Neil Brown wrote:
The array wasn't resyncing at the time was it?
I can confirm that the array was not resyncing. Running the same iostat as yesterday I now see that the reads are properly balanced. However, the IO load is a tenth of what it was yesterday. So it looks to me like the read-balancing somehow broke down yesterday - probably due to high pressure.
I've attached a graph of the IO activity in this period (taken from Munin). As writes (on top side) are written to both sda and sdb only the latter shows. On the bottom side of the graph you can see that reads from sda totally dominate.
-- Per Andreas Buer
