# This program is copyright 2011 Percona Inc.
# Feedback and improvements are welcome.
#
# THIS PROGRAM IS PROVIDED "AS IS" AND WITHOUT ANY EXPRESS OR IMPLIED
# WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF
# MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE.
#
# This program is free software; you can redistribute it and/or modify it under
# the terms of the GNU General Public License as published by the Free Software
# Foundation, version 2; OR the Perl Artistic License.  On UNIX and similar
# systems, you can issue `man perlgpl' or `man perlartistic' to read these
# licenses.
#
# You should have received a copy of the GNU General Public License along with
# this program; if not, write to the Free Software Foundation, Inc., 59 Temple
# Place, Suite 330, Boston, MA  02111-1307  USA.
# ###########################################################################
# Diskstats package
# ###########################################################################
{
# Package: Diskstats
# This package implements most of the logic in the old shell pt-diskstats;
# it parses data from /proc/diskstats, calculcates deltas, and prints those.

package Diskstats;

use strict;
use warnings FATAL => 'all';
use English qw(-no_match_vars);
use constant PTDEBUG => $ENV{PTDEBUG} || 0;

use IO::Handle;
use List::Util qw( max first );

sub new {
   my ( $class, %args ) = @_;

   my @required_args = qw(OptionParser);
   foreach my $arg ( @required_args ) {
      die "I need a $arg argument" unless $args{$arg};
   }
   my ($o) = @args{@required_args};

   my $self = {
      # Defaults
      filename           => '/proc/diskstats',
      column_regex       => qr/cnc|rt|busy|prg|time|io_s/,
      device_regex       => qr/(?=)/,
      block_size         => 512,
      out_fh             => \*STDOUT,
      filter_zeroed_rows => $o->get('zero-rows') ? undef : 1,
      sample_time        => $o->get('sample-time') || 0,
      interactive        => 0,

      _stats_for         => {},
      _ordered_devs      => [],
      _ts                => {},
      _first             => 1,

      # Internal for now, but might need APIfying.
      _save_curr_as_prev => 1,
      _print_header      => 1,
   };

   if ( $o->get('memory-for-speed') ) {
      PTDEBUG && _d('Diskstats', "Called with memory-for-speed");
      eval {
         require Memoize;
         Memoize::memoize('_parse_diskstats_line');
      };
      if ($EVAL_ERROR) {
         warn "Can't trade memory for speed: $EVAL_ERROR. Continuing as usual.";
      }
   }

   my %pod_to_attribute = (
      columns => 'column_regex',
      devices => 'device_regex'
   );
   for my $key ( grep { defined $o->get($_) } keys %pod_to_attribute ) {
      my $re = $o->get($key) || '(?=)';
      $self->{ $pod_to_attribute{$key} } = qr/$re/i;
   }

   # If they passed us an attribute explicitly, we use those.
   for my $attribute ( grep { !/^_/ && defined $args{$_} } keys %$self ) {
      $self->{$attribute} = $args{$attribute};
   }

   return bless $self, $class;
}

# The next lot are accessors, plus some convenience functions.

sub curr_ts {
   my ($self) = @_;
   return $self->{_ts}->{curr} || 0;
}

sub set_curr_ts {
   my ($self, $val) = @_;
   $self->{_ts}->{curr} = $val || 0;
}

sub prev_ts {
   my ($self) = @_;
   return $self->{_ts}->{prev} || 0;
}

sub set_prev_ts {
   my ($self, $val) = @_;
   $self->{_ts}->{prev} = $val || 0;
}

sub first_ts {
   my ($self) = @_;
   return $self->{_ts}->{first} || 0;
}

sub set_first_ts {
   my ($self, $val) = @_;
   $self->{_ts}->{first} = $val || 0;
}

sub filter_zeroed_rows {
   my ($self) = @_;
   return $self->{filter_zeroed_rows};
}

sub set_filter_zeroed_rows {
   my ($self, $new_val) = @_;
   $self->{filter_zeroed_rows} = $new_val;
}

sub sample_time {
   my ($self) = @_;
   return $self->{sample_time};
}

sub set_sample_time {
   my ($self, $new_val) = @_;
   if (defined($new_val)) {
      $self->{sample_time} = $new_val;
   }
}

sub interactive {
   my ($self, $new_val) = @_;
   if (defined($new_val)) {
      $self->{interactive} = $new_val;
   }
   return $self->{interactive};
}

