mirror of
https://github.com/python/cpython.git
synced 2025-10-26 19:24:34 +00:00
gh-135953: Reduce memory usage of stack collectors (#138875)
The stack collector base class keeps all frames until export() is called, which causes significant unnecessary memory usage. Instead, we can process the frames on the fly in the collect call by dispatching the aggregation logic to the subclass through the process_frames method. Co-authored-by: Pablo Galindo Salgado <pablogsal@gmail.com>
This commit is contained in:
László Kiss Kollár
GitHub
parent
efc08c5fbf
commit
3e06cfcaee
5 changed files with 292 additions and 174 deletions
|
|
@ -7,51 +7,51 @@
|
|||
import os
|
||||
|
||||
from .collector import Collector
|
||||
from .string_table import StringTable
|
||||
|
||||
|
||||
class StackTraceCollector(Collector):
|
||||
def __init__(self):
|
||||
self.call_trees = []
|
||||
self.function_samples = collections.defaultdict(int)
|
||||
|
||||
def _process_frames(self, frames):
|
||||
"""Process a single thread's frame stack."""
|
||||
if not frames:
|
||||
return
|
||||
|
||||
# Store the complete call stack (reverse order - root first)
|
||||
call_tree = list(reversed(frames))
|
||||
self.call_trees.append(call_tree)
|
||||
|
||||
# Count samples per function
|
||||
for frame in frames:
|
||||
self.function_samples[frame] += 1
|
||||
|
||||
def collect(self, stack_frames):
|
||||
for frames in self._iter_all_frames(stack_frames):
|
||||
self._process_frames(frames)
|
||||
if not frames:
|
||||
continue
|
||||
self.process_frames(frames)
|
||||
|
||||
def process_frames(self, frames):
|
||||
pass
|
||||
|
||||
|
||||
class CollapsedStackCollector(StackTraceCollector):
|
||||
def __init__(self):
|
||||
self.stack_counter = collections.Counter()
|
||||
|
||||
def process_frames(self, frames):
|
||||
call_tree = tuple(reversed(frames))
|
||||
self.stack_counter[call_tree] += 1
|
||||
|
||||
def export(self, filename):
|
||||
stack_counter = collections.Counter()
|
||||
for call_tree in self.call_trees:
|
||||
# Call tree is already in root->leaf order
|
||||
lines = []
|
||||
for call_tree, count in self.stack_counter.items():
|
||||
stack_str = ";".join(
|
||||
f"{os.path.basename(f[0])}:{f[2]}:{f[1]}" for f in call_tree
|
||||
)
|
||||
stack_counter[stack_str] += 1
|
||||
lines.append((stack_str, count))
|
||||
|
||||
lines.sort(key=lambda x: (-x[1], x[0]))
|
||||
|
||||
with open(filename, "w") as f:
|
||||
for stack, count in stack_counter.items():
|
||||
for stack, count in lines:
|
||||
f.write(f"{stack} {count}\n")
|
||||
print(f"Collapsed stack output written to {filename}")
|
||||
|
||||
|
||||
class FlamegraphCollector(StackTraceCollector):
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
self.stats = {}
|
||||
self._root = {"samples": 0, "children": {}}
|
||||
self._total_samples = 0
|
||||
self._func_intern = {}
|
||||
self._string_table = StringTable()
|
||||
|
||||
def set_stats(self, sample_interval_usec, duration_sec, sample_rate, error_rate=None):
|
||||
"""Set profiling statistics to include in flamegraph data."""
|
||||
|
|
@ -65,11 +65,13 @@ def set_stats(self, sample_interval_usec, duration_sec, sample_rate, error_rate=
|
|||
def export(self, filename):
|
||||
flamegraph_data = self._convert_to_flamegraph_format()
|
||||
|
||||
# Debug output
|
||||
# Debug output with string table statistics
|
||||
num_functions = len(flamegraph_data.get("children", []))
|
||||
total_time = flamegraph_data.get("value", 0)
|
||||
string_count = len(self._string_table)
|
||||
print(
|
||||
f"Flamegraph data: {num_functions} root functions, total samples: {total_time}"
|
||||
f"Flamegraph data: {num_functions} root functions, total samples: {total_time}, "
|
||||
f"{string_count} unique strings"
|
||||
)
|
||||
|
||||
if num_functions == 0:
|
||||
|
|
@ -98,105 +100,105 @@ def _format_function_name(func):
|
|||
return f"{funcname} ({filename}:{lineno})"
|
||||
|
||||
def _convert_to_flamegraph_format(self):
|
||||
"""Convert call trees to d3-flamegraph format with optimized hierarchy building"""
|
||||
if not self.call_trees:
|
||||
return {"name": "No Data", "value": 0, "children": []}
|
||||
|
||||
unique_functions = set()
|
||||
for call_tree in self.call_trees:
|
||||
unique_functions.update(call_tree)
|
||||
|
||||
func_to_name = {
|
||||
func: self._format_function_name(func) for func in unique_functions
|
||||
}
|
||||
|
||||
root = {"name": "root", "children": {}, "samples": 0}
|
||||
|
||||
for call_tree in self.call_trees:
|
||||
current_node = root
|
||||
current_node["samples"] += 1
|
||||
|
||||
for func in call_tree:
|
||||
func_name = func_to_name[func] # Use pre-computed name
|
||||
|
||||
if func_name not in current_node["children"]:
|
||||
current_node["children"][func_name] = {
|
||||
"name": func_name,
|
||||
"func": func,
|
||||
"children": {},
|
||||
"samples": 0,
|
||||
"filename": func[0],
|
||||
"lineno": func[1],
|
||||
"funcname": func[2],
|
||||
}
|
||||
|
||||
current_node = current_node["children"][func_name]
|
||||
current_node["samples"] += 1
|
||||
|
||||
def convert_node(node, min_samples=1):
|
||||
if node["samples"] < min_samples:
|
||||
return None
|
||||
|
||||
source_code = None
|
||||
if "func" in node:
|
||||
source_code = self._get_source_lines(node["func"])
|
||||
|
||||
result = {
|
||||
"name": node["name"],
|
||||
"value": node["samples"],
|
||||
"""Convert aggregated trie to d3-flamegraph format with string table optimization."""
|
||||
if self._total_samples == 0:
|
||||
return {
|
||||
"name": self._string_table.intern("No Data"),
|
||||
"value": 0,
|
||||
"children": [],
|
||||
"strings": self._string_table.get_strings()
|
||||
}
|
||||
|
||||
if "filename" in node:
|
||||
result.update(
|
||||
{
|
||||
"filename": node["filename"],
|
||||
"lineno": node["lineno"],
|
||||
"funcname": node["funcname"],
|
||||
}
|
||||
def convert_children(children, min_samples):
|
||||
out = []
|
||||
for func, node in children.items():
|
||||
samples = node["samples"]
|
||||
if samples < min_samples:
|
||||
continue
|
||||
|
||||
# Intern all string components for maximum efficiency
|
||||
filename_idx = self._string_table.intern(func[0])
|
||||
funcname_idx = self._string_table.intern(func[2])
|
||||
name_idx = self._string_table.intern(self._format_function_name(func))
|
||||
|
||||
child_entry = {
|
||||
"name": name_idx,
|
||||
"value": samples,
|
||||
"children": [],
|
||||
"filename": filename_idx,
|
||||
"lineno": func[1],
|
||||
"funcname": funcname_idx,
|
||||
}
|
||||
|
||||
source = self._get_source_lines(func)
|
||||
if source:
|
||||
# Intern source lines for memory efficiency
|
||||
source_indices = [self._string_table.intern(line) for line in source]
|
||||
child_entry["source"] = source_indices
|
||||
|
||||
