It will be a huge improvement for performance analysis with FlameScope.
There is an implementation of FlameScope front, could you plan to provide data format support for FlameScope?
My understanding is that Flamescope consumes files under the 'perf' file format so I guess this means async-profiler would become able to output 'perf' files. I'm wondering what is the benefit compared to running perf directly? Honest question, I've just found out about Flamescope (thanks to this issue) and might be missing something :)
@miaozhuojun , do you want FlameScope for every kind? alloc/ cpu/ etc. I will take a look at it over the holidays.
Are there any plans to support this? Either the ‘perf’ or ‘nflxprofile’ formats for cpu alone (mem would be nice too) would be really helpful so that it can interplay with the flame scope heat maps. With current flame graphs there is no insight into what a program was doing at any moment in time, however the pairing of heat map + flame graph let’s you analyze sub second cpu usage patterns.
The alternative to building it in is supplying a conversion script to run after profiling. I didn’t see any documentation on the format of the trace otherwise would love to hack a script together for it!
I made the converter that translates async-profiler output to nflxprofile format.
This makes possible to view profiles collected by async-profiler with FlameScope.
Usage:
java -jar converter.jar input.jfr output.nflx
However, I believe it's better to address the issue from the other side:
FlameScope should support .jfr profiles
Let me explain why.
Therefore, if you are looking for a better integration of async-profiler and FlameScope, I suggest to
submit an issue in FlameScope project instead.
Most helpful comment
I made the converter that translates async-profiler output to nflxprofile format.
This makes possible to view profiles collected by async-profiler with FlameScope.
Usage:
However, I believe it's better to address the issue from the other side:
Let me explain why.
Therefore, if you are looking for a better integration of async-profiler and FlameScope, I suggest to
submit an issue in FlameScope project instead.