[1.0.3] backports

CHANGELOG-1.0.md

@@ -4,6 +4,16 @@
 
 ### Added
 
+- [#2976](https://github.com/epiphany-platform/epiphany/issues/2976) - Allow for custom Terraform scripts
+- [#1076](https://github.com/epiphany-platform/epiphany/issues/1076) - Add sorting entries in the inventory file
+
+### Fixed
+
+- [#2989](https://github.com/epiphany-platform/epiphany/issues/2989) - Task `Remove swap from /etc/fstab` does not remove swap entry from file
+- [#2653](https://github.com/epiphany-platform/epiphany/issues/2653) - Epicli is failing in air-gapped infra mode
+- [#3004](https://github.com/epiphany-platform/epiphany/issues/3004) - `disk_size_gb` hardcoded in `/usr/local/epicli/data/azure/defaults/infrastructure/virtual-machine.yml`
+- [#2934](https://github.com/epiphany-platform/epiphany/issues/2934) - Duplicated entries in image-registry defaults (LTS)
+- [#3065](https://github.com/epiphany-platform/epiphany/issues/3065) - Flag `delete_os_disk_on_termination` has no effect when removing cluster
 - [#2996](https://github.com/epiphany-platform/epiphany/issues/2996) - Introduce the new configuration field to change a component name
 
 ## [1.0.2] 2022-01-26

core/src/epicli/.devcontainer/Dockerfile

@@ -4,13 +4,8 @@ ARG USERNAME=vscode
 ARG USER_UID=1000
 ARG USER_GID=$USER_UID
 
-COPY cert/cert.md cert/*.crt /usr/local/share/ca-certificates/
-COPY cert/cert.md cert/*.pem /
-COPY cert/config-pre.sh /
-
 RUN : INSTALL APT REQUIREMENTS \
  && export DEBIAN_FRONTEND=noninteractive \
- && /bin/bash /config-pre.sh \
  && apt-get -q update \
  && apt-get -q install -y --no-install-recommends \
     apt-utils dialog \
@@ -54,14 +49,11 @@ RUN : INSTALL PIP REQUIREMENTS \
  && pip install --disable-pip-version-check --no-cache-dir --default-timeout=100 \
     --requirement /requirements.txt \
  && pip install --disable-pip-version-check --no-cache-dir --default-timeout=100 \
-    poetry pylint pytest setuptools twine wheel
-
-COPY cert/config-post.sh /
+    poetry pylint pytest pytest_mock setuptools twine wheel
 
 RUN : SETUP USER, CERTS AND OTHERS \
  && groupadd --gid $USER_GID $USERNAME \
  && useradd -s /bin/bash --uid $USER_UID --gid $USER_GID -m $USERNAME \
  && echo $USERNAME ALL=\(root\) NOPASSWD:ALL > /etc/sudoers.d/$USERNAME \
  && chmod ug=r,o= /etc/sudoers.d/$USERNAME \
- && setcap 'cap_net_bind_service=+ep' /usr/bin/ssh \
- && /bin/bash /config-post.sh
+ && setcap 'cap_net_bind_service=+ep' /usr/bin/ssh

core/src/epicli/cli/engine/providers/any/APIProxy.py

@@ -1,6 +1,6 @@
 from cli.helpers.doc_list_helpers import select_first
 from cli.helpers.Log import Log
-from cli.models.AnsibleHostModel import AnsibleHostModel
+from cli.models.AnsibleHostModel import AnsibleOrderedHostModel
 
 
 class APIProxy:
@@ -22,5 +22,8 @@ def get_ips_for_feature(self, component_key):
             for machine in component_config.machines:
                 machine_doc = select_first(self.config_docs,
                                            lambda x: x.kind == 'infrastructure/machine' and x.name == machine)
-                result.append(AnsibleHostModel(machine_doc.specification.hostname, machine_doc.specification.ip))
+                result.append(AnsibleOrderedHostModel(machine_doc.specification.hostname, machine_doc.specification.ip))
+
+        result.sort()
+
         return result