# What this method does is thee-fold:
# It sets or returns the currently set filehandle, kind of like a poor man's
# select(); but also, it checks whenever said filehandle is open. If it's not,
# it defaults to STDOUT.

sub out_fh {
   my ( $self, $new_fh ) = @_;

                  # ->opened comes from IO::Handle.
   if ( $new_fh && ref($new_fh) && $new_fh->opened ) {
      $self->{out_fh} = $new_fh;
   }
   if ( !$self->{out_fh} || !$self->{out_fh}->opened ) {
      $self->{out_fh} = \*STDOUT;
   }
   return $self->{out_fh};
}

sub column_regex {
   my ( $self, $new_re ) = @_;
   if ($new_re) {
      return $self->{column_regex} = $new_re;
   }
   return $self->{column_regex};
}

sub device_regex {
   my ( $self, $new_re ) = @_;
   if ($new_re) {
      return $self->{device_regex} = $new_re;
   }
   return $self->{device_regex};
}

sub filename {
   my ( $self, $new_filename ) = @_;
   if ( $new_filename ) {
      return $self->{filename} = $new_filename;
   }
   return $self->{filename};
}

sub block_size {
   my $self = shift;
   return $self->{block_size};
}

# Returns a list of devices seen. You may pass an arrayref argument to
# replace the internal list, but consider using clear_ordered_devs and
# add_ordered_dev instead.

sub ordered_devs {
   my ( $self, $replacement_list ) = @_;
   if ( $replacement_list ) {
      $self->{_ordered_devs} = $replacement_list;
   }
   return @{ $self->{_ordered_devs} };
}

sub add_ordered_dev {
   my ( $self, $new_dev ) = @_;
   if ( !$self->{_seen_devs}->{$new_dev}++ ) {
      push @{ $self->{_ordered_devs} }, $new_dev;
   }
   return;
}

# clear_stuff methods. Like the name says, they clear state stored inside
# the object.

sub clear_state {
   my ($self) = @_;
   $self->{_first} = 1;
   $self->{_print_header} = 1;
   $self->clear_curr_stats();
   $self->clear_prev_stats();
   $self->clear_first_stats();
   $self->clear_ts();
   $self->clear_ordered_devs();
}

sub clear_ts {
   my ($self) = @_;
   $self->{_ts} = {};
}

sub clear_ordered_devs {
   my $self = shift;
   $self->{_seen_devs} = {};
   $self->ordered_devs( [] );
}

sub _clear_stats_common {
   my ( $self, $key, @args ) = @_;
   if (@args) {
      for my $dev (@args) {
         $self->{$key}->{$dev} = {};
      }
   }
   else {
      $self->{$key} = {};
   }
}

sub clear_curr_stats {
   my ( $self, @args ) = @_;
   $self->_clear_stats_common( "_stats_for", @args );
}

sub clear_prev_stats {
   my ( $self, @args ) = @_;
   $self->_clear_stats_common( "_prev_stats_for", @args );
}

sub clear_first_stats {
   my ( $self, @args ) = @_;
   $self->_clear_stats_common( "_first_stats_for", @args );
}

sub stats_for {
   my ( $self, $dev ) = @_;
   $self->{_stats_for} ||= {};
   if ($dev) {
      return $self->{_stats_for}->{$dev};
   }
   return $self->{_stats_for};
}

sub prev_stats_for {
   my ( $self, $dev ) = @_;
   $self->{_prev_stats_for} ||= {};
   if ($dev) {
      return $self->{_prev_stats_for}->{$dev};
   }
   return $self->{_prev_stats_for};
}

sub first_stats_for {
   my ( $self, $dev ) = @_;
   $self->{_first_stats_for} ||= {};
   if ($dev) {
      return $self->{_first_stats_for}->{$dev};
   }
   return $self->{_first_stats_for};
}

sub has_stats {
   my ($self) = @_;
   my $stats  = $self->stats_for;

   for my $key ( keys %$stats ) {
      return 1 if $stats->{$key} && %{ $stats->{$key} }
   }

   return;
}

sub _save_curr_as_prev {
   my ( $self, $curr ) = @_;

   if ( $self->{_save_curr_as_prev} ) {
      $self->{_prev_stats_for} = $curr;
      for my $dev (keys %$curr) {
         $self->{_prev_stats_for}->{$dev}->{sum_ios_in_progress} +=
            $curr->{$dev}->{ios_in_progress};
      }
      $self->set_prev_ts($self->curr_ts());
   }

