path: root/sw_config/bmra/patched_preflight.yml

# SPDX-FileCopyrightText: 2021 Intel Corporation.
#
# SPDX-License-Identifier: Apache-2.0

---
# Preflight: ALL checks must PASS
# Only assert issues (do NOT change anything)
#
# Manual run: 'ansible-playbook -i inventory.ini playbooks/preflight.yml --flush-cache'
#
# Summary:
# On Ansible Host (localhost):
# - Check Ansible version (match)
# - Check Python version (min)
# - Check Group Vars (exist)
# - Check CMK Hosts (valid targets)
# On All targets (k8s-cluster):
# - Check Linux Distro
# - Check Hostnames (match Inventory)
# - Check CMK Config (isolcpus defined)
# - Check isolcpus Total (not more than actual)
# - Check isolcpus IDs (valid on system)
# - Check isolcpus OS Reserved (not 0,1,etc)
# On Worker Nodes Only (kube-node):
# - Check DP Interfaces (is not empty)
# - Check DP Interfaces Name (optional)
# - Check DP Interfaces Bus Info (pciid)
# - Check QAT Devices Bus Info (pciid)
# - Check QAT SRIOV VFs (max)
# - Check SGX configuration
# - Check OVS DPDK Dependencies (for 1G Hugepages)
# - Check VPP Dependencies (for 2M Hugepages)
# - Check CNI Dependencies (for OVS DPDK or VPP and Hugepages)
# - Check SST (not on RHEL 8.2 or old OSs)
# - Warn BIOS VT-d (should be enabled)
# - Warn BIOS Hyper-Threading (should be enabled)
# - Warn collectd (kernel update needed on old OSs)
# - Check OVS DPDK Version Compatability (for OVS support)

# additional vars required:
#  bmra_supported_ansible: # must be version
#  bmra_supported_python: # min version
#  bmra_supported_distros: [] # list
#  bmra_supported_distros_versions: [] # list


##################################
# Prerequisites for Ansible Host #
##################################
- hosts: 127.0.0.1
  connection: local
  vars:
    bmra_supported_ansible: 2.9.20
    bmra_supported_python: 2.7

  tasks:

    - debug: msg="Ansible version is {{ ansible_version.string }}"
    - name: Check Ansible Version
      assert:
        that: (ansible_version.full is version_compare(bmra_supported_ansible, '=='))
        msg: "Ansible version must be {{ bmra_supported_ansible }}. Please update"

    - debug: msg="Python version is {{ ansible_python_version }}"
    - name: Check Python Version
      assert:
        that: (ansible_python_version is version_compare(bmra_supported_python, '>='))
        msg: "Python version must be at least {{ bmra_supported_python }}. Please update"

    - name: Read Group Vars
      stat:
        path: "{{ inventory_dir }}/group_vars/"
      register: group_vars_details

    - name: Check Group Vars
      assert:
        that: "group_vars_details.stat.exists and group_vars_details.stat.isdir"
        msg: "File group_vars/all.yml does NOT exist. Must be created per Guide"

    - debug:
        msg:
          - cmk_enabled = {{ cmk_enabled }} (group_vars/all.yml)
          - cmk_use_all_hosts = {{ cmk_use_all_hosts }} (group_vars/all.yml)
          - cmk_hosts_list = {{ cmk_hosts_list | default('') }} (group_vars/all.yml)
          - all targets = {{ groups['all'] }} (inventory.ini)
      when: cmk_enabled is defined  # CMK expected true for all profiles except basic

    - name: Check Intel CMK Hosts
      assert:
        that: "item in groups['all']"
        msg: "Hostname '{{ item }}' is NOT a valid target from inventory.  Please correct the cmk_hosts_list or disable the CMK feature in group vars"
      with_items: "{{ (cmk_hosts_list.split(',') if (cmk_hosts_list is defined and cmk_hosts_list | length > 0) else []) }}"
      when: cmk_enabled is defined and cmk_enabled and not cmk_use_all_hosts


##############################################
# Prerequisites for Control and Worker Nodes #
##############################################
- hosts: k8s-cluster
  vars:
    bmra_supported_distros: [CentOS, RedHat, Ubuntu]
    bmra_supported_distros_versions: ['7.6', '7.8', '7.9', '8.2', '8.3', '8.4', '18.04', '20.04', '21.04']
    isolcpus_ranges: []
    isolcpus_discretes: []

  tasks:

