2009-08-29.

I remember doing this back in 2006, before discovering that 
the quality of normal SATA cables (rather than purpose-made
hot-swap cases) is such that removing and changing one of 
a group of disks will almost always result in extra ones 
getting temporarily jogged out of service too, and that it's
therefore not worth trying the hot-anything such systems.

Now, I want to remove a SATA disk from a running system, and
would rather not turn off.  But I know that it has hung at 
previous times when the disk just was pulled out.
Thanks to the web, the following (having ascertained that
0:0:0:0 is the SCSI name of the SATA disk -- each SATA disk
port is regarded as a separately numbered SCSI host) came
to light:

echo x >/sys/bus/scsi/devices/0\:0\:0\:0/delete

This worked fine: immediate sound of shutting down of a
disk (the disk was already unmounted), then no trouble from
the system about the disappearance of a disk.  This was on 
the expansion-card SiS (silicon systems) SATA controller; 
it's not sure that the integrated (old) Intel controller's
driver would support this hot-removal.


2009-09-29.

An alternative suggestion for removal:

echo 1 > /sys/block/sdd/device/delete

where sdd is the 'common' device name.
Would the '1' and 'x' be equivalent?


But on plugging in another SATA disk, even on the hotplug
chassis, there is not immediate detection: one has to ask
for a scan for new devices:

echo "- - -" > /sys/class/scsi_host/host0/scan

(or host1 or more, if on another controller).  This caused
the new disk to be recognised, just as though hotplugged 
on usb.