gistfile1.pp

# == Class riak::tuning
#
# This class performs system performance tuning
# This is based on @Cornellio's work from
# https://github.com/Cornellio/puppet-riak/blob/master/manifests/baseconfig.pp
# We're trying to implement the recommendations in
# http://docs.basho.com/riak/1.3.2/cookbooks/Linux-Performance-Tuning/
# if we're on AWS, we should implement the recommendations in
# http://docs.basho.com/riak/1.3.2/cookbooks/Performance-Tuning-AWS/
# (specifically, use the deadline scheduler on EBS volumes)

# We want to:
# - increase open files limit - http://docs.basho.com/riak/1.3.2/cookbooks/Open-Files-Limit/
#
# - set noatime,data=writeback,barrier=0 on the volume we've mounted riak data to
# - sysctl.conf kernel settings, mostly TCP settings
# vm.swappiness = 0
# net.ipv4.tcp_max_syn_backlog = 40000
# net.core.somaxconn=4000
# net.ipv4.tcp_timestamps = 0
# net.ipv4.tcp_sack = 1
# net.ipv4.tcp_window_scaling = 1
# net.ipv4.tcp_fin_timeout = 15
# net.ipv4.tcp_keepalive_intvl = 30
# net.ipv4.tcp_tw_reuse = 1
# if we have a 10Gb network, we want:
# net.core.rmem_default = 8388608
# net.core.rmem_max = 8388608
# net.core.wmem_default = 8388608
# net.core.wmem_max = 8388608
# net.core.netdev_max_backlog = 10000

# - ulimit settings
# - disable swap entirely
class riak::tuning {
  case $::osfamily {
    'Debian', 'RedHat', 'Amazon': {


      # set kernel io scheduler
      # http://docs.basho.com/riak/latest/ops/tuning/linux/
      #   Scheduler recommendations:

      #   The noop scheduler when deploying on iSCSI over HBAs, or any hardware-based RAID.
      #   The deadline scheduler when using SSD-based storage.
      # by default, set to deadline
      # if EBS, set to deadline
      # if SSD, set to noop

      # increase the scheduler depth to 1024
      # echo 1024 > /sys/block/sda/queue/nr_requests
      # how about setting it permanently?

      # filesystem options
      # use ext4 on linux
      # use ZFS on BSD or Solaris
      mount { '/':
        options => 'defaults,noatime,barrier=0,data=writeback',
      }

      # Optional I/O Settings
      # If your cluster is experiencing excessive I/O blocking, the following settings may help prevent disks from being overwhelmed during periods of high write activity at the expense of peak performance for spiky workloads:

      # vm.dirty_background_ratio = 0
      # vm.dirty_background_bytes = 209715200
      # vm.dirty_ratio = 40
      # vm.dirty_bytes = 0
      # vm.dirty_writeback_centisecs = 100
      # vm.dirty_expire_centisecs = 200

      # These settings have been tested and benchmarked by Basho in nodes with 16 GB of RAM.



      mount { '/':
        ensure  => 'mounted',
        dump    => '1',
        fstype  => 'ext4',
        options => 'noatime,data=writeback,barrier=0',
        pass    => '1',
        target  => '/etc/fstab',
      }

      augeas { 'kernel-params':
        context => '/files/etc/sysctl.conf',
        changes => [
          "set fs.file-max[1] ${riak::sysctl_fs_file_max}",
          'set vm.swappiness[1] 0',
          'set net.ipv4.tcp_max_syn_backlog[1] 40000',
          'set net.core.somaxconn[1] 4000',
          'set net.ipv4.tcp_timestamps[1] 0',
          'set net.ipv4.tcp_sack[1] 1',
          'set net.ipv4.tcp_window_scaling[1] 1',
          'set net.ipv4.tcp_fin_timeout[1] 15',
          'set net.ipv4.tcp_keepalive_intvl[1] 30',
          'set net.ipv4.tcp_tw_reuse[1] 1',
          'set net.core.rmem_default[1] 8388608',
          'set net.core.rmem_max[1] 8388608',
          'set net.core.wmem_default[1] 8388608',
          'set net.core.wmem_max[1] 8388608',
          'set net.core.netdev_max_backlog[1] 10000',
        ],
        before  => Exec['load-kernel-params'],
      }
    }
}