   return;
}

sub _save_curr_as_first {
   my ($self, $curr) = @_;

   if ( $self->{_first} ) {
      $self->{_first_stats_for} = {
         # 1-level deep copy of the original structure. Should
         # be enough.
         map { $_ => {%{$curr->{$_}}} } keys %$curr
      };
      $self->set_first_ts($self->curr_ts());
      $self->{_first} = undef;
   }
}

sub _save_stats {
   my ( $self, $stats ) = @_;
   return $self->{_stats_for} = $stats;
}

sub trim {
   my ($c) = @_;
   $c =~ s/^\s+//;
   $c =~ s/\s+$//;
   return $c;
}

sub col_ok {
   my ( $self, $column ) = @_;
   my $regex = $self->column_regex();
   return ($column =~ $regex) || (trim($column) =~ $regex);
}

sub dev_ok {
   my ( $self, $device ) = @_;
   my $regex = $self->device_regex();
   return $device =~ $regex;
}

our @columns_in_order = (
   # Column        # Format   # Key name
   [ "   rd_s" => "%7.1f",   "reads_sec", ],
   [ "rd_avkb" => "%7.1f",   "avg_read_sz", ],
   [ "rd_mb_s" => "%7.1f",   "mbytes_read_sec", ],
   [ "rd_io_s" => "%7.1f",   "ios_read_sec", ],
   [ "rd_mrg"  => "%5.0f%%", "read_merge_pct", ],
   [ "rd_cnc"  => "%6.1f",   "read_conc", ],
   [ "  rd_rt" => "%7.1f",   "read_rtime", ],
   [ "   wr_s" => "%7.1f",   "writes_sec", ],
   [ "wr_avkb" => "%7.1f",   "avg_write_sz", ],
   [ "wr_mb_s" => "%7.1f",   "mbytes_written_sec", ],
   [ "wr_io_s" => "%7.1f",   "ios_written_sec", ],
   [ "wr_mrg"  => "%5.0f%%", "write_merge_pct", ],
   [ "wr_cnc"  => "%6.1f",   "write_conc", ],
   [ "  wr_rt" => "%7.1f",   "write_rtime", ],
   [ "busy"    => "%3.0f%%", "busy", ],
   [ "in_prg"  => "%6d",     "in_progress", ],
   [ "   io_s" => "%7.1f",   "s_spent_doing_io", ],
   [ " qtime"   => "%6.1f",   "qtime", ],
   [ " stime"   => "%5.1f",   "stime", ],
);

{

   my %format_for = ( map { ( $_->[0] => $_->[1] ) } @columns_in_order, );

   sub _format_for {
      my ( $self, $col ) = @_;
      return $format_for{$col};
   }

}

{

   my %column_to_key = ( map { ( $_->[0] => $_->[2] ) } @columns_in_order, );

   sub _column_to_key {
      my ( $self, $col ) = @_;
      return $column_to_key{$col};
   }

}

# Method: design_print_formats()
#   What says on the label. Returns three things: the format for the header
#   and the data, and an arrayref of the columns used to make it.
#
# Parameters:
#   %args - Arguments
#
# Optional Arguments:
#   columns             - An arrayref with column names. If absent,
#                         uses ->col_ok to decide which columns to use.
#   max_device_length   - How much space to leave for device names.
#                         Defaults to 6.
#

sub design_print_formats {
   my ( $self,       %args )    = @_;
   my ( $dev_length, $columns ) = @args{qw( max_device_length columns )};
   $dev_length ||= max 6, map length, $self->ordered_devs;
   my ( $header, $format );

   # For each device, print out the following: The timestamp offset and
   # device name.
   $header = $format = qq{%5s %-${dev_length}s };

   if ( !$columns ) {
      @$columns = grep { $self->col_ok($_) } map { $_->[0] } @columns_in_order;
   }
   elsif ( !ref($columns) || ref($columns) ne ref([]) ) {
      die "The columns argument to design_print_formats should be an arrayref";
   }

   $header .= join " ", @$columns;
   $format .= join " ", map $self->_format_for($_), @$columns;

   return ( $header, $format, $columns );
}

{
# This is hot code. In any given run it could end up being called
# thousands of times, so beware: Here could be dragons.
my @diskstats_fields = qw(
   reads  reads_merged  read_sectors      ms_spent_reading
   writes writes_merged written_sectors   ms_spent_writing
   ios_in_progress      ms_spent_doing_io ms_weighted
);
# This allows parse_diskstats_line() to be overriden, but also to be
# memoized without a normalization function.