# Recurse
|
||||
child_entry["children"] = convert_children(
|
||||
node["children"], min_samples
|
||||
)
|
||||
out.append(child_entry)
|
||||
|
||||
if source_code:
|
||||
result["source"] = source_code
|
||||
|
||||
# Recursively convert children
|
||||
child_nodes = []
|
||||
for child_name, child_node in node["children"].items():
|
||||
child_result = convert_node(child_node, min_samples)
|
||||
if child_result:
|
||||
child_nodes.append(child_result)
|
||||
|
||||
# Sort children by sample count (descending)
|
||||
child_nodes.sort(key=lambda x: x["value"], reverse=True)
|
||||
result["children"] = child_nodes
|
||||
|
||||
return result
|
||||
# Sort by value (descending) then by name index for consistent ordering
|
||||
out.sort(key=lambda x: (-x["value"], x["name"]))
|
||||
return out
|
||||
|
||||
# Filter out very small functions (less than 0.1% of total samples)
|
||||
total_samples = len(self.call_trees)
|
||||
total_samples = self._total_samples
|
||||
min_samples = max(1, int(total_samples * 0.001))
|
||||
|
||||
converted_root = convert_node(root, min_samples)
|
||||
|
||||
if not converted_root or not converted_root["children"]:
|
||||
return {"name": "No significant data", "value": 0, "children": []}
|
||||
root_children = convert_children(self._root["children"], min_samples)
|
||||
if not root_children:
|
||||
return {
|
||||
"name": self._string_table.intern("No significant data"),
|
||||
"value": 0,
|
||||
"children": [],
|
||||
"strings": self._string_table.get_strings()
|
||||
}
|
||||
|
||||
# If we only have one root child, make it the root to avoid redundant level
|
||||
if len(converted_root["children"]) == 1:
|
||||
main_child = converted_root["children"][0]
|
||||
main_child["name"] = f"Program Root: {main_child['name']}"
|
||||
if len(root_children) == 1:
|
||||
main_child = root_children[0]
|
||||
# Update the name to indicate it's the program root
|
||||
old_name = self._string_table.get_string(main_child["name"])
|
||||
new_name = f"Program Root: {old_name}"
|
||||
main_child["name"] = self._string_table.intern(new_name)
|
||||
main_child["stats"] = self.stats
|
||||
main_child["strings"] = self._string_table.get_strings()
|
||||
return main_child
|
||||
|
||||
converted_root["name"] = "Program Root"
|
||||
converted_root["stats"] = self.stats
|
||||
return converted_root
|
||||
return {
|
||||
"name": self._string_table.intern("Program Root"),
|
||||
"value": total_samples,
|
||||
"children": root_children,
|
||||
"stats": self.stats,
|
||||
"strings": self._string_table.get_strings()
|
||||
}
|
||||
|
||||
def process_frames(self, frames):
|
||||
# Reverse to root->leaf
|
||||
call_tree = reversed(frames)
|
||||
self._root["samples"] += 1
|
||||
self._total_samples += 1
|
||||
|
||||
current = self._root
|
||||
for func in call_tree:
|
||||
func = self._func_intern.setdefault(func, func)
|
||||
children = current["children"]
|
||||
node = children.get(func)
|
||||
if node is None:
|
||||
node = {"samples": 0, "children": {}}
|
||||
children[func] = node
|
||||
node["samples"] += 1
|
||||
current = node
|
||||
|
||||
def _get_source_lines(self, func):
|
||||
filename, lineno, funcname = func
|
||||
filename, lineno, _ = func
|
||||
|
||||
try:
|
||||
# Get several lines around the function definition
|
||||
lines = []
|
||||
start_line = max(1, lineno - 2)
|
||||
end_line = lineno + 3
|
||||
|
|
@ -210,7 +212,6 @@ def _get_source_lines(self, func):
|
|||
return lines if lines else None
|
||||
|
||||
except Exception:
|
||||
# If we can't get source code, return None
|
||||
return None
|
||||
|
||||
def _create_flamegraph_html(self, data):
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue