On Tue, 7 May 2019 03:33:27 +0200 (CEST) Roy Sigurd Karlsbakk <roy@xxxxxxxxxxxxx> wrote: > Hi all > > I have sometimes vitnessed a RAID set slowing down tremendously and after some research, finding a single drive performing very badly (often down to 1-2% of what it should do). In zfs, I found this with zpool iostat (IIRC, that was some time back), but I'm not using zfs at home, just md. A friend just had a similar issue, so I tried nosing around looking for some counters to tell me what was lagging, but found none. Luckily, the raid only had six drives, so we tried hdparm -t on each of them, and one of them stood out with a speed of well below 1MB/s (the others were around 100MB/s, these being a diversity of old 1TB drives). Then I checked a drive that was kicked out of my home raid the other day, apparently for no reason (smart data looking ok etc, same thing with my friend's disk) and same thing there - perhaps 2MB/s on a Western Digital RE4 (6 years spinning time in one hour at this moment), which should be something like 150MB/s or thereabout. If you catch this slowdown event while it is occuring, you can run iostat -x /dev/sd? 2 to get a summary of disk I/O stats, including "utilization" percentage. The slow disks will display 100% in the rightmost column, while others will be mostly idle. -- With respect, Roman