core/src/epicli/cli/engine/providers/azure/APIProxy.py

@@ -5,7 +5,7 @@
 from cli.helpers.Log import LogPipe, Log
 from cli.helpers.doc_list_helpers import select_first
 from cli.helpers.naming_helpers import resource_name, cluster_tag
-from cli.models.AnsibleHostModel import AnsibleHostModel
+from cli.models.AnsibleHostModel import AnsibleOrderedHostModel
 
 class APIProxy:
     def __init__(self, cluster_model, config_docs):
@@ -54,7 +54,7 @@ def get_ips_for_feature(self, component_key):
         look_for_public_ip = self.cluster_model.specification.cloud.use_public_ips
         cluster = cluster_tag(self.cluster_prefix, self.cluster_name)
         running_instances = self.run(self, f'az vm list-ip-addresses --ids $(az resource list --query "[?type==\'Microsoft.Compute/virtualMachines\' && tags.{component_key} == \'\' && tags.cluster == \'{cluster}\'].id" --output tsv)')
-        result = []
+        result: List[AnsibleOrderedHostModel] = []
         for instance in running_instances:
             if isinstance(instance, list):
                 instance = instance[0]   
@@ -66,7 +66,10 @@ def get_ips_for_feature(self, component_key):
                 ip = instance['virtualMachine']['network']['publicIpAddresses'][0]['ipAddress']
             else:
                 ip = instance['virtualMachine']['network']['privateIpAddresses'][0]
-            result.append(AnsibleHostModel(name, ip))
+            result.append(AnsibleOrderedHostModel(name, ip))
+
+        result.sort()
+
         return result
 
     def get_storage_account_primary_key(self, storage_account_name):

core/src/epicli/cli/engine/schema/DefaultMerger.py

@@ -30,7 +30,10 @@ def merge_parent(self, files, doc):
             merged_parent['version'] = VERSION
             merge_objdict(merged_parent, doc)
             return merged_parent
+        default_config = select_first(self.docs, lambda x: x.name == 'default' and x.kind == doc.kind)
         default_doc = select_first(files, lambda x: x.name == 'default')
+        if default_config is not None:
+            merge_objdict(default_doc, default_config)
         default_doc['version'] = VERSION
         merge_objdict(default_doc, doc)
-        return default_doc
+        return default_doc

core/src/epicli/cli/engine/terraform/TerraformTemplateGenerator.py

@@ -12,8 +12,9 @@ def __init__(self, cluster_model, infrastructure):
 
     def run(self):
         terraform_output_dir = get_terraform_path(self.cluster_model.specification.name)
-        # Remove generated .tf files (not tfstate).
-        remove_files_matching_glob(terraform_output_dir, '*.tf')
+
+        # Only remove epicli generated .tf files, not tfstate or user created files.
+        remove_files_matching_glob(terraform_output_dir, '[0-9][0-9][0-9]_*.tf')
 
         templates = filter(lambda x: x.kind != 'infrastructure/cloud-init-custom-data', self.infrastructure)
         for idx, doc in enumerate(templates):

core/src/epicli/cli/epicli.py

@@ -227,7 +227,9 @@ def delete_parser(subparsers):
     sub_parser._action_groups.append(optional)
 
     def run_delete(args):
-        if not query_yes_no('Do you really want to delete your cluster?'):
+        if not query_yes_no('''You are trying to delete your cluster.
+If your configuration does not allow to keep the existing disks used in the cluster, you will lose your data.
+Make sure your data is safe. Do you really want to delete your cluster?'''):
             return 0
         adjust_paths_from_build(args)
         with DeleteEngine(args) as engine:

core/src/epicli/cli/models/AnsibleHostModel.py

@@ -1,4 +1,20 @@
 class AnsibleHostModel:
-    def __init__(self, name, ip):
-        self.name = name
-        self.ip = ip
+    def __init__(self, name: str, ip: str):
+        self.name: str = name
+        self.ip: str = ip
+
+    def __eq__(self, other) -> bool:
+        return (self.name == other.name and
+                self.ip == other.ip)
+
+    def __lt__(self, other) -> bool:
+        pass
+
+
+class AnsibleOrderedHostModel(AnsibleHostModel):
+    """
+    Sortable variant of AnsibleHostModel
+    """
+
+    def __lt__(self, other) -> bool:
+        return self.name < other.name

core/src/epicli/data/azure/defaults/infrastructure/virtual-machine.yml

@@ -20,7 +20,6 @@ specification:
     sku: 18.04-LTS
     version: "18.04.202103151" # Never put latest on anything! Need to always pin the version number but testing we can get away with it
   storage_os_disk:
-    delete_on_termination: false
     managed: false
     caching: ReadWrite
     create_option: FromImage

core/src/epicli/data/azure/terraform/infrastructure/virtual-machine.j2

@@ -49,10 +49,6 @@ resource "azurerm_virtual_machine" "{{ specification.name }}" {
   # windows specific stuff here maybe... hopefully never.
   {%- endif %}
 
-  {%- if specification.storage_os_disk.managed != true %}
-  delete_os_disk_on_termination = "{{ specification.storage_os_disk.delete_on_termination | lower }}"
-  {%- endif %}
-
   storage_os_disk {
     name          = "{{ specification.name }}-os-disk"
     caching       = "{{ specification.storage_os_disk.caching }}"

core/src/epicli/data/azure/validation/infrastructure/virtual-machine.yml

@@ -41,8 +41,6 @@ properties:
   storage_os_disk:
     type: object
     properties:
-      delete_on_termination:
-        type: boolean
       managed:
         type: boolean
       caching:

core/src/epicli/data/common/ansible/playbooks/roles/common/tasks/main.yml

@@ -26,11 +26,11 @@
     - sysctl
 
 - name: Remove swap from /etc/fstab
-  mount:
-    backup: yes
-    fstype: swap
-    path: swap
+  lineinfile:
+    path: /etc/fstab
+    regexp: ^(?!#).*\sswap\s.*\d\s+\d$
     state: absent
+    backup: yes
   tags:
     - disableswap
 

core/src/epicli/data/common/ansible/playbooks/roles/preflight/tasks/check-routing.yml

@@ -0,0 +1,46 @@
+---
+- name: Check routing configuration
+  become: true
+  command: ip route show default 0.0.0.0/0
+  register: ip_route_show_default
+
+- name: Assert default route exists
+  assert:
+    that: ip_route_show_default.stdout_lines | length > 0
+    fail_msg: >-
+      No default route configured. At least one is required, read more in troubleshooting document.
+    quiet: true
+
+- name: Validate metric values if multiple default routes exist
+  when: ip_route_show_default.stdout_lines | length > 1
+  block:
+    - name: Get metric values
+      become: true
+      shell: |-
+        set -o pipefail && \
+        ip route show default 0.0.0.0/0 | awk '{if (! /metric/) print 0; else for (x=1;x<NF;x++) if ($x == "metric") print $(x+1) }'
+      register: default_routing_configuration_metric_values
+      args:
+        executable: /bin/bash
+
+    - name: Assert two most prioritized default routes have unique metric
+      assert:
+        that: "default_routing_configuration_metric_values.stdout_lines[0] != default_routing_configuration_metric_values.stdout_lines[1]"
+        fail_msg:  >-
+          At least two default routes have the same metric value.
+          Check routing configuration, read more in troubleshooting document.
+
+- include_vars:
+    file: roles/common/vars/main.yml
+    name: common_vars
+
+- name: Validate if ansible_default_ipv4.address matches address from inventory
+  when:
+    - common_vars.provider == "any"
+    - common_vars.specification.cloud is undefined
+  assert:
+    that: ansible_default_ipv4.address == ansible_host
+    fail_msg:  >-
+      ansible_default_ipv4.address is {{ ansible_default_ipv4.address }} but inventory uses ip: {{ ansible_host }}.
+      Check default routing configuration, read more in troubleshooting document.
+    quiet: true

core/src/epicli/data/common/ansible/playbooks/roles/preflight/tasks/main.yml

@@ -3,6 +3,8 @@
 
 - include_tasks: check-os.yml
 
+- include_tasks: check-routing.yml
+
 - name: Check if existing PostgreSQL needs to be migrated to distribution installed from PostgreSQL repository
   block:
 