# Magic goto, removes this function from the return stack. Haven't
# benchmarked it, but ostensibly faster.
sub parse_diskstats_line  { shift; goto &_parse_diskstats_line }
sub _parse_diskstats_line {
   my ( $line, $block_size ) = @_;
   my $dev;
   keys my %dev_stats = 30; # Pre-expand the amount of buckets for this hash.

#   The following split replaces this:
#         $line =~ /^
#            # Disk format
#               \s*   (\d+)    # major
#               \s+   (\d+)    # minor
#               \s+   (.+?)    # Device name
#               \s+   (\d+)    # # of reads issued
#               \s+   (\d+)    # # of reads merged
#               \s+   (\d+)    # # of sectors read
#               \s+   (\d+)    # # of milliseconds spent reading
#               \s+   (\d+)    # # of writes completed
#               \s+   (\d+)    # # of writes merged
#               \s+   (\d+)    # # of sectors written
#               \s+   (\d+)    # # of milliseconds spent writing
#               \s+   (\d+)    # # of IOs currently in progress
#               \s+   (\d+)    # # of milliseconds spent doing IOs
#               \s+   (\d+)    # weighted # of milliseconds spent doing IOs
#               \s*$/x
#
#   Since we assume that device names can't have spaces.

   # Assigns the first two elements of the list created by split() into
   # %dev_stats as the major and minor, the third element into $dev,
   # and the remaining elements back into %dev_stats.
   if ( 14 == (( @dev_stats{qw( major minor )}, $dev, @dev_stats{@diskstats_fields} ) =
         split " ", $line, 14 ) )
   {
      $dev_stats{read_kbs}    =
         ( $dev_stats{read_bytes} = $dev_stats{read_sectors}
                                  * $block_size ) / 1024;
      $dev_stats{written_kbs} =
         ( $dev_stats{written_bytes} = $dev_stats{written_sectors}
                                     * $block_size ) / 1024;
      $dev_stats{ios_requested} = $dev_stats{reads}
                                + $dev_stats{writes};

      $dev_stats{ios_in_bytes}  = $dev_stats{read_bytes}
                                + $dev_stats{written_bytes};

      return ( $dev, \%dev_stats );
   }
   elsif ((@dev_stats{qw( major minor )}, $dev,
           @dev_stats{ qw( reads read_sectors writes written_sectors ) }) =
           $line =~ /^
            # Partition format
               \s*   (\d+)    # major
               \s+   (\d+)    # minor
               \s+   (.+?)    # Device name
               \s+   (\d+)    # # of reads issued
               \s+   (\d+)    # # of sectors read
               \s+   (\d+)    # # of writes issued
               \s+   (\d+)    # # of sectors written
               \s*$/x)
   {
      for my $key ( @diskstats_fields ) {
         # Unintiialized values should be 0
         $dev_stats{$key} ||= 0;
      }
      # Copypaste from above, should probably abstract, but it would make
      # the common case slower.
      $dev_stats{read_bytes} = $dev_stats{read_sectors} * $block_size;
      $dev_stats{written_bytes} =
        $dev_stats{written_sectors} * $block_size;
      $dev_stats{read_kbs}    = $dev_stats{read_bytes} / 1024;
      $dev_stats{written_kbs} = $dev_stats{written_bytes} / 1024;
      $dev_stats{ios_requested} = $dev_stats{reads} + $dev_stats{writes};
      $dev_stats{ios_in_bytes}  = $dev_stats{read_bytes}
                                + $dev_stats{written_bytes};

      return ( $dev, \%dev_stats );
   }
   else {
      return;
   }
}
}

# Method: parse_from()
#   Parses data from one of the sources.
#
# Parameters:
#   %args - Arguments
#
# Optional Arguments:
#   filehandle       - Reads data from a filehandle by calling readline()
#                      on it.
#   data             - Reads data one line at a time.
#   filename         - Opens a filehandle to the file and reads it one
#                      line at a time.
#   sample_callback  - Called each time a sample is processed, passed
#                      the latest timestamp.
#

sub parse_from {
    my ( $self, %args ) = @_;

    my $lines_read = $args{filehandle}
      ? $self->parse_from_filehandle( @args{qw( filehandle sample_callback )} )
      : $args{data}
      ? $self->parse_from_data( @args{qw( data sample_callback )} )
      : $self->parse_from_filename( @args{qw( filename sample_callback )} );
    return $lines_read;
}


sub parse_from_filename {
   my ( $self, $filename, $sample_callback ) = @_;

