fuel-library/deployment/puppet/l23network

L23network

Puppet module for configuring network interfaces on 2nd and 3rd level (802.1q vlans, access ports, NIC-bondind, assign IP addresses, dhcp, and interfaces without IP addresses). Can work together with Open vSwitch or standart linux way. At this moment support Centos 6.3+ (RHEL6) and Ubuntu 12.04 or above.

L23network module have a same behavior for both operation systems.

WARNING!!! This is a L23network v1.1, it contains some incompatibles with earlier versions. Be carefully.

Usage

Initializing

Place this module at /etc/puppet/modules/l23network or another directory with your puppet modules.

Include L23network module and initialize it. It is recommended to do it on the early stage:

#Network configuration
stage {'netconfig':
  before  => Stage['main'],
}
class { 'l23network':
  use_ovs => true,
  use_lnx => true,
  stage   => 'netconfig'
}

Initialization class 'l23network' has following incoming parameters and its default values:

class { 'l23network':
  use_ovs                   => false,
  use_lnx                   => true,
  install_ovs               => $use_ovs,
  install_brtool            => $use_lnx,
  install_ethtool           => $use_lnx,
  install_bondtool          => $use_lnx,
  install_vlantool          => $use_lnx,
  ovs_modname               => undef,
  ovs_datapath_package_name => undef,
  ovs_common_package_name   => undef,
}

For highly customized configurations you can redefine each of ones. For example, if you plan to use open vSwitch you should enable it:

class {'l23network': 
  use_ovs=>true
}

L2 network features configuation

Current layout is:

• bridges -- A "Bridge" is a virtual ethernet L2 switch. You can plug ports into it.
• ports -- A Port is an interface you plug into the bridge. It may be virtual or native interface.

Then in your manifest you can either use it as a parameterized classes:

l23network::l2::bridge{"br-mgmt": }
l23network::l2::port{"eth0": bridge => "br-mgmt"}

l23network::l2::bridge{"br-ex": provider => ovs }
l23network::l2::port{"eth1": bridge => "br-ex" }
l23network::l2::port{"ve0": bridge => "br-ex" }
l23network::l2::port{"ve1": bridge => "br-ex" }

L2::Bridge

This resource implemented for configire of bridge.

l23network::l2::bridge { 'br1':
  ensure          => present,
  stp             => true,  # or false
  vendor_specific => {
    .....
  },
  provider        => lnx,
}

Non-obligatory fields:

• stp -- enable/disable STP for bridge
• bpdu_forward -- enable/disable BPDU forward on bridge
• bridge_id -- bridge_id for STP protocol.
• vendor_specific -- vendor_specific hash (see below)
• delay_while_up -- delay, in seconds, which will happens each time while node will boot after interface up.

L2::Port

Resource for configuring port L2 options. Only L2 options. For configuring L3 options -- use L23network::l3::ifconfig resource

l23network::l2::port { 'eth1':
  mtu     => 9000,   # MTU value, unchanged if absent.
  onboot  => true,   # whether port has UP state after setup or node boot
  ethtool => {
    .....
  },
  vendor_specific => {
    .....
  },
  provider => lnx
}

l23network::l2::port { 'eth1.101':
  ensure    => present,
  bridge    => 'br1',  # port can be a member of bridge. 
                       # If no value given this property was unchanged, 
                       # if given 'absent' port will be excluded from any
                       # bridges.
  onboot    => true,
  delay_while_up => 10
  provider  => lnx
}

Alternative VLAN definition (not recommended for 'lnx' provider)

l23network::l2::port { 'vlan77':
  vlan_id   => 77,
  vlan_dev  => eth1,
  provider  => lnx
}

L2::Bond

It's a special type of port. Designed for bonding two or more interfaces. Detail description of bonding feature you can read here: https://www.kernel.org/doc/Documentation/networking/bonding.txt If you plan use LACP -- we highly recommend do not use OVS. Also we don't recommend insert native linux bonds to OVS bridges. This case works, but leads many heavy diagnostic surprises.

