With generic event caching now implemented, there is no roadblock to the second part of the I/O Tracing support.
The remaining question is how file I/O should be displayed.
If we want to mirror block I/O, we would display it per file. I reckon that this view is impossible to display.
Alternatively, I would suggest we display it per thread.
If we still want to have filenames in some sort of capacity (such as event metadata) we have to have the "BPF talk", as filenames are pointers to kernel memory and can't be retrieved otherwise.
Continuation of #194
With generic event caching now implemented, there is no roadblock to the second part of the I/O Tracing support.
The remaining question is how file I/O should be displayed.
If we want to mirror block I/O, we would display it per file. I reckon that this view is impossible to display.
Alternatively, I would suggest we display it per thread.
If we still want to have filenames in some sort of capacity (such as event metadata) we have to have the "BPF talk", as filenames are pointers to kernel memory and can't be retrieved otherwise.