1Demonstrations of mdflush, the Linux eBPF/bcc version. 2 3 4The mdflush tool traces flushes at the md driver level, and prints details 5including the time of the flush: 6 7# ./mdflush 8Tracing md flush requests... Hit Ctrl-C to end. 9TIME PID COMM DEVICE 1003:13:49 16770 sync md0 1103:14:08 16864 sync md0 1203:14:49 496 kworker/1:0H md0 1303:14:49 488 xfsaild/md0 md0 1403:14:54 488 xfsaild/md0 md0 1503:15:00 488 xfsaild/md0 md0 1603:15:02 85 kswapd0 md0 1703:15:02 488 xfsaild/md0 md0 1803:15:05 488 xfsaild/md0 md0 1903:15:08 488 xfsaild/md0 md0 2003:15:10 488 xfsaild/md0 md0 2103:15:11 488 xfsaild/md0 md0 2203:15:11 488 xfsaild/md0 md0 2303:15:11 488 xfsaild/md0 md0 2403:15:11 488 xfsaild/md0 md0 2503:15:11 488 xfsaild/md0 md0 2603:15:12 488 xfsaild/md0 md0 2703:15:13 488 xfsaild/md0 md0 2803:15:15 488 xfsaild/md0 md0 2903:15:19 496 kworker/1:0H md0 3003:15:49 496 kworker/1:0H md0 3103:15:55 18840 sync md0 3203:16:49 496 kworker/1:0H md0 3303:17:19 496 kworker/1:0H md0 3403:20:19 496 kworker/1:0H md0 3503:21:19 496 kworker/1:0H md0 3603:21:49 496 kworker/1:0H md0 3703:25:19 496 kworker/1:0H md0 38[...] 39 40This can be useful for correlation with latency outliers or spikes in disk 41latency, as measured using another tool (eg, system monitoring). If spikes in 42disk latency often coincide with md flush events, then it would make flushing 43a target for tuning. 44 45Note that the flush events are likely to originate from higher in the I/O 46stack, such as from file systems. This traces md processing them, and the 47timestamp corresponds with when md began to issue the flush to disks. 48