l23network::l2::bond { 'bond0':
  interfaces      => ['eth1', 'eth2'],
  bridge          => 'br0',  # obligatory only for OVS provider
  mtu             => 9000,
  onboot          => true,
  bond_properties => {  # bond configuration properties (see bonding.txt)
    mode             => '803.1ad',
    lacp_rate        => 'slow',
    xmit_hash_policy => 'encap3+4'
  },
  interface_properties => {  # config properties for included ifaces
    ethtool => {
      .....
    },
  },
  vendor_specific => {
    .....
  },
  delay_while_up => 45
  provider => lnx,
}

Bond mode and xmit_hash_policy configuration has some differences for lnx and ovs providers:

For lnx provider mode can be:

• balance-rr (default)
• active-backup
• balance-xor
• broadcast
• 802.3ad
• balance-tlb
• balance-alb

For 802.3ad (LACP), balance-xor, balance-tlb and balance-alb cases should be defined xmit_hash_policy as one of:

• layer2 (default)
• layer2+3
• layer3+4
• encap2+3
• encap3+4

For ovs provider mode can be:

• active-backup
• balance-slb (default)
• balance-tcp

Field xmit_hash_policy shouldn't use for any mode. For balance-tcp mode lacp bond-property should be set to 'active' or 'passive' value.

While bond will created also will created ports, included to the bond. This ports will be created as slave ports for this bond with properties, listed in interface_properties field. If you want more flexibility, you can create this ports by l23network::l2::port resource and shouldn't define interface_properties field.

MTU field, setted for bond interface will be passed to interfaces, included to the bond automatically.

I recommend use delay_while_up property, while configure LACP bonds, because such bonds may take some time for settle.

For some providers (ex: ovs) bridge field is obligatory.

L2::Patch

It's a patchcord for connecting two bridges. Architecture limitation: two bridges may be connected only by one patchcord. Name for patchcord interfaces calculated automatically and can't changed in configuration.

OVS provider can connect OVS-to-OVS and OVS-to-LNX bridges. If you connect OVS-to-LNX bridges, you SHOULD put OVS bridge first in order.

l23network::l2::patch { 'patch__br0--br1':
  bridges => ['br0','br1'],
  vendor_specific => {
    .....
  },
}

Naming conviency

Each low-level puppet patchcord resource l2_patch has his name in 'bridge__%bridge1%--%bridge2%' format, and bridges provided in alphabetical order for all providers. This resource also contain 'bridges' property. It's a array of two bridge names. Order of names depends of provider implementation. For example, 'ovs' provider bridge names listed in alphabetical order for OVS-to-OVS connectivity, and ovs-bridge always first for OVS-to-LNX bridges connectivity.

Each L2_patch instance contains read-only 'jacks' property. It's a array of two names of jacks, 'inserted' to each bridge. This property has the same ordering style, that a 'bridges' property for this provider.

If patchcord connect two bridges different nature, the 'cross' flag will be setting to 'true'.

Ethtool hash and offloading settings

You can manage offloading and another options, controlled by ethtool utility, for any resources, that has ethtool hash as one of incoming properties. Ethtool field look like hash of hashes. Keys of the external hash -- are a section names from ethtool manual. Ones maps to an internal hashes. Internal hashes -- is a option to value mappings. Option names corresponds to ethtool output option naming. For example, you can see list of offloading options by executing 'ethtool -k eth0'. Ethtool options are pre-defined and stateful. All implemented sections and options you can see bellow:

