Merge pull request #792 from dgoodwin/upgrade

Upgrade Improvements

2 files changed, 216 insertions, 22 deletions

diff --git a/playbooks/adhoc/upgrades/library/openshift_upgrade_config.py b/playbooks/adhoc/upgrades/library/openshift_upgrade_config.py
new file mode 100755
index 000000000..60f4fd8b8
--- /dev/null
+++ b/playbooks/adhoc/upgrades/library/openshift_upgrade_config.py
@@ -0,0 +1,117 @@
+#!/usr/bin/python
+# -*- coding: utf-8 -*-
+# vim: expandtab:tabstop=4:shiftwidth=4
+
+"""Ansible module for modifying OpenShift configs during an upgrade"""
+
+import os
+import shutil
+import yaml
+
+from datetime import datetime
+
+DOCUMENTATION = '''
+---
+module: openshift_upgrade_config
+short_description: OpenShift Upgrade Config
+author: Jason DeTiberus
+requirements: [ ]
+'''
+EXAMPLES = '''
+'''
+
+def get_cfg_dir():
+    """Return the correct config directory to use."""
+    cfg_path = '/etc/origin/'
+    if not os.path.exists(cfg_path):
+        cfg_path = '/etc/openshift/'
+    return cfg_path
+
+
+def upgrade_master_3_0_to_3_1(backup):
+    """Main upgrade method for 3.0 to 3.1."""
+    changed = False
+
+    # Facts do not get transferred to the hosts where custom modules run,
+    # need to make some assumptions here.
+    master_config = os.path.join(get_cfg_dir(), 'master/master-config.yaml')
+
+    master_cfg_file = open(master_config, 'r')
+    config = yaml.safe_load(master_cfg_file.read())
+    master_cfg_file.close()
+
+    # Remove v1beta3 from apiLevels:
+    if 'apiLevels' in config and \
+        'v1beta3' in config['apiLevels']:
+        config['apiLevels'].remove('v1beta3')
+        changed = True
+    if 'apiLevels' in config['kubernetesMasterConfig'] and \
+        'v1beta3' in config['kubernetesMasterConfig']['apiLevels']:
+        config['kubernetesMasterConfig']['apiLevels'].remove('v1beta3')
+        changed = True
+
+    # Add the new master proxy client certs:
+    # TODO: re-enable this once these certs are generated during upgrade:
+#    if 'proxyClientInfo' not in config['kubernetesMasterConfig']:
+#        config['kubernetesMasterConfig']['proxyClientInfo'] = {
+#            'certFile': 'master.proxy-client.crt',
+#            'keyFile': 'master.proxy-client.key'
+#       }
+
+    if changed:
+        if backup:
+            timestamp = datetime.now().strftime('%Y%m%d%H%M%S')
+            basedir = os.path.split(master_config)[0]
+            backup_file = os.path.join(basedir, 'master-config.yaml.bak-%s'
+                                       % timestamp)
+            shutil.copyfile(master_config, backup_file)
+        # Write the modified config:
+        out_file = open(master_config, 'w')
+        out_file.write(yaml.safe_dump(config, default_flow_style=False))
+        out_file.close()
+
+    return changed
+
+
+def upgrade_master(from_version, to_version, backup):
+    """Upgrade entry point."""
+    if from_version == '3.0':
+        if to_version == '3.1':
+            return upgrade_master_3_0_to_3_1(backup)
+
+
+def main():
+    """ main """
+    # disabling pylint errors for global-variable-undefined and invalid-name
+    # for 'global module' usage, since it is required to use ansible_facts
+    # pylint: disable=global-variable-undefined, invalid-name
+    global module
+
+    module = AnsibleModule(
+        argument_spec=dict(
+            from_version=dict(required=True, choices=['3.0']),
+            to_version=dict(required=True, choices=['3.1']),
+            role=dict(required=True, choices=['master']),
+            backup=dict(required=False, default=True, type='bool')
+        ),
+        supports_check_mode=True,
+    )
+
+    from_version = module.params['from_version']
+    to_version = module.params['to_version']
+    role = module.params['role']
+    backup = module.params['backup']
+
+    changed = False
+    if role == 'master':
+        changed = upgrade_master(from_version, to_version, backup)
+
+    return module.exit_json(changed=changed)
+
+# ignore pylint errors related to the module_utils import
+# pylint: disable=redefined-builtin, unused-wildcard-import, wildcard-import
+# import module snippets
+from ansible.module_utils.basic import *
+
+if __name__ == '__main__':
+    main()
diff --git a/playbooks/adhoc/upgrades/upgrade.yml b/playbooks/adhoc/upgrades/upgrade.yml
index ae1d0127c..09f991b1d 100644
--- a/playbooks/adhoc/upgrades/upgrade.yml
+++ b/playbooks/adhoc/upgrades/upgrade.yml
@@ -1,4 +1,57 @@
 ---
+- name: Update deployment type
+  hosts: OSEv3
+  roles:
+  - openshift_facts
+  post_tasks: # technically tasks are run after roles, but post_tasks is a bit more explicit.
+  - openshift_facts:
+      role: common
+      local_facts:
+        deployment_type: "{{ deployment_type }}"
+
+- name: Verify upgrade can proceed
+  hosts: masters
+  tasks:
+  # Checking the global deployment type rather than host facts, this is about
+  # what the user is requesting.
+    - fail: msg="Deployment type enterprise not supported for upgrade"
+      when: deployment_type == "enterprise"
+
+- name: Backup etcd
+  hosts: masters
+  vars:
+    embedded_etcd: "{{ openshift.master.embedded_etcd }}"
+    timestamp: "{{ lookup('pipe', 'date +%Y%m%d%H%M%S') }}"
+  roles:
+  - openshift_facts
+  tasks:
+  - stat: path=/var/lib/openshift
+    register: var_lib_openshift
+  - name: Create origin symlink if necessary
+    file: src=/var/lib/openshift/ dest=/var/lib/origin state=link
+    when: var_lib_openshift.stat.exists == True
+  - name: Check available disk space for etcd backup
+    # We assume to be using the data dir for all backups.
+    shell: >
+      df --output=avail -k {{ openshift.common.data_dir }} | tail -n 1
+    register: avail_disk
+
+  - name: Check current embedded etcd disk usage
+    shell: >
+      du -k {{ openshift.master.etcd_data_dir }} | tail -n 1 | cut -f1
+    register: etcd_disk_usage
+    when: embedded_etcd | bool
+
+  - name: Abort if insufficient disk space for etcd backup
+    fail: msg="{{ etcd_disk_usage.stdout }} Kb disk space required for etcd backup, {{ avail_disk.stdout }} Kb available."
+    when: (embedded_etcd | bool) and (etcd_disk_usage.stdout|int > avail_disk.stdout|int)
+  - name: Install etcd (for etcdctl)
+    yum: pkg=etcd state=latest
+  - name: Generate etcd backup
+    command: etcdctl backup --data-dir={{ openshift.master.etcd_data_dir }} --backup-dir={{ openshift.common.data_dir }}/etcd-backup-{{ timestamp }}
+  - name: Display location of etcd backup
+    debug: msg="Etcd backup created in {{ openshift.common.data_dir }}/etcd-backup-{{ timestamp }}"
+
 - name: Upgrade base package on masters
   hosts: masters
   roles:
@@ -9,22 +62,58 @@
     - name: Upgrade base package
       yum: pkg={{ openshift.common.service_type }}{{ openshift_version  }} state=latest
 
