Replace the switches with direct connections

Contributors netapp-pcarriga

If you have a two-node cluster that uses cluster interconnect switches, replace the switches with direct, back-to-back connections between the nodes. This is a nondisruptive operation.

Before you begin
  • The cluster must be healthy and consist of two nodes connected by cluster switches, and the nodes must be running the same ONTAP release.

  • Each node must have the required number of dedicated cluster ports that provide redundant cluster interconnect connections to support your system configuration, for example, two redundant ports for a system with two dedicated cluster interconnect ports on each node.

About this task

The following procedure removes the cluster switches in a two-node cluster and replaces each connection to the switch with a direct connection to the partner node.

Cluster switches replaced with direct connections

The examples in the following procedure show nodes that are using "e0a" and "e0b" as cluster ports. Your nodes might be using different cluster ports as they vary by system.

Steps
  1. Change the privilege level to advanced, entering y when prompted to continue:

    set -privilege advanced

    The advanced prompt *> appears.

  2. ONTAP 9.3 and later supports automatic detection of switchless clusters, which is enabled by default.

    You can verify that detection of switchless clusters is enabled by running the advanced privilege command:

    network options detect-switchless-cluster show

    The following example output shows if the option is enabled.

    cluster::*> network options detect-switchless-cluster show
       (network options detect-switchless-cluster show)
    Enable Switchless Cluster Detection: true

    If "Enable Switchless Cluster Detection” is false, contact support.

  3. If AutoSupport is enabled on this cluster, suppress automatic case creation by invoking an AutoSupport message:

    system node autosupport invoke -node * -type all -message MAINT=<number_of_hours>h

    where h is the duration of the maintenance window in hours. The message notifies technical support of this maintenance task so that they can suppress automatic case creation during the maintenance window.

    In the following example, the command suppresses automatic case creation for two hours:

    cluster::*> system node autosupport invoke -node * -type all -message MAINT=2h
  4. Organize the cluster ports on each switch into groups so that the cluster ports in group1 go to cluster switch1 and the cluster ports in group2 go to cluster switch2. These groups are required later in the procedure.

  5. Identify the cluster ports and verify link status and health:

    network port show -ipspace Cluster

    In the following example for nodes with cluster ports "e0a" and "e0b", one group is identified as “node1:e0a” and “node2:e0a” and the other group as “node1:e0b” and “node2:e0b”. Your nodes might be using different cluster ports because they vary by system.

    Cluster switch connections between node1 and node2

    Verify that the ports have a value of up for the “Link” column and a value of healthy for the “Health Status” column, as shown in the following example:

    cluster::> network port show -ipspace Cluster
    Node: node1
                                                                     Ignore
                                                 Speed(Mbps) Health  Health
    Port  IPspace   Broadcast Domain Link  MTU   Admin/Oper	 Status  Status
    ----- --------- ---------------- ----- ----- ----------- ------- -------
    e0a   Cluster   Cluster          up    9000  auto/10000  healthy false
    e0b   Cluster   Cluster          up    9000  auto/10000  healthy false
    
    Node: node2
                                                                     Ignore
                                                 Speed(Mbps) Health  Health
    Port  IPspace   Broadcast Domain Link  MTU   Admin/Oper	 Status  Status
    ----- --------- ---------------- ----- ----- ----------- ------- -------
    e0a   Cluster   Cluster          up    9000  auto/10000  healthy false
    e0b   Cluster   Cluster          up    9000  auto/10000  healthy false
    4 entries were displayed.
  6. Confirm that all the cluster LIFs are on their home ports.

    Verify that the “is-home” column is true for each of the cluster LIFs:

    network interface show -vserver Cluster -fields is-home

    The system displays output similar to the following example:

    cluster::*> net int show -vserver Cluster -fields is-home
    (network interface show)
    vserver  lif          is-home
    -------- ------------ --------
    Cluster  node1_clus1  true
    Cluster  node1_clus2  true
    Cluster  node2_clus1  true
    Cluster  node2_clus2  true
    4 entries were displayed.

    If there are cluster LIFs that are not on their home ports, revert those LIFs to their home ports:

    network interface revert -vserver Cluster -lif *

  7. Disable auto-revert for the cluster LIFs:

    network interface modify -vserver Cluster -lif * -auto-revert false

  8. Verify that all ports listed in the previous step are connected to a network switch:

    network device-discovery show -port cluster_port

    The “Discovered Device” column should be the name of the cluster switch that the port is connected to. The following example shows that cluster ports "e0a" and "e0b" are correctly connected to cluster switches "cs1" and "cs2".

    cluster::> network device-discovery show -port e0a|e0b
      (network device-discovery show)
    Node/     Local  Discovered
    Protocol  Port   Device (LLDP: ChassisID)  Interface  Platform
    --------- ------ ------------------------- ---------- ----------
    node1/cdp
              e0a    cs1                       0/11       BES-53248
              e0b    cs2                       0/12       BES-53248
    node2/cdp
              e0a    cs1                       0/9        BES-53248
              e0b    cs2                       0/9        BES-53248
    4 entries were displayed.
  9. Verify the cluster connectivity:

    cluster ping-cluster -node local

  10. Verify that the cluster is healthy:

    cluster ring show

    All units must be either master or secondary.