ethtool => {
  offload => {
      rx-checksumming              => true or false,
      tx-checksumming              => true or false,
      scatter-gather               => true or false,
      tcp-segmentation-offload     => true or false,
      udp-fragmentation-offload    => true or false,
      generic-segmentation-offload => true or false,
      generic-receive-offload      => true or false,
      large-receive-offload        => true or false,
      rx-vlan-offload              => true or false,
      tx-vlan-offload              => true or false,
      ntuple-filters               => true or false,
      receive-hashing              => true or false,
      rx-fcs                       => true or false,
      rx-all                       => true or false,
      highdma                      => true or false,
      rx-vlan-filter               => true or false,
      fcoe-mtu                     => true or false,
      l2-fwd-offload               => true or false,
      loopback                     => true or false,
      tx-nocache-copy              => true or false,
      tx-gso-robust                => true or false,
      tx-fcoe-segmentation         => true or false,
      tx-gre-segmentation          => true or false,
      tx-ipip-segmentation         => true or false,
      tx-sit-segmentation          => true or false,
      tx-udp_tnl-segmentation      => true or false,
      tx-mpls-segmentation         => true or false,
      tx-vlan-stag-hw-insert       => true or false,
      rx-vlan-stag-hw-parse        => true or false,
      rx-vlan-stag-filter          => true or false,
  },
  #settings => {
  #   duplex => 'half',
  #   mdix   => off
  #}
}

L3 network configuration

L3::Ifconfig

Resource for configuring IP addresses on interface. Only L3 options. For configuring L2 options -- use corresponded L2 resource.

l23network::l3::ifconfig { 'eth1.101':
  ensure           => present,
  ipaddr           => ['192.168.10.3/24', '10.20.30.40/25'],
  gateway          => '192.168.10.1',
  gateway_metric   => 10,  # different Ifconfig resources should not has
                           # gateways with same metrics
  vendor_specific => {
    .....
  },
}

DHCP or address-less interfaces

l23network::l3::ifconfig {"eth2": ipaddr=>'dhcp'}
l23network::l3::ifconfig {"eth3": ipaddr=>'none'}

Option ipaddr can contains array of IP addresses (even setup one ipaddr), 'dhcp', or 'none' string.

CIDR-notated form of IP address is required.

Default gateway

l23network::l3::ifconfig {"eth1":
  ipaddr         => ['192.168.2.5/24'],
  gateway        => '192.168.2.1',
  gateway_metric => 10,
}

if gateway_metric omited, gateway will be setup without metric definition.

Network Scheme

Network scheme is a hierarchical-based manner for define network topology for host. In following examples I use yaml format for represent it. Main idea:

• when we got undeployed server we have some number of NICs. NICs, managed by puppet should be listed in interfaces section. It is giveg.
• The result of our network configuration process is a some network topology on the host and some interfaces with assignet IP addresses (or without IPs). It's a endpoints.
• Interfaces become endpoints by successive transformations. I try explain how it works in the following document: Transformations. How they work.

Example of typical network scheme:

---
network_scheme:
  version: 1.1
  provider: lnx
  interfaces:
    eth1:
      mtu: 7777
    eth2:
      mtu: 9000
  transformations:
    - action: add-br
      name: br1
    - action: add-port
      name: eth1
      bridge: br1
    - action: add-br
      name: br-mgmt
    - action: add-port
      name: eth1.101
      bridge: br-mgmt
    - action: add-br
      name: br-ex
    - action: add-port
      name: eth1.102
      bridge: br-ex
    - action: add-br
      name: br-storage
    - action: add-port
      name: eth1.103
      bridge: br-storage
    - action: add-br
      name: br-prv
      provider: ovs
    - action: add-port
      name: eth2
      bridge: br-prv
      provider: ovs
  endpoints:
    br-mgmt:
      IP:
        - 192.168.101.3/24
      gateway: 192.168.101.1
      gateway-metric: 100
      routes:
        - net: 192.168.210.0/24
          via: 192.168.101.1
        - net: 192.168.211.0/24
          via: 192.168.101.1
        - net: 192.168.212.0/24
          via: 192.168.101.1
    br-ex:
      gateway: 192.168.102.1
      IP:
        - 192.168.102.3/24
    br-storage:
      IP:
        - 192.168.103.3/24
    br-prv:
      IP: none
  roles:
    management: br-mgmt
    ceph: br-mgmt
    private: br-prv
    fw-admin: br1
    ex: br-ex
    floating: br-ex
    storage: br-storage

