Open milianw opened 8 years ago
The per-process I/O report is already the sum of every I/O operation made to all the FDs belonging to a process, and the per-file report combines all the I/O made by every process that have opened this file. I am not sure what you mean exactly, could you maybe provide an example ?
For the blocks, it is more tricky, because we do not have a definitive mapping between syscalls and block operations, especially with caching/writeback. It really is a best effort here because we do not have all the instrumentation required.
Right now when I use the
lttng-iousagetop
command, I get useful Per-process I/O, Per-file I/O and Block I/O reports. What I'm missing is a way to combine this, to get e.g. the per-process I/O for every file and ideally also for every block.That would allow one to figure out which file was written to by what process, and on what block that file lives.