-- name: Re-Run cluster configuration to apply latest configuration changes
-  include: ../../common/openshift-cluster/config.yml
+- name: Evaluate oo_first_master
+  hosts: localhost
   vars:
-    g_etcd_group: "{{ 'etcd' }}"
     g_masters_group: "{{ 'masters' }}"
-    g_nodes_group: "{{ 'nodes' }}"
-    openshift_cluster_id: "{{ cluster_id | default('default') }}"
-    openshift_deployment_type: "{{ deployment_type }}"
+  tasks:
+    - name: Evaluate oo_first_master
+      add_host:
+        name: "{{ groups[g_masters_group][0] }}"
+        groups: oo_first_master
+        ansible_ssh_user: "{{ g_ssh_user | default(omit) }}"
+        ansible_sudo: "{{ g_sudo | default(omit) }}"
+      when: g_masters_group in groups and (groups[g_masters_group] | length) > 0
+
+# TODO: ideally we would check the new version, without installing it. (some
+# kind of yum repoquery? would need to handle openshift -> atomic-openshift
+# package rename)
+- name: Perform upgrade version checking
+  hosts: oo_first_master
+  tasks:
+    - name: Determine new version
+      command: >
+        rpm -q --queryformat '%{version}' {{ openshift.common.service_type }}
+      register: _new_version
+
+- name: Ensure AOS 3.0.2 or Origin 1.0.6
+  hosts: oo_first_master
+  tasks:
+    fail: This playbook requires Origin 1.0.6 or Atomic OpenShift 3.0.2 or later
+    when: _new_version.stdout | version_compare('1.0.6','<') or ( _new_version.stdout | version_compare('3.0','>=' and _new_version.stdout | version_compare('3.0.2','<') )
+
+- name: Verify upgrade can proceed
+  hosts: oo_first_master
+  tasks:
+  # Checking the global deployment type rather than host facts, this is about
+  # what the user is requesting.
+  - fail: msg="Deployment type 'enterprise' must be updated to 'openshift-enterprise' for upgrade to proceed"
+    when: deployment_type == "enterprise" and (_new_version.stdout | version_compare('1.0.7', '>=') or _new_version.stdout | version_compare('3.1', '>='))
 