   $filename ||= $self->filename();

   open my $fh, "<", $filename
     or die "Cannot parse $filename: $OS_ERROR";
   my $lines_read = $self->parse_from_filehandle( $fh, $sample_callback );
   close $fh or die "Cannot close: $OS_ERROR";

   return $lines_read;
}

# Method: parse_from_filehandle()
#   Parses data received from using readline() on the filehandle. This is
#   particularly useful, as you could pass in a filehandle to a pipe, or
#   a tied filehandle, or a PerlIO::Scalar handle. Or your normal
#   run of the mill filehandle.
#
# Parameters:
#   filehandle       - 
#   sample_callback  - Called each time a sample is processed, passed
#                      the latest timestamp.
#

sub parse_from_filehandle {
   my ( $self, $filehandle, $sample_callback ) = @_;
   return $self->_load( $filehandle, $sample_callback );
}

# Method: parse_from_data()
#   Similar to parse_from_filehandle, but uses a reference to a scalar
#   as a filehandle
#
# Parameters:
#   data             - A normal Perl scalar, or a ref to a scalar.
#   sample_callback  - Same as parse_from_filehandle.
#
sub parse_from_data {
   my ( $self, $data, $sample_callback ) = @_;

   open( my $fh, "<", ref($data) ? $data : \$data )
     or die "Couldn't parse data: $OS_ERROR";
   my $lines_read = $self->parse_from_filehandle( $fh, $sample_callback );
   close $fh or die "";

   return $lines_read;
}

# Method: _load()
#   !!!!INTERNAL!!!!!
#   Reads from the filehandle, either saving the data as needed if dealing
#   with a diskstats-formatted line, or if it finds a TS line and has a
#   callback, defering to that.

sub _load {
   my ( $self, $fh, $sample_callback ) = @_;
   my $block_size = $self->block_size();
   my $current_ts = 0;
   my $new_cur    = {};

   while ( my $line = <$fh> ) {
      if ( my ( $dev, $dev_stats ) = $self->parse_diskstats_line($line, $block_size) )
      {
         $new_cur->{$dev} = $dev_stats;
         $self->add_ordered_dev($dev);
      }
      elsif ( my ($new_ts) = $line =~ /TS\s+([0-9]+(?:\.[0-9]+)?)/ ) {
         if ( $current_ts && %$new_cur ) {
            $self->_save_curr_as_prev( $self->stats_for() );
            $self->_save_stats($new_cur);
            $self->set_curr_ts($current_ts);
            $self->_save_curr_as_first( $new_cur );
            $new_cur = {};
         }
         if ($sample_callback) {
            $self->$sample_callback($current_ts);
         }
         $current_ts = $new_ts;
      }
      else {
         chomp($line);
         warn "Line $INPUT_LINE_NUMBER: [$line] isn't in the diskstats format";
      }
   }

   if ( $current_ts ) {
      if ( %{$new_cur} ) {
         $self->_save_curr_as_prev( $self->stats_for() );
         $self->_save_stats($new_cur);
         $self->set_curr_ts($current_ts);
         $self->_save_curr_as_first( $new_cur );
         $new_cur = {};
      }
      if ($sample_callback) {
         $self->$sample_callback($current_ts);
      }
   }
   # Seems like this could be useful.
   return $INPUT_LINE_NUMBER;
}

sub _calc_read_stats {
   my ( $self, %args ) = @_;

   my @required_args = qw( delta_for elapsed devs_in_group );
   foreach my $arg ( @required_args ) {
      die "I need a $arg argument" unless $args{$arg};
   }
   my ($delta_for, $elapsed, $devs_in_group) = @args{ @required_args };

   my %read_stats = (
      reads_sec       => $delta_for->{reads} / $elapsed,
      read_requests   => $delta_for->{reads_merged} + $delta_for->{reads},
      mbytes_read_sec => $delta_for->{read_kbs} / $elapsed / 1024,
      ios_read_sec    => $delta_for->{ms_spent_reading} / 1000,
      read_conc       => $delta_for->{ms_spent_reading} /
                           $elapsed / 1000 / $devs_in_group,
   );

   if ( $delta_for->{reads} > 0 ) {
      $read_stats{read_rtime} =
        $delta_for->{ms_spent_reading} / $delta_for->{reads};
      $read_stats{avg_read_sz} =
        $delta_for->{read_kbs} / $delta_for->{reads};
   }
   else {
      $read_stats{read_rtime}  = 0;
      $read_stats{avg_read_sz} = 0;
   }

   $read_stats{read_merge_pct} =
     $read_stats{read_requests} > 0
     ? 100 * $delta_for->{reads_merged} / $read_stats{read_requests}
     : 0;

   return %read_stats;
}

sub _calc_write_stats {
   my ( $self, %args ) = @_;

   my @required_args = qw( delta_for elapsed devs_in_group );
   foreach my $arg ( @required_args ) {
      die "I need a $arg argument" unless $args{$arg};
   }
   my ($delta_for, $elapsed, $devs_in_group) = @args{ @required_args };

   my %write_stats = (
      writes_sec     => $delta_for->{writes} / $elapsed,
      write_requests => $delta_for->{writes_merged} + $delta_for->{writes},
      mbytes_written_sec  => $delta_for->{written_kbs} / $elapsed / 1024,
      ios_written_sec    => $delta_for->{ms_spent_writing} / 1000,
      write_conc         => $delta_for->{ms_spent_writing} /
        $elapsed / 1000 /
        $devs_in_group,
   );

   if ( $delta_for->{writes} > 0 ) {
      $write_stats{write_rtime} =
        $delta_for->{ms_spent_writing} / $delta_for->{writes};
      $write_stats{avg_write_sz} =
        $delta_for->{written_kbs} / $delta_for->{writes};
   }
   else {
      $write_stats{write_rtime}  = 0;
      $write_stats{avg_write_sz} = 0;
   }

   $write_stats{write_merge_pct} =
     $write_stats{write_requests} > 0
     ? 100 * $delta_for->{writes_merged} / $write_stats{write_requests}
     : 0;

   return %write_stats;
}


# Compute the numbers for reads and writes together, the things for
# which we do not have separate statistics.

sub _calc_misc_stats {
   my ( $self, %args ) = @_;

   my @required_args = qw( delta_for elapsed devs_in_group stats );
   foreach my $arg ( @required_args ) {
      die "I need a $arg argument" unless $args{$arg};
   }
   my ($delta_for, $elapsed, $devs_in_group, $stats) = @args{ @required_args };
   my %extra_stats;

   # Busy is what iostat calls %util.  This is the percent of
   # wall-clock time during which the device has I/O happening.
   $extra_stats{busy} =
      100 *
      $delta_for->{ms_spent_doing_io} /
      ( 1000 * $elapsed * $devs_in_group );

   my $number_of_ios        = $stats->{ios_requested};
   my $total_ms_spent_on_io = $delta_for->{ms_spent_reading}
                            + $delta_for->{ms_spent_writing};

   if ( $number_of_ios ) {
      $extra_stats{qtime} = $total_ms_spent_on_io / $number_of_ios;
      $extra_stats{stime} = $delta_for->{ms_spent_doing_io} / $number_of_ios;
   }
   else {
      $extra_stats{qtime} = 0;
      $extra_stats{stime} = 0;
   }

   $extra_stats{s_spent_doing_io} = $total_ms_spent_on_io / 1000;

   $extra_stats{line_ts} = $self->compute_line_ts(
      first_ts   => $self->first_ts(),
      curr_ts    => $self->curr_ts(),
   );

   return %extra_stats;
}

sub _calc_delta_for {
   my ( $self, $curr, $against ) = @_;
   my %deltas = (
      map { ( $_ => ($curr->{$_} || 0) - ($against->{$_} || 0) ) }
        qw(
         reads reads_merged read_sectors ms_spent_reading
         writes writes_merged written_sectors ms_spent_writing
         read_kbs written_kbs
         ms_spent_doing_io ms_weighted
        )
   );
   return \%deltas;
}

sub _calc_stats_for_deltas {
   my ( $self, $elapsed ) = @_;
   my @end_stats;
   my @devices = $self->ordered_devs();

   my $devs_in_group = $self->compute_devs_in_group();

