#!/usr/bin/perl -w

# By Marc MERLIN <marc_soft@merlins.org>
# $Id: swdisksusp 353 2011-09-24 20:52:16Z svnuser $

# Home page:
# http://marc.merlins.org/perso/linux/post_2010-08-03_Spinning-Down-WD20EADS-Drives-and-Fixing-Load-Cycle.html

# Manually spin down drives that will not spin down with hdparm -S x /dev/device
# like they should.
# This script is meant to be run as "swdisksusp &" as it logs to stderr It
# sleeps 10 seconds between each try, so it's kind of pointless to set the
# timeout to less than 10 seconds, which you wouldn't want to do for spinning
# down anyway.
# iostat is required as well as kernel support for it before this script
# will work.

# Western Digital green drives like the WD20EADS are affected by this. They
# are so busy trying to park their heads every 6 seconds (watch Load Cycle in
# SMART) that their firmware seems to ignore the -S timeout value altogether
# (incidently both for parking heads or spinning down the platters :( ).

# If your system tickles the drive in any way (SMART query, including hddtemp,
# or apparently LVM or software raid or something else on my system woke the
# drives up constantly), your Load Cycle count will shoot through the roof (up
# to 2000/day for me, or about 1.5 years of expected parking/unparking lifetime
# with a limit of about 1 million). Note that in my case I was not keeping the
# filesystem busy since iostat did show no disk access and spinning down the
# drives with hdparm -y /dev/device showed that they stayed spun down for 10 to
# 20mn at a time. 
# It's not entirely clear why those drives seem to unpark when no data
# is requested from them.
# You can check Load Cycle Count with:
# smartctl -A /dev/device | grep Load_Cycle_Count

# To stop this unfortunate behaviour, you need to run the unapproved but working
# wdidle3 from http://support.wdc.com/product/download.asp
# (yes it says it's not supported on all but 3 drives, but it has worked for me and others.
# I personally set /S300 and it has stopped my parking).


use strict;

my $time = time;
# BEGIN Variables to set are
my %watchdevs = (
    "sdh" => $time,
    "sdi" => $time,
    "sdj" => $time,
    "sdk" => $time,
    "sdl" => $time,
);

# How many seconds of inactivity before we spin down a drive
# you can override on the command line
my $sleep_timeout = 300;
# END   Variables to set are

my $stderr = "";
my $DEBUG = 0;
if (defined $ARGV[0] and $ARGV[0] eq "--debug")
{
    $DEBUG = 1;
    $stderr = "-s";
    shift @ARGV;
}
$sleep_timeout = $ARGV[0] if (defined $ARGV[0] and $ARGV[0]);

print STDERR "Will monitor ".join(", ", keys (%watchdevs))." and spin them down after ${sleep_timeout}s of inactivity\n";

while (1) 
{
    # Show 2 seconds of iostat, and skip the first run that shows init counters
    # Any devices shown in the subsequent run saw data traffic.
    open(IOSTAT, "iostat -z 1 2 |");
    # Skip first 5 lines of headers.
    for (my $i=1; $i<5; $i++)
    {
	$_ = <IOSTAT>;
    }
    # and keep looking for output until the the first batch of drives is fully shown.
    while (<IOSTAT>)
    {
	last if /^avg-cpu:/;
    }
#    avg-cpu:  %user   %nice %system %iowait  %steal   %idle
#	       18.30    0.00    5.51    1.50    0.00   74.69
#
#    Device:            tps   Blk_read/s   Blk_wrtn/s   Blk_read   Blk_wrtn
#    sdh               3.00         0.00         4.00          0          4
#    sdi               3.00         0.00         4.00          0          4
#    sdj               8.00        48.00        52.00         48         52
#    sdk               8.00        48.00        52.00         48         52
#    sdl               3.00         0.00         4.00          0          4
#    md7               7.00         0.00        48.00          0         48
#    dm-2              6.00         0.00        48.00          0         48

    $time = time;
    # Grab all the device names that saw any update (read or write) since init.
    while (<IOSTAT>)
    {
	my $dev;

	/^([hs]d\S)\s+/ or next;
	$dev = $1;

	#warn "Testing $dev at $time\n";
	# $watchdevs{$dev} contains a positive time for a last update with the drive running.
	# and a negative time if that last update was before the drive was put to sleep.
	if (defined $watchdevs{$dev})
	{
	    system("logger $stderr \"$dev came out of sleep after ".($time + $watchdevs{$dev})."s of inactivity\"") if ($watchdevs{$dev} < 0);
	    $watchdevs{$dev} = $time;
	}
    }
    close(IOSTAT);

    foreach my $dev (keys %watchdevs)
    {
	if ($watchdevs{$dev} > 0 and $time - $watchdevs{$dev} > $sleep_timeout)
	{
	    if ($stderr)
	    {
		system("hdparm -y /dev/$dev");
	    }
	    else
	    {
		system("hdparm -y /dev/$dev &>/dev/null");
	    }
	    system("logger $stderr \"Putting $dev to sleep after ${sleep_timeout}s of inactivity\"");
	    $watchdevs{$dev} = -$time;
	}
    }
    sleep 10;
}

# vim:sts=4:sw=4