microsoft
Adding a UserCommand to convert etl files to speedscope.json or updating NetperfToSpeedScope
I've noticed that there is a UserCommand called NetperfToSpeedScope however giving that method a provided etl file, as it is stated in its comments results in an error due to the etlx file not being found. I am unsure if there is a way to give the UserCommand a .netperf file since as far as I can tell from documentation the .netperf format is deprecated but this would likely cause the program to also look for a .etlx file which may not exist since I cant find where these are generated.
My main question is, is it worth the time to put in a PR to add a UserCommand to convert an etl file to SpeedScope, or does the current method work but need some more specific files that I am not looking for?
If NetPerf is not deprecated, how would I obtain this data using PerfView so I can quickly convert traces to the SpeedScope format?
@EmosewaMC, I recommend that you look at https://docs.microsoft.com/en-us/dotnet/core/diagnostics/dotnet-trace. It is possible to convert from nettrace to speedscope via dotnet-trace convert. You can also specify the output format directly if using dotnet-trace to collect.
The main question which I kind of buried in here is how can I convert an etl file to a speedscope format? This would work for nettrace but my main goal is to take a trace with PerfView, and convert that resulting etl file to a speedscope without needing to do so via the CPUStacks gui. Is this possible in the current version of PerfView?
Oh, I see. Sorry, I mis-read your question. I am not aware of a way to do this via the command line. It is available through the Save View As menu option in the Stack Viewer. Totally reasonable to create a UserCommand for this if you like.
Although not PerfView I have needed this functionality as well and have added it to ETWAnalyzer. I have copied the PerfView code which is very well designed. That was easy.
https://github.com/Siemens-Healthineers/ETWAnalyzer
It gives you also the option to sum all threads into one to get an overview of the total CPU consumption which I use sometimes to see what is costing too much CPU in total.