  11. Set up the switchless configuration for the ports in group 1.

    Important To avoid potential networking issues, you must disconnect the ports from group1 and reconnect them back-to-back as quickly as possible, for example, in less than 20 seconds.
    1. Disconnect all the cables from the ports in group1 at the same time.

      In the following example, the cables are disconnected from port "e0a" on each node, and cluster traffic continues through the switch and port "e0b" on each node:

      ClusterSwitch1 disconnected
    2. Cable the ports in group1 back-to-back.

      In the following example, "e0a" on node1 is connected to "e0a" on node2:

      Direct connection between ports "e0a"
  12. The switchless cluster network option transitions from false to true. This might take up to 45 seconds. Confirm that the switchless option is set to true:

    network options switchless-cluster show

    The following example shows that the switchless cluster is enabled:

    cluster::*> network options switchless-cluster show
    Enable Switchless Cluster: true
  13. Verify that the cluster network is not disrupted:

    cluster ping-cluster -node local

  14. Set up the switchless configuration for the ports in group 2.

    Important To avoid potential networking issues, you must disconnect the ports from group2 and reconnect them back-to-back as quickly as possible, for example, in less than 20 seconds.
    1. Disconnect all the cables from the ports in group2 at the same time.

      In the following example, the cables are disconnected from port "e0b" on each node, and cluster traffic continues through the direct connection between the "e0a" ports:

      ClusterSwitch2 disconnected
    2. Cable the ports in group2 back-to-back.

      In the following example, "e0a" on node1 is connected to "e0a" on node2 and "e0b" on node1 is connected to "e0b" on node2:

      Direct connection between ports on node1 and node2
  15. Verify that the ports on both nodes are correctly connected:

    network device-discovery show -port cluster_port

    The following example shows that cluster ports "e0a" and "e0b" are correctly connected to the corresponding port on the cluster partner:

    cluster::> net device-discovery show -port e0a|e0b
      (network device-discovery show)
    Node/      Local  Discovered
    Protocol   Port   Device (LLDP: ChassisID)  Interface  Platform
    ---------- ------ ------------------------- ---------- ----------
    node1/cdp
               e0a    node2                     e0a        AFF-A300
               e0b    node2                     e0b        AFF-A300
    node1/lldp
               e0a    node2 (00:a0:98:da:16:44) e0a        -
               e0b    node2 (00:a0:98:da:16:44) e0b        -
    node2/cdp
               e0a    node1                     e0a        AFF-A300
               e0b    node1                     e0b        AFF-A300
    node2/lldp
               e0a    node1 (00:a0:98:da:87:49) e0a        -
               e0b    node1 (00:a0:98:da:87:49) e0b        -
    8 entries were displayed.
  16. Reenable auto-revert for the cluster LIFs:

    network interface modify -vserver Cluster -lif * -auto-revert true

  17. Verify that all LIFs are home. This might take a few seconds:

    network interface show -vserver Cluster -lif lif_name

    The LIFs have been reverted if the “Is Home” column is true, as shown for node1_clus2 and node2_clus2 in the following example:

    cluster::> network interface show -vserver Cluster -fields curr-port,is-home
    vserver  lif           curr-port is-home
    -------- ------------- --------- -------
    Cluster  node1_clus1   e0a       true
    Cluster  node1_clus2   e0b       true
    Cluster  node2_clus1   e0a       true
    Cluster  node2_clus2   e0b       true
    4 entries were displayed.

    If any cluster LIFS have not returned to their home ports, revert them manually:

    network interface revert -vserver Cluster -lif lif_name

  18. Check the cluster status of the nodes from the system console of either node:

    cluster show

    The following example shows epsilon on both nodes to be false:

    Node  Health  Eligibility Epsilon
    ----- ------- ----------- --------
    node1 true    true        false
    node2 true    true        false
    2 entries were displayed.
  19. Confirm connectivity between the cluster ports:

    cluster ping-cluster local

  20. If you suppressed automatic case creation, reenable it by invoking an AutoSupport message:

    system node autosupport invoke -node * -type all -message MAINT=END

  21. Change the privilege level back to admin:

    set -privilege admin

Related information