   # Read "For each device that passes the dev_ok regex, and we have stats for"
   foreach my $dev_and_curr (
         map {
            my $curr = $self->dev_ok($_) && $self->stats_for($_);
            $curr ? [ $_, $curr ] : ()
         }
         @devices )
   {
      my $dev     = $dev_and_curr->[0];
      my $curr    = $dev_and_curr->[1];
      my $against = $self->delta_against($dev);

      my $delta_for       = $self->_calc_delta_for( $curr, $against );
      my $in_progress     = $curr->{"ios_in_progress"};
      my $tot_in_progress = $against->{"sum_ios_in_progress"} || 0;

      # Compute the per-second stats for reads, writes, and overall.
      my %stats = (
         $self->_calc_read_stats(
            delta_for     => $delta_for,
            elapsed       => $elapsed,
            devs_in_group => $devs_in_group,
         ),
         $self->_calc_write_stats(
            delta_for     => $delta_for,
            elapsed       => $elapsed,
            devs_in_group => $devs_in_group,
         ),
         in_progress =>
           $self->compute_in_progress( $in_progress, $tot_in_progress ),
      );

      my %extras = $self->_calc_misc_stats(
         delta_for     => $delta_for,
         elapsed       => $elapsed,
         devs_in_group => $devs_in_group,
         stats         => \%stats,
      );

      @stats{ keys %extras } = values %extras;

      $stats{dev} = $dev;

      push @end_stats, \%stats;
   }
   return @end_stats;
}

sub _calc_deltas {
   my ( $self ) = @_;

   my $elapsed = $self->curr_ts() - $self->delta_against_ts();
   die "Time elapsed is [$elapsed]" unless $elapsed;

   return $self->_calc_stats_for_deltas($elapsed);
}

sub print_header {
   my ($self, $header, @args) = @_;
   if ( $self->{_print_header} ) {
      printf { $self->out_fh() } $header . "\n", @args;
   }
}

sub print_rows {
   my ($self, $format, $cols, $stat) = @_;
   if ( $self->filter_zeroed_rows() ) {
      # Conundrum: What is "zero"?
      # Is 0.000001 zero? How about 0.1?
      # Here the answer is "it looks like zero after formatting";
      # unfortunately, we lack the formats at this point. We could
      # fetch them again, but that's a pain, so instead we use
      # %7.1f, which is what most of them are anyway, and should
      # work for nearly all cases.
      return unless grep {
            sprintf("%7.1f", $_) != 0
         } @{$stat}{ @$cols };
   }
   printf { $self->out_fh() } $format . "\n",
           @{$stat}{ qw( line_ts dev ), @$cols };
}

sub print_deltas {
   my ( $self, %args ) = @_;

   my ( $header, $format, $cols ) = $self->design_print_formats(
      # Not required args, because design_print_formats picks sane defaults.
      max_device_length => $args{max_device_length},
      columns           => $args{columns},
   );

   return unless $self->delta_against_ts();

   @$cols = map { $self->_column_to_key($_) } @$cols;
   my ( $header_callback, $rows_callback ) = @args{qw( header_callback rows_callback )};

   if ( $header_callback ) {
      $self->$header_callback( $header, "#ts", "device" );
   }
   else {
      $self->print_header( $header, "#ts", "device" );
   }

   for my $stat ( $self->_calc_deltas() ) {
      if ($rows_callback) {
         $self->$rows_callback( $format, $cols, $stat );
      }
      else {
         $self->print_rows( $format, $cols, $stat );
      }
   }
}

sub compute_line_ts {
   my ( $self, %args ) = @_;
   return sprintf( "%5.1f", $args{first_ts} > 0
                            ? $args{curr_ts} - $args{first_ts}
                            : 0 );
}

sub compute_in_progress {
   my ( $self, $in_progress, $tot_in_progress ) = @_;
   return $in_progress;
}

sub compute_devs_in_group {
   return 1;
}

sub delta_against {
   die 'You must override delta_against() in a subclass';
}

sub delta_against_ts {
   die 'You must override delta_against_ts() in a subclass';
}

sub group_by {
   die 'You must override group_by() in a subclass';
}

sub _d {
   my ($package, undef, $line) = caller 0;
   @_ = map { (my $temp = $_) =~ s/\n/\n# /g; $temp; }
        map { defined $_ ? $_ : 'undef' }
        @_;
   print STDERR "# $package:$line $PID ", join(' ', @_), "\n";
}

1;

}
# ###########################################################################
# End Diskstats package
# ###########################################################################