core/src/epicli/data/common/defaults/configuration/image-registry.yml

@@ -74,18 +74,22 @@ specification:
         file_name: kube-scheduler-v1.14.6.tar
       - name: "k8s.gcr.io/kube-proxy:v1.14.6"
         file_name: kube-proxy-v1.14.6.tar
-      - name: "k8s.gcr.io/etcd:3.3.10"
-        file_name: etcd-3.3.10.tar
-      - name: "k8s.gcr.io/coredns:1.3.1"
-        file_name: coredns-1.3.1.tar
+      # Disable this dual entry because of validation issues.
+      # We leave it for reference.
+      # - name: "k8s.gcr.io/etcd:3.3.10"
+      #   file_name: etcd-3.3.10.tar
+      # - name: "k8s.gcr.io/coredns:1.3.1"
+      #   file_name: coredns-1.3.1.tar
       - name: "coredns/coredns:1.5.0"
         file_name: coredns-1.5.0.tar
-      - name: "k8s.gcr.io/pause:3.1"
-        file_name: pause-3.1.tar
-      - name: "quay.io/coreos/flannel:v0.11.0-amd64"
-        file_name: flannel-v0.11.0-amd64.tar
-      - name: "quay.io/coreos/flannel:v0.11.0"
-        file_name: flannel-v0.11.0.tar
+      # Disable this dual entry because of validation issues.
+      # We leave it for reference
+      # - name: "k8s.gcr.io/pause:3.1"
+      #   file_name: pause-3.1.tar
+      # - name: "quay.io/coreos/flannel:v0.11.0-amd64"
+      #   file_name: flannel-v0.11.0-amd64.tar
+      # - name: "quay.io/coreos/flannel:v0.11.0"
+      #   file_name: flannel-v0.11.0.tar
       - name: "calico/cni:v3.8.1"
         file_name: cni-v3.8.1.tar
       - name: "calico/kube-controllers:v3.8.1"
@@ -129,12 +133,14 @@ specification:
         file_name: kube-scheduler-v1.17.4.tar
       - name: "k8s.gcr.io/kube-proxy:v1.17.4"
         file_name: kube-proxy-v1.17.4.tar
-      - name: "k8s.gcr.io/etcd:3.4.3-0"
-        file_name: etcd-3.4.3-0.tar
-      - name: "k8s.gcr.io/coredns:1.6.5"
-        file_name: coredns-1.6.5.tar
-      - name: "k8s.gcr.io/pause:3.1"
-        file_name: pause-3.1.tar
+      # Disable this dual entry because of validation issues.
+      # We leave it for reference.
+      # - name: "k8s.gcr.io/etcd:3.4.3-0"
+      #   file_name: etcd-3.4.3-0.tar
+      # - name: "k8s.gcr.io/coredns:1.6.5"
+      #   file_name: coredns-1.6.5.tar
+      # - name: "k8s.gcr.io/pause:3.1"
+      #   file_name: pause-3.1.tar
       # flannel
       - name: "quay.io/coreos/flannel:v0.11.0-amd64"
         file_name: flannel-v0.11.0-amd64.tar

core/src/epicli/tests/engine/providers/any/test_APIProxy.py

@@ -0,0 +1,26 @@
+from pytest_mock import MockerFixture
+
+from cli.engine.providers.any.APIProxy import APIProxy
+from cli.models.AnsibleHostModel import AnsibleOrderedHostModel
+from tests.engine.providers.data.APIProxy_data import CLUSTER_MODEL, CONFIG_DOC
+
+
+def test_get_ips_for_feature(mocker: MockerFixture):
+    """
+    Make sure that hostnames in inventory are sorted.
+    """
+
+    mocker.patch('cli.engine.providers.any.APIProxy.Log')
+    proxy = APIProxy(CLUSTER_MODEL('any'), CONFIG_DOC())
+
+    EXPECTED_RESULT = [
+        AnsibleOrderedHostModel('service-vm-0', '20.73.105.240'),
+        AnsibleOrderedHostModel('service-vm-1', '20.73.105.188'),
+        AnsibleOrderedHostModel('service-vm-2', '20.73.105.18'),
+        AnsibleOrderedHostModel('service-vm-3', '20.73.105.33'),
+        AnsibleOrderedHostModel('service-vm-4', '20.73.105.54')
+    ]
+
+    result = proxy.get_ips_for_feature('service')
+
+    assert EXPECTED_RESULT == result