Codecov Report

Merging #138 (c630217) into master (db980b5) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #138   +/-   ##
=======================================
  Coverage   27.47%   27.47%           
=======================================
  Files          20       20           
  Lines         404      404           
=======================================
  Hits          111      111           
  Misses        286      286           
  Partials        7        7           

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update db980b5...c630217. Read the comment docs.

Could you try a CLI flame map visualization? I think it is a better way to verify CLI command and ebpf profiling result. No matter for codes, PR and demonstration in the future.

I could not find a great visualization to show the profiling analysis. But I could save the analysis as an svg file, then user could open this file on the browser and show the performance issue.
Here is a demo file: https://www.brendangregg.com/FlameGraphs/cpu-bash-flamegraph.svg

And If we are using the CLI with E2E, the yaml format is more perfect to use. the graph as a dump file is for the user to use.

But I could save the analysis as an svg file,

SVG is good enough.

And If we are using the CLI with E2E, the yaml format is more perfect to use. the graph as a dump file is for the user to use.

Agree.

I think the html should only output when set --dispaly=html/graph?

I think the html should only output when set --dispaly=html/graph?

Yes, the analysis default is displayed to graph, and it would output a file. Should the user need to force setting it into the command-line?

Yes, the analysis default is displayed to graph, and it would output a file. Should the user need to force setting it into the command-line?

What was the default output for profiling trace? I think we should keep consistent.

Yes, the analysis default is displayed to graph, and it would output a file. Should the user need to force setting it into the command-line?

What was the default output for profiling trace? I think we should keep consistent.

In the default, profiling trace analysis and profiling ebpf analysis output is the graph that generates the perf HTML to the current directory, filename format: flame_graph_{current_timestamp}.html.

How profiling trace analysis gets a graph out? I don't remember we could do that.

How profiling trace analysis gets a graph out? I don't remember we could do that.

trace profiling data have the dumpCount attribute could help on getting the count of per stack. then it is also dependent on the tree mode(id and parentId to build). So we could use them to build a perf HTML that can see which methods need more time to execute.

Could you share a screenshot here? I think we didn't have this feature before. This is new too. Please mention this, breaking change, ebpf profiling in change logs separatedly.

Could you share a screenshot here? I think we didn't have this feature before. This is new too. Please mention this, breaking change, ebpf profiling in change logs separatedly.

Updated. Because the demo of trace profiling it's just using Thread#sleep. So it could only get one stack.

Make sense to me.

Wait for @kezhenxu94 to review.