 - name: Upgrade masters
   hosts: masters
   vars:
     openshift_version: "{{ openshift_pkg_version | default('') }}"
   tasks:
+    - name: Upgrade to latest available kernel
+      yum: pkg=kernel state=latest
+    - name: display just the deployment_type variable for the current host
+      debug:
+        var: hostvars[inventory_hostname]
     - name: Upgrade master packages
-      yum: pkg={{ openshift.common.service_type }}-master{{ openshift_version }} state=latest
+      command: yum update -y {{ openshift.common.service_type }}-master{{ openshift_version }}
+    - name: Upgrade master configuration.
+      openshift_upgrade_config: from_version=3.0 to_version=3.1 role=master
     - name: Restart master services
       service: name="{{ openshift.common.service_type}}-master" state=restarted
 
@@ -32,26 +121,14 @@
   hosts: nodes
   vars:
     openshift_version: "{{ openshift_pkg_version | default('') }}"
+  roles:
+  - openshift_facts
   tasks:
     - name: Upgrade node packages
-      yum: pkg={{ openshift.common.service_type }}-node{{ openshift_version }} state=latest
+      command: yum update -y {{ openshift.common.service_type }}-node{{ openshift_version }}
     - name: Restart node services
       service: name="{{ openshift.common.service_type }}-node" state=restarted
 
-- name: Determine new master version
-  hosts: oo_first_master
-  tasks:
-    - name: Determine new version
-      command: >
-        rpm -q --queryformat '%{version}' {{ openshift.common.service_type }}-master
-      register: _new_version
-
-- name: Ensure AOS 3.0.2 or Origin 1.0.6
-  hosts: oo_first_master
-  tasks:
-    fail: This playbook requires Origin 1.0.6 or Atomic OpenShift 3.0.2 or later
-    when: _new_version.stdout | version_compare('1.0.6','<') or ( _new_version.stdout | version_compare('3.0','>=' and _new_version.stdout | version_compare('3.0.2','<') )
-
 - name: Update cluster policy
   hosts: oo_first_master
   tasks: