writeback.bt(8) System Manager's Manual writeback.bt(8)

writeback.bt - Trace file system writeback events with details. Uses bpftrace/eBPF.

writeback.bt

This traces when file system dirtied pages are flushed to disk by kernel writeback, and prints details including when the event occurred, and the duration of the event. This can be useful for correlating these times with other performance problems, and if there is a match, it would be a clue that the problem may be caused by writeback. How quickly the kernel does writeback can be tuned: see the kernel docs, eg, vm.dirty_writeback_centisecs.

This uses the tracepoint:writeback:writeback_start and tracepoint:writeback:writeback_written tracepoints.

Since this uses BPF, only the root user can use this tool.

CONFIG_BPF and bpftrace.

# writeback.bt

Time that the writeback event completed, in %H:%M:%S format.
Device name in major:minor number format.
Pages written during writeback.
Reason for the writeback event. This may be "background", "vmscan", "sync", "periodic", etc.
Duration of the writeback event in milliseconds.

Since writeback events are expected to be infrequent (<10/sec), the overhead of this tool is expected to be negligible (near 0%).

This is from bpftrace.

https://github.com/bpftrace/bpftrace

Also look in the bpftrace distribution for a companion _examples.txt file containing example usage, output, and commentary for this tool.

Linux

Unstable - in development.

Brendan Gregg

biosnoop.bt(8)

2018-09-14 USER COMMANDS