Skip to content

Instantly share code, notes, and snippets.

Last active January 3, 2024 22:50
Show Gist options
  • Star 1 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save softwaredoug/b0ddee0941c9dd5b3922a68f64622f32 to your computer and use it in GitHub Desktop.
Save softwaredoug/b0ddee0941c9dd5b3922a68f64622f32 to your computer and use it in GitHub Desktop.
VisualVM Forward Calls Stacktrace to Flamegraph format
"""Parses a VisualVM Forward Call Tree CSV file and outputs a flamegraph.
In VisualVM CPU Snapshot, select "Export", and choose "Forward Call Tree" and save
as CSV. The input will look something like:
"org.apache.spark.util.collection.ExternalSorter.writePartitionedFile(", "1,252 ms (100.0%)"
" Self time", "1 ms (0.01%)"
" org.apache.spark.util.collection.ExternalSorter.someOtherFunction(", "1,150 ms (95.0%)"
" Self time", "75 ms (5.0%)"
(Other calls from of someOtherFunction)
"org.apache.spark.util.collection.ExternalSorter.doAThing(", "1,150 ms (95.0%)"
The output is a flattened callstack with the time spent in each function consumable by the Brendan Gregg's script.
org.apache.spark.util.collection.ExternalSorter.writePartitionedFile( 1252
org.apache.spark.util.collection.ExternalSorter.writePartitionedFile(;org.apache.spark.util.collection.ExternalSorter.someOtherFunction( 1150
and so on
python <path to csv file> > stacks.txt
The output can be used with to generate a flamegraph as follows:
./ --title="My Flamegraph" stacks.txt > flamegraph.svg
Open the file in a browser to view the flamegraph.
import csv
import sys
def read_csv(filename):
"""Reads a CSV file and returns a list of lists."""
with open(filename, 'r') as f:
reader = csv.reader(f)
# Skip headers
return list(reader)
def flatten_callstack(table):
"""Flattens a callstack table."""
stacks = []
for idx, row in enumerate(table):
call = row[0]
my_time = float(row[1].replace(',', '').split()[0])
my_hits = int(row[3].replace(',', ''))
call_strip = call.strip()
if call_strip == 'Self time':
spaces_before = len(call) - len(call.lstrip())
my_stack = [call_strip]
# Read backwards to get calls indented less than my call
for prev_row in reversed(table[:idx]):
prev_call = prev_row[0]
prev_call_strip = prev_call.strip()
if prev_call_strip == 'Self time':
prev_spaces_before = len(prev_call) - len(prev_call.lstrip())
if prev_spaces_before < spaces_before:
# This is a parent call, prepend it, remember current indentation level
# so we don't get siblings
my_stack.insert(0, prev_call_strip)
spaces_before = prev_spaces_before
stacks.append((my_stack, my_time, my_hits))
return stacks
def to_flamegraph_format(stacks):
"""Converts a list of stacks to flamegraph format."""
for stack, _, hits in stacks:
print(';'.join(stack), hits)
def main(jmx_path):
data = read_csv(jmx_path)
flattened = flatten_callstack(data)
if __name__ == '__main__':
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment