On Mon, 15 Sep 2014 12:52:07 +0200 Marco Schindler <marco.schindler@xxxxxxxxx> wrote: > > On 15.09.2014, at 02:18, NeilBrown <neilb@xxxxxxx> wrote: > > > It would help to get "udevadm monitor" info to correlate with this. > > Presumably some uevent is generated when the spindown happens. udev might > > respond to this by reading from the device, which defeats the purpose... > > > > > >> > >> here’s the output of blktrace -d /dev/sda during that time. > >> https://dl.dropboxusercontent.com/u/3464720/blktrace.tar.bz2 > > > > That suggest that something is reading the metadata from the device almost > > constantly. Mostly a 'kworker' thread. I don't know what would cause that. > > > > Let's look at the 'udevadm monitor' trace first and see what that shows. > > here’s the output of udevadm monitor during the spindown cycle while mdadm-3.3 is installed. > > monitor will print the received events for: > UDEV - the event which udev sends out after rule processing > KERNEL - the kernel uevent > > KERNEL[299719.336261] change /devices/pci0000:00/0000:00:06.0/0000:09:00.0/host0/port-0:1/end_device-0:1/target0:0:1/0:0:1:0/block/sdb (block) > UDEV [299720.646760] change /devices/pci0000:00/0000:00:06.0/0000:09:00.0/host0/port-0:1/end_device-0:1/target0:0:1/0:0:1:0/block/sdb (block) > KERNEL[299780.202901] change /devices/pci0000:00/0000:00:06.0/0000:09:00.0/host0/port-0:0/end_device-0:0/target0:0:0/0:0:0:0/block/sda (block) > UDEV [299781.567308] change /devices/pci0000:00/0000:00:06.0/0000:09:00.0/host0/port-0:0/end_device-0:0/target0:0:0/0:0:0:0/block/sda (block) > KERNEL[299841.090818] change /devices/pci0000:00/0000:00:06.0/0000:09:00.0/host0/port-0:1/end_device-0:1/target0:0:1/0:0:1:0/block/sdb (block) > UDEV [299842.407035] change /devices/pci0000:00/0000:00:06.0/0000:09:00.0/host0/port-0:1/end_device-0:1/target0:0:1/0:0:1:0/block/sdb (block) > > please note that the issue immediately disappears when downgrading to mdadm-3.2 without touching anything else. > I see the udev rules have been updated in mdadm-3.3.. Getting a "change" even on spindown is causing the problem I suspect. A change in 3.3.1 causes "mdadm -I" to be run on a device when it 'changes'. That will read from the device which will wake it up. (commit 25392f5fc59f96fb76 - revert it and the symptom will probably go away). I really think the "bug" here is that the change event is emitted on 'spindown', but maybe the bug is that the exact meaning of 'change' isn't well documented. I can probably get "mdadm -I" to use O_EXCL which will fail on devices already in an array, but I'm not sure that is a complete solution. You could still get wakeups on other devices. Can you rung the 'udevadm monitor' again, but this time with '--property'. Maybe there is some property associated with spindown events which we can use to ignore them. NeilBrown
Attachment:
signature.asc
Description: PGP signature
