't have any comments for this series so:

I think even the test can be follow up as the code isn't on the critical path.

Thanks,
Ian

Quoted 33 lines
Quoted 33 lines

Committer Testing:
- Tested with a perf.data file generated for single and multiple cpu cores enabled.
- Uploaded the stdout into profiler.firefox.com and verified the output.
- Verified the output with the output generated by the existing script as mentioned here [2].
Method:
- perf record -F 99 -a -g -- sleep 5
- perf script firefox-gecko-converter.py > output.json
- upload output.json to profiler.firefox.com

[1] https://github.com/firefox-devtools/profiler/blob/main/docs-developer/gecko-profile-format.md
[2] https://perf.wiki.kernel.org/index.php/Tutorial#Firefox_Profiler

Anup Sharma (6):
perf scripts python: Add initial script file with usage information perf scripts python: Extact necessary information from process event perf scripts python: Add classes and conversion functions perf scripts python: Add trace end processing and PRODUCT and
CATEGORIES information
perf scripts python: implement internal get or create frame, stack and
string function
perf scripts python: Implement add sample function and thread
processing

 .../scripts/python/firefox-gecko-converter.py | 339 ++++++++++++++++++
 1 file changed, 339 insertions(+)
 create mode 100644 tools/perf/scripts/python/firefox-gecko-converter.py

--
2.34.1

#11 lore raw
Quoted 21 lines · Ian Rogers
Quoted 20 lines · Anup Sharma
Quoted 7 lines · Anup Sharma

This patch series adds support for Firefox's gecko profile format. The format is documented here [1].

I have incorporated several changes based on feedback from the previous version of the patch.

Changes in v3:

a small typo here. It should be v4 instead of v3 (changes in v4).

Quoted 10 lines

- Implemented object-oriented programming (OOP) approach for Thread and Sample to improve code organization and maintainability.
- Enhanced user experience by introducing argparse for changing color settings of user and kernel symbols during execution.
- Ensured proper module encapsulation by adding imports wherever necessary.
- Improved code readability by adding descriptive comments and type annotations.

These changes focus on adopting OOP principles, enhancing user interaction with argparse, and making the code more readable with comments and type information.

Quoted 7 lines
Quoted 5 lines
Quoted 5 lines

TODO:
- use cpu_mode to determine the category instead of finding it from symbol name.
- Write a test.
- add direct execution command for this script under script/python/bin directory.

Thanks Anup, is there a minimal TODO so that some code can be merged?

Yaah, only the first TODO is left, will discuss during office hours.

Quoted 2 lines

I didn't have any comments for this series so:

Thanks for the review.

Quoted 1 line

I think even the test can be follow up as the code isn't on the critical path.

By the way, I've finished writing the test and will send it very soon.

Quoted 36 lines

Thanks,
Ian

Quoted 33 lines
Quoted 33 lines

Committer Testing:
- Tested with a perf.data file generated for single and multiple cpu cores enabled.
- Uploaded the stdout into profiler.firefox.com and verified the output.
- Verified the output with the output generated by the existing script as mentioned here [2].
Method:
- perf record -F 99 -a -g -- sleep 5
- perf script firefox-gecko-converter.py > output.json
- upload output.json to profiler.firefox.com

[1] https://github.com/firefox-devtools/profiler/blob/main/docs-developer/gecko-profile-format.md
[2] https://perf.wiki.kernel.org/index.php/Tutorial#Firefox_Profiler

Anup Sharma (6):
perf scripts python: Add initial script file with usage information perf scripts python: Extact necessary information from process event perf scripts python: Add classes and conversion functions perf scripts python: Add trace end processing and PRODUCT and
CATEGORIES information
perf scripts python: implement internal get or create frame, stack and
string function
perf scripts python: Implement add sample function and thread
processing

 .../scripts/python/firefox-gecko-converter.py | 339 ++++++++++++++++++
 1 file changed, 339 insertions(+)
 create mode 100644 tools/perf/scripts/python/firefox-gecko-converter.py

--
2.34.1

#12 lore raw
In reply to Anup Sharma

Em Wed, Jul 19, 2023 at 04:34:28PM +0530, Anup Sharma escreveu:

Quoted 35 lines · Anup Sharma
Quoted 7 lines

This patch series adds support for Firefox's gecko profile format. The format is documented here [1].

I have incorporated several changes based on feedback from the previous version of the patch.

Changes in v3:

a small typo here. It should be v4 instead of v3 (changes in v4).

Quoted 26 lines

- Implemented object-oriented programming (OOP) approach for Thread and Sample to improve code organization and maintainability.
- Enhanced user experience by introducing argparse for changing color settings of user and kernel symbols during execution.
- Ensured proper module encapsulation by adding imports wherever necessary.
- Improved code readability by adding descriptive comments and type annotations.

These changes focus on adopting OOP principles, enhancing user interaction with argparse, and making the code more readable with comments and type information.

TODO:
- use cpu_mode to determine the category instead of finding it from symbol name.
- Write a test.
- add direct execution command for this script under script/python/bin directory.

Committer Testing:
- Tested with a perf.data file generated for single and multiple cpu cores enabled.
- Uploaded the stdout into profiler.firefox.com and verified the output.
- Verified the output with the output generated by the existing script as mentioned here [2].
Method:
- perf record -F 99 -a -g -- sleep 5
- perf script firefox-gecko-converter.py > output.json
- upload output.json to profiler.firefox.com

I see it in the cover letter, will try now.

Quoted 21 lines
Quoted 21 lines

[1] https://github.com/firefox-devtools/profiler/blob/main/docs-developer/gecko-profile-format.md
[2] https://perf.wiki.kernel.org/index.php/Tutorial#Firefox_Profiler

Anup Sharma (6):
perf scripts python: Add initial script file with usage information perf scripts python: Extact necessary information from process event perf scripts python: Add classes and conversion functions perf scripts python: Add trace end processing and PRODUCT and
CATEGORIES information
perf scripts python: implement internal get or create frame, stack and
string function
perf scripts python: Implement add sample function and thread
processing

 .../scripts/python/firefox-gecko-converter.py | 339 ++++++++++++++++++
 1 file changed, 339 insertions(+)
 create mode 100644 tools/perf/scripts/python/firefox-gecko-converter.py
Signature · 1 line
2.34.1
Signature · 1 line
- Arnaldo
#13 lore raw

Em Thu, Jul 20, 2023 at 12:14:10PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 37 lines

Em Wed, Jul 19, 2023 at 04:34:28PM +0530, Anup Sharma escreveu:

Quoted 35 lines · Anup Sharma
Quoted 7 lines

This patch series adds support for Firefox's gecko profile format. The format is documented here [1].

I have incorporated several changes based on feedback from the previous version of the patch.

Changes in v3:

a small typo here. It should be v4 instead of v3 (changes in v4).

Quoted 26 lines

- Implemented object-oriented programming (OOP) approach for Thread and Sample to improve code organization and maintainability.
- Enhanced user experience by introducing argparse for changing color settings of user and kernel symbols during execution.
- Ensured proper module encapsulation by adding imports wherever necessary.
- Improved code readability by adding descriptive comments and type annotations.

These changes focus on adopting OOP principles, enhancing user interaction with argparse, and making the code more readable with comments and type information.

TODO:
- use cpu_mode to determine the category instead of finding it from symbol name.
- Write a test.
- add direct execution command for this script under script/python/bin directory.

Committer Testing:
- Tested with a perf.data file generated for single and multiple cpu cores enabled.
- Uploaded the stdout into profiler.firefox.com and verified the output.
- Verified the output with the output generated by the existing script as mentioned here [2].
Method:
- perf record -F 99 -a -g -- sleep 5
- perf script firefox-gecko-converter.py > output.json
- upload output.json to profiler.firefox.com

Quoted 1 line

I see it in the cover letter, will try now.

I followed the instructions, uploaded a output.json produced and it seems to work, good job!

- Arnaldo