Example of typical network scheme with bonds and disabling offloads:

---
network_scheme:
  version: "1.1"
  provider: lnx
  interfaces:
    eth1:
      mtu: 9000
    eth2:
    eth3:
  transformations:
    - action: add-br
      name: br1
    - action: add-port
      bridge: br1
      name: eth1
      ethtool:
        offload:
          tcp-segmentation-offload: off
          udp-fragmentation-offload: off
          generic-segmentation-offload: off
          generic-receive-offload: off
          large-receive-offload: off
    - action: add-br
      name: br2
    - action: add-bond
      name: bond23
      bridge: br2
      interfaces:
        - eth2
        - eth3
      mtu: 9000
      interface_properties:
        ethtool:
          offload:
            tcp-segmentation-offload: off
            udp-fragmentation-offload: off
      bond_properties:
        mode: balance-rr
        xmit_hash_policy: encap3+4
        updelay: 10
        downdelay: 40
        use_carrier: 0
    - action: add-br
      name: br-mgmt
    - action: add-port
      name: bond23.101
      bridge: br-mgmt
    - action: add-br
      name: br-ex
    - action: add-port
      name: bond23.102
      bridge: br-ex
    - action: add-br
      name: br-storage
    - action: add-port
      name: bond23.103
      bridge: br-storage
  endpoints:
    br-mgmt:
      IP:
        - 192.168.101.3/24
      gateway: 192.168.101.1
      gateway-metric: 100
    br-ex:
      gateway: 192.168.102.1
      IP:
        - 192.168.102.3/24
    br-storage:
      IP:
        - 192.168.103.3/24
  roles:
    fw-admin: br1
    ex: br-ex
    management: br-mgmt
    storage: br-storage

Vendor_specific hash

Vendor_specific field - is a hash, empty by default, required only for plug-ins. It allows plugin developers not to change custom type code for adding non-standart parameters. Due to inheriting and extending puppet type (not the provider one), is a non-trivial task. Plugin developers may pass any data structures by this hash and its subhashes. All data from this hash pass to the provider transparently.

Debugging

For debug purpose you can use following puppet calls for get prefetchable properties for existing resources. Please note, that bridges and bonds in linux are a port too, and present in l2_port output with corresponded flags (if_type).

# puppet resource -vd --trace l23_stored_config
# puppet resource -vd --trace l2_port
# puppet resource -vd --trace l2_bridge
# puppet resource -vd --trace l2_bond
# puppet resource -vd --trace l3_ifconfig
# puppet resource -vd --trace l3_route

This commands may be fail before 1st configuration networking by L23network because some kernel modules may wasn't loaded or some command-line tools wasn't installed.

Internals

Each L23network resource has interface trought puppet 'define' resource. This define may conains some non difficult logic, define provider for low-level resources and call two low level resources:

• l23_stored_config -- for modifying OS config files
• low level resource for configuring it in runtime (e.x: l2_bridge)

L23_stored_config custom type

This resource is implemented to manage interface config files. Each possible parameter should be described in resource type.

This resource allows to forget about ERB templates, because in some cases (i.e. bridge + port with same name + ip address for this port) we should modify config file content three times.

l23_stored_config { 'br1':
  onboot   => true,
  method   => manual,
  mtu      => 1500,
  ethtool => {
    .....
  },
  provider => lnx_ubuntu
}

Place of config files location defined inside provider for corresponded operation system and provider. Provider name for l23_stored_config depends from operation system (may be with version) and network provider (native linux, ovs, etc...)

References

• Transformations. How they work.

---

When I started working on this module I was inspired by https://github.com/ekarlso/puppet-vswitch. Endre, big thanks...