mdadm fails to start the raid array which i created on 2 usb drives. The reason is that the /dev entries for them are not created before mdadm tries start the array. The array is raid1 using 2 external 320 gig hds. Their dev entries are /dev/sdb* and /dev/sdc*. In the mean time i just start and mount the array manually in local.start.
what /dev entries ? you mean /dev/sdb* and /dev/sdc* ?
(In reply to comment #1) > what /dev entries ? you mean /dev/sdb* and /dev/sdc* ? > yea the entries for the usb drives.
not a mdadm bug
This should be a baselayout bug where raid is being started too early. When do the dev entries get created? after usbfs starts?
The dev entries should be created as soon as the related sysfs dirs exist. Then udev cares about them and creates the devices. But if for example your ehci_hcd is a module, then usb is started up as soon as udevd is started. The code scanning for the devices attached to the usb-ata/scsi chips can take a while. Perhaps this just takes longer than udevd needs to be started up, and thus mdadm is started too early. A. Either try to compile all modules into the kernel to give it more time for startup, or B. write a dummy rcscript addon that just waits (with a timeout) for your devices to appear.
Workaround: use usb-storage.delay_use=1 kernel parameter.