    - debug: msg="Linux distribution on target is {{ ansible_distribution }} {{ ansible_distribution_version }} {{ ansible_distribution_release }}"
    - name: Check Linux Distro and Version
      assert:
        that: "ansible_distribution in bmra_supported_distros and ansible_distribution_version in bmra_supported_distros_versions"
        msg:
          - Linux distribution {{ ansible_distribution }} {{ ansible_distribution_version }} on target '{{ inventory_hostname }}' is NOT supported
          - Must be one of {{ bmra_supported_distros }} and version {{ bmra_supported_distros_versions }}

#    - name: Check Linux Across Cluster
# TODO ?? Linux OS must be the same on all targets (no mix-n-match)

    - name: regather network facts in case hostname recently changed
      setup:
        gather_subset: network
    - debug: msg="Inventory target '{{ inventory_hostname }}' has the actual system hostname '{{ ansible_hostname }}'"
    - name: Check Inventory Hostnames
      debug:
        msg:
          - "Target '{{ inventory_hostname }}' in inventory does NOT match the actual system hostname '{{ ansible_hostname }}'."
          - "If it's done intentionally, please ignore this message."
      when:
        - inventory_hostname != ansible_hostname

# Early check if SELinux is configured properly
    - block:
      - name: "Collect packages facts"
        package_facts:
      - debug:
          msg:
            - "Current SELinux status:"
            - "status: {{ ansible_selinux.status | default('') }}"
            - "policy version:  {{ ansible_selinux.policyvers | default('') }}"
            - "type: {{ ansible_selinux.type | default('') }}"
            - "mode: {{ ansible_selinux.mode | default('') }}"
            - "config_mode: {{ ansible_selinux.config_mode | default('') }}"

      - name: check selinux condition possibly causing system boot failure
        debug:
          msg:
            - "Current SELinux setup might cause the system possibly will not boot up on next reboot."
            - "Please, check SELinux settings and set it up according to the documentation."
        when:
          - "'selinux-policy' not in ansible_facts.packages"
          - "'selinux-policy-targeted' not in ansible_facts.packages"
      when:
        - ansible_os_family == "RedHat"

# STORY: "cmk requires isolcpus to be configured"
    - block:
      - debug:
          msg:
            - cmk_enabled = {{ cmk_enabled }} (group_vars/all.yml)
            - cmk_use_all_hosts = {{ cmk_use_all_hosts }} (group_vars/all.yml)
            - cmk_hosts_list = {{ cmk_hosts_list | default('') }} (group_vars/all.yml)
            - cmk_shared_num_cores = {{ cmk_shared_num_cores }} (group_vars/all.yml)
            - cmk_exclusive_num_cores = {{ cmk_exclusive_num_cores }} (group_vars/all.yml)
            - isolcpus_enabled = {{ isolcpus_enabled }} (host_vars)
            - isolcpus = {{ isolcpus }} (host_vars)
            - ansible_processor_count = {{ ansible_processor_count }}
            - ansible_processor_cores = {{ ansible_processor_cores }}
            - ansible_processor_threads_per_core = {{ ansible_processor_threads_per_core }}
            - ansible_processor_vcpus = {{ ansible_processor_vcpus }}
            - CPUs Reserved for OS = 0...{{ ansible_processor_count - 1 }}
#            - CPUs Reserved for OS = {{ lookup('sequence','0-{{ ansible_processor_count - 1 }}').split(',') }} # [E207] Nested jinja pattern

      - name: Check Intel CMK Config
        assert:
          that: ({{ cmk_enabled }} and {{ isolcpus_enabled }} and "{{ isolcpus }}" | length > 0)
          msg:
            - Incorrect configuration pertaining Intel CMK. Conflicting or improper values detected
            - When Intel CMK is enabled, CPUs isolation ('isolcpus') must be set according to the example file for host_vars. Please correct the configuration

      - name: Split isolcpus Groups
        set_fact:
          isolcpus_groups: "{{ isolcpus.split(',') }}"

      - debug: msg="isolcpus_groups = {{ isolcpus_groups }}"

      - name: Filter isolcpus Ranges
        set_fact:
          isolcpus_ranges: "{{ isolcpus_ranges + [item] }}"
        with_items: "{{ isolcpus_groups }}"
        when: ("-" in item)

      - debug: msg="isolcpus_ranges = {{ isolcpus_ranges }}"

      - name: Filter isolcpus Discretes
        set_fact:
          isolcpus_discretes: "{{ isolcpus_discretes + [item] }}"
        with_items: "{{ isolcpus_groups }}"
        when: ("-" not in item)

      - debug: msg="isolcpus_discretes = {{ isolcpus_discretes }}"

      - name: Build isolcpus List
        set_fact:
          isolcpus_list: "{{ isolcpus_list | default([]) | union(isolcpus_discretes) | union([item]) }}"
        with_sequence: "{{ isolcpus_ranges }}"

      - debug: msg="isolcpus_list = {{ isolcpus_list }}"

      - name: Check isolcpus Total
        assert:
          that: "{{ isolcpus_list | length }} <= ansible_processor_vcpus"
          msg:
            - Incorrect configuration pertaining isolcpus. Conflicting or improper values detected
            - The number of isolcpus {{ isolcpus_list | length }}, exceeds total CPUs on target {{ ansible_processor_vcpus }}. Please correct the configuration
        when: isolcpus is defined

      - name: Check isolcpus IDs
        assert:
          that: "item | int <= ansible_processor_vcpus"
          msg:
            - Incorrect configuration pertaining isolcpus. Conflicting or improper values detected
            - The CPU ID {{ item }} set for isolcpus is NOT actually present on target. Please correct the configuration
        with_items: "{{ isolcpus_list }}"
        when: isolcpus is defined

#TODO relationship between cmk shared/exclusive cores and isolcpus

      when:
        - cmk_enabled is defined
        - (not cmk_hosts_list is defined) or (inventory_hostname in cmk_hosts_list)  #CMK expected true for all profiles except basic
      # {% if not cmk_use_all_hosts %}
        - "'kube-node' in group_names"
      # {% endif %}


####################################
# Prerequisites for Worker Node(s) #
####################################
- hosts: kube-node
  vars:
    phy_nics_pciids: []

  tasks:

# STORY: "nic bus info specified is present on system"
    - debug:
        msg: "Dataplane (DP) interface(s) defined in host_vars = {{ dataplane_interfaces }}"
      when: dataplane_interfaces is defined

    - name: Check DP Interfaces
      assert:
        that: "dataplane_interfaces != []"
        msg: "Dataplane (DP) interface(s) on target '{{ ansible_hostname }}' must be set in host_vars. Please correct the configuration"
      when:
        - dataplane_interfaces is defined
        - (update_nic_drivers is defined and update_nic_drivers) or
          (install_ddp_packages is defined and install_ddp_packages) or
          (sriov_cni_enabled is defined and sriov_cni_enabled) or
          (sriov_network_operator_enabled is defined and sriov_network_operator_enabled)

    - debug:
        msg: "Network interfaces present on target '{{ ansible_hostname }}' = {{ ansible_interfaces }}"

    - name: Read Physical NICs PCIIDs
      set_fact:
        phy_nics_pciids: "{{ phy_nics_pciids + [ ansible_facts[item]['pciid'] ] }}"
      with_items: "{{ ansible_interfaces }}"
      when: ansible_facts[item]['pciid'] is defined and ansible_facts[item]['type'] == "ether"

    - debug: msg="PCI Slots for the NICs on target '{{ ansible_hostname }}' = {{ phy_nics_pciids }}"

    - name: Check DP Interfaces Names
      assert:
        that: ("{{ item.name }}" in {{ ansible_interfaces }})
        msg: "Dataplane interface '{{ item.name }}' defined in host_vars does NOT exist on target. Please correct the configuration"
      with_items: "{{ dataplane_interfaces }}"
      when: dataplane_interfaces is defined and dataplane_interfaces != []
      ignore_errors: True

    - name: Check DP Interfaces Bus Info
      assert:
        that: ("{{ item.bus_info }}" in "{{ phy_nics_pciids }}")
        msg: "Dataplane interface '{{ item.name }}' defined with PCI ID '{{ item.bus_info }}' does NOT exist on target. Please correct the configuration"
      with_items: "{{ dataplane_interfaces }}"
      when: dataplane_interfaces is defined and dataplane_interfaces != []
      ignore_errors: True


# QAT Devices list is okay to be left empty (default), but if was defined, device(s) must exist on target
    - debug:
        msg: "QAT device(s) defined in host_vars = {{ qat_devices }}"
      when: qat_devices is defined

    - name: Read QAT PCIIDs
      shell: lshw -businfo -numeric | grep -i quickassist
      register: lshw_qat
      ignore_errors: True
      when: qat_devices is defined

    - debug:
        msg: "QAT devices found on target = {{ lshw_qat.stdout }}"
      when: qat_devices is defined

    - name: Check QAT Devices' Bus Info
      assert:
        that: ("{{ item.qat_id }}" in """{{ lshw_qat.stdout }}""")
        msg: "QAT device '{{ item.qat_dev }}' defined with PCI ID '{{ item.qat_id }}' does NOT exist on target. Please correct the configuration"
      with_items: "{{ qat_devices }}"
      when: qat_devices is defined and qat_devices != []
      ignore_errors: True

# STORY: "qat_sriov_numvfs should not exceed max supported (16) per each dev_ID"
    - debug:
        msg:
          - qat_sriov_numvfs for {{ item.qat_id }} = {{ item.qat_sriov_numvfs }} (host_vars)
          - update_qat_drivers = {{ update_qat_drivers }} (host_vars)
      with_items: "{{ qat_devices }}"
      when: qat_devices is defined and qat_devices != [] # update_qat_drivers expected as 'true' for all profiles except basic

    - name: Check QAT SRIOV VFs
      assert:
        that: ({{ item.qat_sriov_numvfs }} <= 16)
        msg:
          - Incorrect configuration pertaining QAT SRIOV. Conflicting or improper values detected
          - When SRIOV VFs are set for QAT, max value is 16 for each ID (max 48 total per card). Please correct the configuration
      with_items: "{{ qat_devices }}"
      when:
        - update_qat_drivers is defined and update_qat_drivers
        - qat_devices is defined and qat_devices != []
# OpenSSL & OpenSSL*Engine must only be configured / installed when update_qat_drivers is set to 'true' and qat_devices is defined in host vars
    - name: check OpenSSL and OpenSSL*Engine requirements
      assert:
        that:
          - update_qat_drivers
          - qat_devices is defined and qat_devices != []
        fail_msg: "OpenSSL & OpenSSL*Engine will only configured if update_qat_drivers is set to 'true' & qat_devices is defined in host vars"
        success_msg: "OpenSSL & OpenSSL*Engine verification completed"
      when: openssl_install is defined and openssl_install

    - name: check KMRA requirements
      assert:
        that:
          - sgx_dp_enabled
        fail_msg: "KMRA installation requires sgx_dp_enabled set to 'true'"
        success_msg: "KMRA requirements verified"
      when: kmra_enabled is defined and kmra_enabled

    - name: check SGX configuration
      assert:
        that:
          - sgx_enabled
        fail_msg: "SGX drivers installation requires sgx_enabled set to 'true'"
        success_msg: "SGX configuration verified"
      when:
        - sgx_dp_enabled is defined and sgx_dp_enabled
        - (ansible_distribution == 'Ubuntu' and ansible_distribution_version != '21.04')
          or (ansible_os_family == 'RedHat' and ansible_distribution_version != '8.4')

    - name: check NFD configuration
      assert:
        that:
          - nfd_enabled
        fail_msg: "SGX DP requires nfd_enabled set to 'true'"
        success_msg: "NFD configuration verified"
      when: sgx_dp_enabled is defined and sgx_dp_enabled

    - name: check kmra_pccs_api_key presence
      assert:
        that:
          - kmra_pccs_api_key is defined
        fail_msg:
          - "kmra_pccs_api_key is not defined"
        success_msg: "kmra_pccs_api_key presence is verified"
      when:
        - kmra_enabled is defined and kmra_enabled

    - name: check PCCS API key length
      assert:
        that:
          - kmra_pccs_api_key | length == 32
        fail_msg: "PCCS API Key should be 32 bytes long"
        success_msg: "PCCS API key length verified"
      when:
        - kmra_enabled is defined and kmra_enabled

    - name: check PCCS API key is not a placeholder
      assert:
        that:
          - kmra_pccs_api_key is defined
          - kmra_pccs_api_key != "ffffffffffffffffffffffffffffffff"
        fail_msg:
          - "Please, visit https://api.portal.trustedservices.intel.com/provisioning-certification and click on 'Subscribe'"
          - "to generate PCCS API key."
          - "PCCS API key is essential for KMRA deployment and usage."
        success_msg: "PCCS API key verified"
      when:
        - kmra_enabled is defined and kmra_enabled

# STORY: "vpp/ovsdpdk require hugepage enabled and configured"
    - debug:
        msg:
          - vpp_enabled = {{ vpp_enabled }} (host_vars)
          - example_net_attach_defs = {{ example_net_attach_defs }} (group_vars/all.yml)
          - userspace_ovs_dpdk = {{ example_net_attach_defs['userspace_ovs_dpdk'] }} (group_vars/all.yml)
          - userspace_vpp = {{ example_net_attach_defs['userspace_vpp'] }} (group_vars/all.yml)
          - sriov_net_dp = {{ example_net_attach_defs['sriov_net_dp'] }} (group_vars/all.yml)
          - userspace_cni_enabled = {{ userspace_cni_enabled }} (host_vars)
          - sriov_cni_enabled = {{ sriov_cni_enabled }} (host_vars)
          - sriov_network_operator_enabled = {{ sriov_network_operator_enabled }} (host_vars)
          - bond_cni_enabled = {{ bond_cni_enabled }} (host_vars)
          - ovs_dpdk_enabled = {{ ovs_dpdk_enabled }} (host_vars)
          - userspace_cni_enabled = {{ userspace_cni_enabled }} (host_vars)
          - hugepages_enabled = {{ hugepages_enabled }} (host_vars)
          - default_hugepage_size = {{ default_hugepage_size }} (host_vars)
          - number_of_hugepages = {{ number_of_hugepages }} (host_vars)
      when: vpp_enabled is defined #host_vars

    - name: Check OVS DPDK Dependencies
      assert:
        that: >-
          ({{ ovs_dpdk_enabled }} and not {{ vpp_enabled }} and {{ hugepages_enabled }} and
          "{{ default_hugepage_size }}" == "1G" and {{ number_of_hugepages }} >= 0)
          or {{ vpp_enabled }}
        msg:
          - Incorrect configuration pertaining OVS DPDK. Conflicting or improper values detected
          - When OVS DPDK is enabled, VPP must be disabled and Hugepages must be set to 1G according to host_vars example. Please correct the configuration
      when: ovs_dpdk_enabled is defined and ovs_dpdk_enabled

    - name: Check VPP Dependencies
      assert:
        that: >-
          ({{ vpp_enabled }} and not {{ ovs_dpdk_enabled }} and {{ hugepages_enabled }} and
          "{{ default_hugepage_size }}" == "2M" and {{ number_of_hugepages }} >= 0)
          or {{ ovs_dpdk_enabled }}
        msg:
          - Incorrect configuration pertaining VPP. Conflicting or improper values detected
          - When VPP is enabled, OVS DPDK must be disabled and Hugepages must be set to 2M according to host_vars example. Please correct the configuration
      when: vpp_enabled is defined and vpp_enabled


# STORY: "cnis require net-attach-defs to be enabled"
    - name: Check CNI Config
      assert:
        that: >-
          ({{ userspace_cni_enabled }} and {{ ovs_dpdk_enabled }} and {{ example_net_attach_defs['userspace_ovs_dpdk'] }} and not {{ vpp_enabled }} and
          not {{ example_net_attach_defs['userspace_vpp'] }} and {{ hugepages_enabled }} and
          "{{ default_hugepage_size }}" == "1G" and {{ number_of_hugepages }} >= 0)
          or ({{ userspace_cni_enabled }} and not {{ ovs_dpdk_enabled }} and not {{ example_net_attach_defs['userspace_ovs_dpdk'] }} and {{ vpp_enabled }}
          and {{ example_net_attach_defs['userspace_vpp'] }} and {{ hugepages_enabled }} and
          "{{ default_hugepage_size }}" == "2M" and {{ number_of_hugepages }} >= 0)
        msg:
          - Incorrect configuration pertaining CNI. Conflicting or improper values detected.
          - When CNI is enabled, either OVS DPDK either VPP must be enabled and Hugepages must be according to example files. Please correct the configuration
      when: userspace_cni_enabled is defined and userspace_cni_enabled


# STORY: "If SST enabled, confirm minimum kernel or kernel_update specified"
    - name: Check SST # see Jira NPF-1545
      assert:
        that: (not sst_bf_configuration_enabled)
        msg: "SST-BF is NOT supported on {{ ansible_distribution }} {{ ansible_distribution_version }}. Please use a different OS or disable this feature"
      when:
        - sst_bf_configuration_enabled is defined
        - (ansible_distribution == "RedHat" and ansible_distribution_version == '8.2') or ansible_distribution_version in ['7.6', '7.8', '7.9', '18.04']
      ignore_errors: True

# STORY: Intel VT-d should be enabled in BIOS
    - name: Check Intel VT-d
      shell: dmesg | grep DMAR | grep remapping
      register: dmesg_dmar_remap
      ignore_errors: True
      changed_when: False

    - debug: msg="dmesg >> {{ dmesg_dmar_remap.stdout }}"

    - name: Warn about Intel VT-d
      fail:
        msg: "Warning: Intel VT-d appears DISABLED on target.  Please check BIOS under 'Advanced > Integrated IO Configuration' and Enable if necessary"
      when: dmesg_dmar_remap.stdout|length  == 0
      ignore_errors: True


# STORY: CPU Hyper-Threading should be enabled in BIOS
    - name: Warn about Hyper-Threading
      fail:
        msg: "Warning: CPU Hyper-Threading is DISABLED on target.  Please check BIOS under 'Advanced > Processor Configuration' and Enable if necessary"
      when: ansible_processor_threads_per_core != 2
      ignore_errors: True


# STORY: "check for collectd. See Jira NPF-1687"
    - name: Warn about collectd
      fail:
        msg: "Warning: On {{ ansible_distribution }} {{ ansible_distribution_version }} collectd won't work unless 'update_kernel' is enabled in group_vars"
      when: ansible_distribution_version in ['7.6', '18.04']
      ignore_errors: True


# STORY: TEMPORARY: "ovs dpdk version requirements"
    - debug:
        msg:
          - install_dpdk = {{ install_dpdk }} (host_vars)
          - dpdk_version = {{ dpdk_version }} (host_vars)
          - ovs_dpdk_enabled = {{ ovs_dpdk_enabled }} (host_vars)
          - ovs_version = {{ ovs_version }} (host_vars)
      when:
        - install_dpdk is defined #host_vars
        - dpdk_version is defined #host_vars
        - ovs_version is defined #host_vars
        - ovs_dpdk_enabled is defined and ovs_dpdk_enabled #host_vars

    - name: Check OVS DPDK compatibility
      assert:
        that: >-
          "{{ ovs_version }} == \"v2.15.0\" and {{ dpdk_version }} >= \"20.11\""
          or "{{ ovs_version }} == \"v2.14.2\" and {{ dpdk_version }} == \"19.11.6\""
          or "{{ ovs_version }} == \"v2.14.1\" and {{ dpdk_version }} == \"19.11.6\""
          or "{{ ovs_version }} == \"v2.14.0\" and {{ dpdk_version }} == \"19.11.6\""
          or "{{ ovs_version }} == \"v2.13.3\" and {{ dpdk_version }} == \"19.11.6\""
          or "{{ ovs_version }} == \"v2.13.2\" and {{ dpdk_version }} == \"19.11.6\""
          or "{{ ovs_version }} == \"v2.13.1\" and {{ dpdk_version }} == \"19.11.6\""
          or "{{ ovs_version }} == \"v2.13.0\" and {{ dpdk_version }} == \"19.11.6\""
        msg: "OVS {{ ovs_version }} does not build with DPDK version {{ dpdk_version }}. Please correct the host_vars configuration"
      when:
        - dpdk_version is defined #host_vars
        - ovs_version is defined #host_vars
        - ovs_dpdk_enabled is defined and ovs_dpdk_enabled #host_vars


    - meta: end_play

#    - name: Print all variables/facts known for a host
#      ansible.builtin.debug:
#        var: hostvars[inventory_hostname]
#        verbosity: 4