Cleanup via Ansible

+ Remove dependency on topo_ scripts that depends on custom SSH()
  that depends on framework itself. This way the cleanup is independent
  of failure in our SSH libs.
+ Simple ansible command can do cleanup of a machine:
  ansible-playbook --inventory inventories/lf_inventory/hosts site.yaml \
  --limit '10.32.8.18' --tags 'cleanup'
+ Add vpp_device reset and cleanup.
+ Remove historical scripts.

- Still in testing beta phase.
  - Need to add SRIOV cleanup.

Signed-off-by: Peter Mikus <pmikus@cisco.com>
Change-Id: I68e23304c7ad01041f51263c328c6e8d9b555cb7

8 files changed, 311 insertions, 0 deletions

diff --git a/resources/tools/testbed-setup/ansible/roles/cleanup/files/reset_vppdevice.sh b/resources/tools/testbed-setup/ansible/roles/cleanup/files/reset_vppdevice.sh
new file mode 100644
index 0000000000..ede2db1273
--- /dev/null
+++ b/resources/tools/testbed-setup/ansible/roles/cleanup/files/reset_vppdevice.sh
@@ -0,0 +1,113 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+function die () {
+    # Print the message to standard error end exit with error code specified
+    # by the second argument.
+    #
+    # Hardcoded values:
+    # - The default error message.
+    # Arguments:
+    # - ${1} - The whole error message, be sure to quote. Optional
+    # - ${2} - the code to exit with, default: 1.
+
+    set +eu
+    warn "${1:-Unspecified run-time error occurred!}"
+    exit "${2:-1}"
+}
+
+
+function set_eligibility_off {
+    # Set Nomad eligibility to ineligible for scheduling. Fail otherwise.
+
+    set -euo pipefail
+
+    node_id="$(nomad node status | grep $(hostname) | cut -d ' ' -f 1)" || die
+    node_status="$(nomad node status | grep $(hostname))" || die
+
+    if [[ "${node_status}" != *"ineligible"* ]]; then
+        nomad node eligibility -disable "${node_id}" || die
+        node_status="$(nomad node status | grep $(hostname))" || die
+        if [[ "${node_status}" != *"ineligible"* ]]; then
+            die "Set eligibility off failed!"
+        fi
+    fi
+}
+
+
+function set_eligibility_on {
+    # Set Nomad eligibility to eligible for scheduling. Fail otherwise.
+
+    set -euo pipefail
+
+    node_id="$(nomad node status | grep $(hostname) | cut -d ' ' -f 1)" || die
+    node_status="$(nomad node status | grep $(hostname))" || die
+
+    if [[ "${node_status}" == *"ineligible"* ]]; then
+        nomad node eligibility -enable "${node_id}" || die
+        node_status="$(nomad node status | grep $(hostname))" || die
+        if [[ "${node_status}" == *"ineligible"* ]]; then
+            die "Set eligibility on failed!"
+        fi
+    fi
+}
+
+
+function restart_vfs_service {
+    # Stop and start VF serice. This will reinitialize VFs and driver mappings.
+
+    set -euo pipefail
+
+    warn "Restarting VFs service (this may take few minutes)..."
+    sudo service csit-initialize-vfs stop || die "Failed to stop VFs service!"
+    sudo service csit-initialize-vfs start || die "Failed to start VFs service!"
+}
+
+
+function wait_for_pending_containers {
+    # Wait in loop for defined amount of time for pending containers to
+    # gracefully quit them. If parameter force is specified. Force kill them.
+
+    # Arguments:
+    # - ${@} - Script parameters.
+
+    set -euo pipefail
+
+    retries=60
+    wait_time=60
+    containers=(docker ps --quiet --filter name=csit*)
+
+    for i in $(seq 1 ${retries}); do
+        mapfile -t pending_containers < <( ${containers[@]} ) || die
+        warn "Waiting for pending containers [${pending_containers[@]}] ..."
+        if [ ${#pending_containers[@]} -eq 0 ]; then
+            break
+        fi
+        sleep "${wait_time}" || die
+    done
+    if [ ${#pending_containers[@]} -ne 0 ]; then
+        if [[ "${1-}" == "force" ]]; then
+            warn "Force killing [${pending_containers[@]}] ..."
+            docker rm --force ${pending_containers[@]} || die
+        else
+            die "Still few containers running!"
+        fi
+    fi
+}
+
+
+function warn () {
+    # Print the message to standard error.
+    #
+    # Arguments:
+    # - ${@} - The text of the message.
+
+    echo "$@" >&2
+}
+
+
+set_eligibility_off || die
+wait_for_pending_containers "${@}" || die
+restart_vfs_service || die
+set_eligibility_on || die
diff --git a/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/kill_containers.yaml b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/kill_containers.yaml
new file mode 100644
index 0000000000..a61aa6ceee
--- /dev/null
+++ b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/kill_containers.yaml
@@ -0,0 +1,28 @@
+---
+# file: roles/cleanup/tasks/kill_containers.yaml
+
+- name: Kill container - Get running Docker containers
+  shell: "docker ps -aq"
+  register: running_containers
+  changed_when: no
+  tags: kill-containers
+
+- name: Kill container - Remove all Docker containers
+  docker_container:
+    name: "{{ item }}"
+    state: absent
+  with_items: "{{ running_containers.stdout_lines }}"
+  tags: kill-containers
+
+- name: Kill container - Get running LXC containers
+  shell: "lxc-ls"
+  register: running_containers
+  changed_when: no
+  tags: kill-containers
+
+- name: Kill container - Remove all LXC containers
+  lxc_container:
+    name: '{{ item }}'
+    state: absent
+  with_items: "{{ running_containers.stdout_lines }}"
+  tags: kill-containers
diff --git a/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/kill_process.yaml b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/kill_process.yaml
new file mode 100644
index 0000000000..4a1180b77f
--- /dev/null
+++ b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/kill_process.yaml
@@ -0,0 +1,27 @@
+---
+# file: roles/cleanup/tasks/kill_process.yaml
+
+- name: Kill process - Get pid of {{ process }}
+  shell: "ps -ef | grep -v grep | grep -w {{ process }} | awk '{print $2}'"
+  when: >
+    process is defined and process != ""
+  register: running_processes
+  tags: kill-process
+
+- name: Kill process - Safe kill {{ process }}
+  shell: "kill {{ item }}"
+  with_items: "{{ running_processes.stdout_lines }}"
+  tags: kill-process
+
+- wait_for:
+    path: "/proc/{{ item }}/status"
+    state: absent
+  with_items: "{{ running_processes.stdout_lines }}"
+  ignore_errors: yes
+  register: killed_processes
+  tags: kill-process
+
+- name: Kill process - Force kill {{ process }}
+  shell: "kill -9 {{ item }}"
+  with_items: "{{ killed_processes.results | select('failed') | map(attribute='item') | list }}"
+  tags: kill-process
diff --git a/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/main.yaml b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/main.yaml
new file mode 100644
index 0000000000..64a55c4672
--- /dev/null
+++ b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/main.yaml
@@ -0,0 +1,31 @@
+---
+# file: roles/cleanup/tasks/main.yaml
+# purpose: Structured per server cleanup tasks.
+# - main:
+#     - tg:
+#         - Run tasks on TG servers only.
+#         - Cleanup processes (T-Rex).
+#     - sut:
+#         - Run tasks on SUT servers only.
+#         - Cleanup file leftovers (logs).
+#         - Cleanup packages (VPP, Honeycomb).
+#         - Cleanup processes (qemu, l3fwd, testpmd, docker, kubernetes)
+#         - Cleanup interfaces.
+#     - vpp_device
+#         - Run tasks on vpp_device servers only.
+#         - Reset SRIOV
+
+- name: tg specific
+  include_tasks: tg.yaml
+  when: "'tg' in group_names"
+  tags: cleanup
+
+- name: sut specific
+  include_tasks: sut.yaml
+  when: "'sut' in group_names"
+  tags: cleanup
+
+- name: vpp_device specific
+  include_tasks: vpp_device.yaml
+  when: "'vpp_device' in group_names"
+  tags: cleanup
diff --git a/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/remove_package.yaml b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/remove_package.yaml
new file mode 100644
index 0000000000..8f5ec8fefe
--- /dev/null
+++ b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/remove_package.yaml
@@ -0,0 +1,31 @@
+---
+# file: roles/cleanup/tasks/remove_package.yaml
+
+- name: Remove package - Fix corrupted apt
+  shell: 'dpkg --configure -a'
+  when: >
+    ansible_distribution == 'Ubuntu'
+  tags: remove-package
+
+- name: Remove package - Check if {{ package }} is installed
+  shell: >
+    "dpkg-query -W -f='${Status}' {{ package }} | grep 'install ok installed'"
+  register: package_is_installed
+  failed_when: no
+  changed_when: no
+  when: >
+    ansible_distribution == 'Ubuntu'
+  tags: remove-package
+
+- name: Remove package - {{ package }}
+  apt:
+    name: '{{ package }}'
+    force: yes
+    purge: yes
+    state: absent
+  when: >
+    package is defined and
+    package != '' and
+    package_is_installed.rc == 0 and
+    ansible_distribution|lower == 'ubuntu'
+  tags: remove-package
diff --git a/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/sut.yaml b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/sut.yaml
new file mode 100644
index 0000000000..5083a96a29
--- /dev/null
+++ b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/sut.yaml
@@ -0,0 +1,52 @@
+---
+# file: roles/cleanup/tasks/sut.yaml
+
+- name: Kill processes - qemu
+  import_tasks: kill_process.yaml
+  vars:
+    process: "qemu"
+  tags: kill-process
+
+- name: Kill processes - l3fwd
+  import_tasks: kill_process.yaml
+  vars:
+    process: "l3fwd"
+  tags: kill-process
+
+- name: Kill processes - testpmd
+  import_tasks: kill_process.yaml
+  vars:
+    process: "testpmd"
+  tags: kill-process
+
+- name: Remove file or dir - HoneyComb logs
+  file:
+    state: absent
+    path: "/var/log/honeycomb"
+  tags: remove-file-dir
+
+- name: Remove file or dir - Core zip file
+  file:
+    state: absent
+    path: "/tmp/*tar.lzo.lrz.xz*"
+  tags: remove-file-dir
+
+- name: Remove file or dir - Core dump file
+  file:
+    state: absent
+    path: "/tmp/*core*"
+  tags: remove-file-dir
+
+- name: Kill containers - Remove all containers
+  import_tasks: kill_containers.yaml
+  tags: kill-containers
+
+- name: Kubernetes - Reset
+  raw: 'kubeadm reset --force'
+  tags: kill-kubernetes
+
+- name: Remove packages - Remove VPP
+  import_tasks: remove_package.yaml
+  vars:
+    package: "*vpp*"
+  tags: remove-package
diff --git a/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/tg.yaml b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/tg.yaml
new file mode 100644
index 0000000000..f58cb59a1a
--- /dev/null
+++ b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/tg.yaml
@@ -0,0 +1,14 @@
+---
+# file: roles/cleanup/tasks/tg.yaml
+
+- name: Kill processes - TRex
+  import_tasks: kill_process.yaml
+  vars:
+      process: "_t-rex"
+  tags: kill-process
+
+- name: Kill processes - WRK
+  import_tasks: kill_process.yaml
+  vars:
+      process: "wrk"
+  tags: kill-process
diff --git a/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/vpp_device.yaml b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/vpp_device.yaml
new file mode 100644
index 0000000000..5b7713a554
--- /dev/null
+++ b/resources/tools/testbed-setup/ansible/roles/cleanup/tasks/vpp_device.yaml
@@ -0,0 +1,15 @@
+---
+# file: roles/cleanup/tasks/vpp_device.yaml
+
+- name: Reset vpp_device binary
+  template:
+   src: 'files/reset_vppdevice.sh'
+   dest: '/usr/local/bin'
+   owner: 'root'
+   group: 'root'
+   mode: '644'
+  tags: reset-sriov
+
+- name: Reset vpp_device
+  raw: 'reset_vppdevice.sh --force'
+  tags: reset-sriov