crossvm/tools/contrib/cros_tracing_analyser

Analyzes cros_tracing event

Extract event_data and timestamp from input file from trace-cmd record and calculate average latency of cros_tracing events

How to list cros_tracing event name

$ cargo run -- list --input trace.dat --count 10

Print list of function names and sum of latency in the trace.dat. Example log:

#1: read: 728685132 usec
#2: readdir: 719231760 usec
#3: lookup: 460496754 usec
#4: open: 38860424 usec
#5: opendir: 38159576 usec
#6: getxattr: 21408816 usec
#7: release: 17821045 usec
#8: releasedir: 17783896 usec
#9: forget: 2942940 usec
#10: getattr: 301824 usec

How to generate histogram

$ cargo run -- histogram --input trace.dat --output histogram.json

To run the python script that generates the histogram plots, you need to install the matplotlib python library.

$ sudo apt-get install python3-matplotlib

To visualize the histogram,

$ python3 histogram.py ./src/histogram.json

And then histograms for each cros_tracing event will be displayed.

How to calculate event average latency

$ cargo run -- average --input trace.dat

calculate the average latency for each virtiofs event and print. Example log:

#0: readdir: 307364 usec
#1: read: 303366 usec
#2: lookup: 71762 usec
#3: open: 34148 usec
#4: opendir: 34132 usec
#5: statfs: 27116 usec
#6: forget: 26754 usec
#7: getxattr: 18714 usec
#8: getattr: 16768 usec
#9: release: 15983 usec
#10: releasedir: 15964 usec
#11: readlink: 15480 usec
#12: flush: 11939 usec

How to generate flamegraph data

$ cargo run -- flamegraph --input trace.dat --output-json tracing_data.json

Extract all events and calculate its latency and output it to a json file compatibile with d3 flamegraph.

To visualize the html page with the flamegraph, you need to run a local webserver. You can do this with a simple python http server:

$ python3 -m http.server

And then open the page at http://localhost:8000/flamegraph.html and the flamegraph will be displayed.

How to apply filters to the flamegraph

$ cargo run -- flamegraph --input trace.dat --output-json tracing_data.json --function "lookup" --count 20

For example this command outputs the data of the top 20 "lookup" functions that are taking the most time: