Merge pull request #2993 from rhcarvalho/success-preflight

Add RPM checks as a byo playbook

16 files changed, 560 insertions, 3 deletions

diff --git a/playbooks/byo/openshift-preflight/README.md b/playbooks/byo/openshift-preflight/README.md
new file mode 100644
index 000000000..b50292eac
--- /dev/null
+++ b/playbooks/byo/openshift-preflight/README.md
@@ -0,0 +1,43 @@
+# OpenShift preflight checks
+
+Here we provide an Ansible playbook for detecting potential roadblocks prior to
+an install or upgrade.
+
+Ansible's default operation mode is to fail fast, on the first error. However,
+when performing checks, it is useful to gather as much information about
+problems as possible in a single run.
+
+The `check.yml` playbook runs a battery of checks against the inventory hosts
+and tells Ansible to ignore intermediate errors, thus giving a more complete
+diagnostic of the state of each host. Still, if any check failed, the playbook
+run will be marked as having failed.
+
+To facilitate understanding the problems that were encountered, we provide a
+custom callback plugin to summarize execution errors at the end of a playbook
+run.
+
+---
+
+*Note that currently the `check.yml` playbook is only useful for RPM-based
+installations. Containerized installs are excluded from checks for now, but
+might be included in the future if there is demand for that.*
+
+---
+
+## Running
+
+With an installation of Ansible 2.2 or greater, run the playbook directly
+against your inventory file. Here is the step-by-step:
+
+1. If you haven't done it yet, clone this repository:
+
+    ```console
+    $ git clone https://github.com/openshift/openshift-ansible
+    $ cd openshift-ansible
+    ```
+
+2. Run the playbook:
+
+    ```console
+    $ ansible-playbook -i <inventory file> playbooks/byo/openshift-preflight/check.yml
+    ```
diff --git a/playbooks/byo/openshift-preflight/check.yml b/playbooks/byo/openshift-preflight/check.yml
new file mode 100644
index 000000000..32673d01d
--- /dev/null
+++ b/playbooks/byo/openshift-preflight/check.yml
@@ -0,0 +1,31 @@
+---
+- hosts: OSEv3
+  roles:
+    - openshift_preflight/init
+
+- hosts: OSEv3
+  name: checks that apply to all hosts
+  gather_facts: no
+  ignore_errors: yes
+  roles:
+    - openshift_preflight/common
+
+- hosts: masters
+  name: checks that apply to masters
+  gather_facts: no
+  ignore_errors: yes
+  roles:
+    - openshift_preflight/masters
+
+- hosts: nodes
+  name: checks that apply to nodes
+  gather_facts: no
+  ignore_errors: yes
+  roles:
+    - openshift_preflight/nodes
+
+- hosts: OSEv3
+  name: verify check results
+  gather_facts: no
+  roles:
+    - openshift_preflight/verify_status
diff --git a/roles/openshift_facts/library/openshift_facts.py b/roles/openshift_facts/library/openshift_facts.py
index 10e30f1c4..616b41c7b 100755
--- a/roles/openshift_facts/library/openshift_facts.py
+++ b/roles/openshift_facts/library/openshift_facts.py
@@ -1246,10 +1246,10 @@ def build_api_server_args(facts):
 def is_service_running(service):
     """ Queries systemd through dbus to see if the service is running """
     service_running = False
-    bus = SystemBus()
-    systemd = bus.get_object('org.freedesktop.systemd1', '/org/freedesktop/systemd1')
-    manager = Interface(systemd, dbus_interface='org.freedesktop.systemd1.Manager')
     try:
+        bus = SystemBus()
+        systemd = bus.get_object('org.freedesktop.systemd1', '/org/freedesktop/systemd1')
+        manager = Interface(systemd, dbus_interface='org.freedesktop.systemd1.Manager')
         service_unit = service if service.endswith('.service') else manager.GetUnit('{0}.service'.format(service))
         service_proxy = bus.get_object('org.freedesktop.systemd1', str(service_unit))
         service_properties = Interface(service_proxy, dbus_interface='org.freedesktop.DBus.Properties')
@@ -1258,6 +1258,8 @@ def is_service_running(service):
         if service_load_state == 'loaded' and service_active_state == 'active':
             service_running = True
     except DBusException:
+        # TODO: do not swallow exception, as it may be hiding useful debugging
+        # information.
         pass
 
     return service_running
diff --git a/roles/openshift_preflight/README.md b/roles/openshift_preflight/README.md
new file mode 100644
index 000000000..b6d3542d3
--- /dev/null
+++ b/roles/openshift_preflight/README.md
@@ -0,0 +1,52 @@
+OpenShift Preflight Checks
+==========================
+
+This role detects common problems prior to installing OpenShift.
+
+Requirements
+------------
+
+* Ansible 2.2+
+
+Role Variables
+--------------
+
+None
+
+Dependencies
+------------
+
+None
+
+Example Playbook
+----------------
+
+```yaml
+---
+- hosts: OSEv3
+  roles:
+    - openshift_preflight/init
+
+- hosts: OSEv3
+  name: checks that apply to all hosts
+  gather_facts: no
+  ignore_errors: yes
+  roles:
+    - openshift_preflight/common
+
+- hosts: OSEv3
+  name: verify check results
+  gather_facts: no
+  roles:
+    - openshift_preflight/verify_status
+```
+
+License
+-------
+
+Apache License Version 2.0
+
+Author Information
+------------------
+
+Customer Success team (dev@lists.openshift.redhat.com)
diff --git a/roles/openshift_preflight/base/library/aos_version.py b/roles/openshift_preflight/base/library/aos_version.py
new file mode 100755
index 000000000..f7fcb6da5
--- /dev/null
+++ b/roles/openshift_preflight/base/library/aos_version.py
@@ -0,0 +1,100 @@
+#!/usr/bin/python
+# vim: expandtab:tabstop=4:shiftwidth=4
+'''
+An ansible module for determining if more than one minor version
+of any atomic-openshift package is available, which would indicate
+that multiple repos are enabled for different versions of the same
+thing which may cause problems.
+
+Also, determine if the version requested is available down to the
+precision requested.
+'''
+
+# import os
+# import sys
+import yum  # pylint: disable=import-error
+from ansible.module_utils.basic import AnsibleModule
+
+
+def main():  # pylint: disable=missing-docstring
+    module = AnsibleModule(
+        argument_spec=dict(
+            version=dict(required=True)
+        ),
+        supports_check_mode=True
+    )
+
+    # NOTE(rhcarvalho): sosiouxme added _unmute, but I couldn't find a case yet
+    # for when it is actually necessary. Leaving it commented out for now,
+    # though this comment and the commented out code related to _unmute should
+    # be deleted later if not proven necessary.
+
+    # sys.stdout = os.devnull  # mute yum so it doesn't break our output
+    # sys.stderr = os.devnull  # mute yum so it doesn't break our output
+
+    # def _unmute():  # pylint: disable=missing-docstring
+    #     sys.stdout = sys.__stdout__
+
+    def bail(error):  # pylint: disable=missing-docstring
+        # _unmute()
+        module.fail_json(msg=error)
+
+    yb = yum.YumBase()  # pylint: disable=invalid-name
+
+    # search for package versions available for aos pkgs
+    expected_pkgs = [
+        'atomic-openshift',
+        'atomic-openshift-master',
+        'atomic-openshift-node',
+    ]
+    try:
+        pkgs = yb.pkgSack.returnPackages(patterns=expected_pkgs)
+    except yum.Errors.PackageSackError as e:  # pylint: disable=invalid-name
+        # you only hit this if *none* of the packages are available
+        bail('Unable to find any atomic-openshift packages. \nCheck your subscription and repo settings. \n%s' % e)
+
+    # determine what level of precision we're expecting for the version
+    expected_version = module.params['version']
+    if expected_version.startswith('v'):  # v3.3 => 3.3
+        expected_version = expected_version[1:]
+    num_dots = expected_version.count('.')
+
+    pkgs_by_name_version = {}
+    pkgs_precise_version_found = {}
+    for pkg in pkgs:
+        # get expected version precision
+        match_version = '.'.join(pkg.version.split('.')[:num_dots + 1])
+        if match_version == expected_version:
+            pkgs_precise_version_found[pkg.name] = True
+        # get x.y version precision
+        minor_version = '.'.join(pkg.version.split('.')[:2])
+        if pkg.name not in pkgs_by_name_version:
+            pkgs_by_name_version[pkg.name] = {}
+        pkgs_by_name_version[pkg.name][minor_version] = True
+
+    # see if any packages couldn't be found at requested version
+    # see if any packages are available in more than one minor version
+    not_found = []
+    multi_found = []
+    for name in expected_pkgs:
+        if name not in pkgs_precise_version_found:
+            not_found.append(name)
+        if name in pkgs_by_name_version and len(pkgs_by_name_version[name]) > 1:
+            multi_found.append(name)
+    if not_found:
+        msg = 'Not all of the required packages are available at requested version %s:\n' % expected_version
+        for name in not_found:
+            msg += '  %s\n' % name
+        bail(msg + 'Please check your subscriptions and enabled repositories.')
+    if multi_found:
+        msg = 'Multiple minor versions of these packages are available\n'
+        for name in multi_found:
+            msg += '  %s\n' % name
+        bail(msg + "There should only be one OpenShift version's repository enabled at a time.")
+
+    # _unmute()
+    module.exit_json(changed=False)
+
+
+if __name__ == '__main__':
+    main()
diff --git a/roles/openshift_preflight/base/library/check_yum_update.py b/roles/openshift_preflight/base/library/check_yum_update.py
new file mode 100755
index 000000000..296ebd44f
--- /dev/null
+++ b/roles/openshift_preflight/base/library/check_yum_update.py
@@ -0,0 +1,116 @@
+#!/usr/bin/python
+# vim: expandtab:tabstop=4:shiftwidth=4
+'''
+Ansible module to test whether a yum update or install will succeed,
+without actually performing it or running yum.
+parameters:
+  packages: (optional) A list of package names to install or update.
+            If omitted, all installed RPMs are considered for updates.
+'''
+
+# import os
+import sys
+import yum  # pylint: disable=import-error
+from ansible.module_utils.basic import AnsibleModule
+
+
+def main():  # pylint: disable=missing-docstring,too-many-branches
+    module = AnsibleModule(
+        argument_spec=dict(
+            packages=dict(type='list', default=[])
+        ),
+        supports_check_mode=True
+    )
+
+    # NOTE(rhcarvalho): sosiouxme added _unmute, but I couldn't find a case yet
+    # for when it is actually necessary. Leaving it commented out for now,
+    # though this comment and the commented out code related to _unmute should
+    # be deleted later if not proven necessary.
+
+    # sys.stdout = os.devnull  # mute yum so it doesn't break our output
+
+    # def _unmute():  # pylint: disable=missing-docstring
+    #     sys.stdout = sys.__stdout__
+
+    def bail(error):  # pylint: disable=missing-docstring
+        # _unmute()
+        module.fail_json(msg=error)
+
+    yb = yum.YumBase()  # pylint: disable=invalid-name
+    # determine if the existing yum configuration is valid
+    try:
+        yb.repos.populateSack(mdtype='metadata', cacheonly=1)
+    # for error of type:
+    #   1. can't reach the repo URL(s)
+    except yum.Errors.NoMoreMirrorsRepoError as e:  # pylint: disable=invalid-name
+        bail('Error getting data from at least one yum repository: %s' % e)
+    #   2. invalid repo definition
+    except yum.Errors.RepoError as e:  # pylint: disable=invalid-name
+        bail('Error with yum repository configuration: %s' % e)
+    #   3. other/unknown
+    #    * just report the problem verbatim
+    except:  # pylint: disable=bare-except; # noqa
+        bail('Unexpected error with yum repository: %s' % sys.exc_info()[1])
+
+    packages = module.params['packages']
+    no_such_pkg = []
+    for pkg in packages:
+        try:
+            yb.install(name=pkg)
+        except yum.Errors.InstallError as e:  # pylint: disable=invalid-name
+            no_such_pkg.append(pkg)
+        except:  # pylint: disable=bare-except; # noqa
+            bail('Unexpected error with yum install/update: %s' %
+                 sys.exc_info()[1])
+    if not packages:
+        # no packages requested means test a yum update of everything
+        yb.update()
+    elif no_such_pkg:
+        # wanted specific packages to install but some aren't available
+        user_msg = 'Cannot install all of the necessary packages. Unavailable:\n'
+        for pkg in no_such_pkg:
+            user_msg += '  %s\n' % pkg
+        user_msg += 'You may need to enable one or more yum repositories to make this content available.'
+        bail(user_msg)
+
+    try:
+        txn_result, txn_msgs = yb.buildTransaction()
+    except:  # pylint: disable=bare-except; # noqa
+        bail('Unexpected error during dependency resolution for yum update: \n %s' %
+             sys.exc_info()[1])
+
+    # find out if there are any errors with the update/install
+    if txn_result == 0:  # 'normal exit' meaning there's nothing to install/update
+        pass
+    elif txn_result == 1:  # error with transaction
+        user_msg = 'Could not perform a yum update.\n'
+        if len(txn_msgs) > 0:
+            user_msg += 'Errors from dependency resolution:\n'
+            for msg in txn_msgs:
+                user_msg += '  %s\n' % msg
+            user_msg += 'You should resolve these issues before proceeding with an install.\n'
+            user_msg += 'You may need to remove or downgrade packages or enable/disable yum repositories.'
+        bail(user_msg)
+    # TODO: it would be nice depending on the problem:
+    #   1. dependency for update not found
+    #    * construct the dependency tree
+    #    * find the installed package(s) that required the missing dep
+    #    * determine if any of these packages matter to openshift
+    #    * build helpful error output
+    #   2. conflicts among packages in available content
+    #    * analyze dependency tree and build helpful error output
+    #   3. other/unknown
+    #    * report the problem verbatim
+    #    * add to this list as we come across problems we can clearly diagnose
+    elif txn_result == 2:  # everything resolved fine
+        pass
+    else:
+        bail('Unknown error(s) from dependency resolution. Exit Code: %d:\n%s' %
+             (txn_result, txn_msgs))
+
+    # _unmute()
+    module.exit_json(changed=False)
+
+
+if __name__ == '__main__':
+    main()
diff --git a/roles/openshift_preflight/common/meta/main.yml b/roles/openshift_preflight/common/meta/main.yml
new file mode 100644
index 000000000..6f23cbf3b
--- /dev/null
+++ b/roles/openshift_preflight/common/meta/main.yml
@@ -0,0 +1,3 @@
+---
+dependencies:
+  - role: openshift_preflight/base
diff --git a/roles/openshift_preflight/common/tasks/main.yml b/roles/openshift_preflight/common/tasks/main.yml
new file mode 100644
index 000000000..f1a4a160e
--- /dev/null
+++ b/roles/openshift_preflight/common/tasks/main.yml
@@ -0,0 +1,21 @@
+---
+# check content available on all hosts
+- when: not openshift.common.is_containerized | bool
+  block:
+
+    - name: determine if yum update will work
+      action: check_yum_update
+      register: r
+
+    - set_fact:
+        oo_preflight_check_results: "{{ oo_preflight_check_results + [r|combine({'_task': 'determine if yum update will work'})] }}"
+
+    - name: determine if expected version matches what is available
+      aos_version:
+        version: "{{ openshift_release }}"
+      when:
+        - deployment_type == "openshift-enterprise"
+      register: r
+
+    - set_fact:
+        oo_preflight_check_results: "{{ oo_preflight_check_results + [r|combine({'_task': 'determine if expected version matches what is available'})] }}"
diff --git a/roles/openshift_preflight/init/meta/main.yml b/roles/openshift_preflight/init/meta/main.yml
new file mode 100644
index 000000000..0bbeadd34
--- /dev/null
+++ b/roles/openshift_preflight/init/meta/main.yml
@@ -0,0 +1,3 @@
+---
+dependencies:
+  - role: openshift_facts
diff --git a/roles/openshift_preflight/init/tasks/main.yml b/roles/openshift_preflight/init/tasks/main.yml
new file mode 100644
index 000000000..bf2d82196
--- /dev/null
+++ b/roles/openshift_preflight/init/tasks/main.yml
@@ -0,0 +1,4 @@
+---
+- name: set common variables
+  set_fact:
+    oo_preflight_check_results: "{{ oo_preflight_check_results | default([]) }}"
diff --git a/roles/openshift_preflight/masters/meta/main.yml b/roles/openshift_preflight/masters/meta/main.yml
new file mode 100644
index 000000000..6f23cbf3b
--- /dev/null
+++ b/roles/openshift_preflight/masters/meta/main.yml
@@ -0,0 +1,3 @@
+---
+dependencies:
+  - role: openshift_preflight/base
diff --git a/roles/openshift_preflight/masters/tasks/main.yml b/roles/openshift_preflight/masters/tasks/main.yml
new file mode 100644
index 000000000..35fb1e3ca
--- /dev/null
+++ b/roles/openshift_preflight/masters/tasks/main.yml
@@ -0,0 +1,31 @@
+---
+# determine if yum install of master pkgs will work
+- when: not openshift.common.is_containerized | bool
+  block:
+
+    - name: main master packages availability
+      check_yum_update:
+        packages:
+          - "{{ openshift.common.service_type }}"
+          - "{{ openshift.common.service_type }}-clients"
+          - "{{ openshift.common.service_type }}-master"
+      register: r
+
+    - set_fact:
+        oo_preflight_check_results: "{{ oo_preflight_check_results + [r|combine({'_task': 'main master packages availability'})] }}"
+
+    - name: other master packages availability
+      check_yum_update:
+        packages:
+          - etcd
+          - bash-completion
+          - cockpit-bridge
+          - cockpit-docker
+          - cockpit-kubernetes
+          - cockpit-shell
+          - cockpit-ws
+          - httpd-tools
+      register: r
+
+    - set_fact:
+        oo_preflight_check_results: "{{ oo_preflight_check_results + [r|combine({'_task': 'other master packages availability'})] }}"
diff --git a/roles/openshift_preflight/nodes/meta/main.yml b/roles/openshift_preflight/nodes/meta/main.yml
new file mode 100644
index 000000000..6f23cbf3b
--- /dev/null
+++ b/roles/openshift_preflight/nodes/meta/main.yml
@@ -0,0 +1,3 @@
+---
+dependencies:
+  - role: openshift_preflight/base
diff --git a/roles/openshift_preflight/nodes/tasks/main.yml b/roles/openshift_preflight/nodes/tasks/main.yml
new file mode 100644
index 000000000..a10e69024
--- /dev/null
+++ b/roles/openshift_preflight/nodes/tasks/main.yml
@@ -0,0 +1,41 @@
+---
+# determine if yum install of node pkgs will work
+- when: not openshift.common.is_containerized | bool
+  block:
+
+    - name: main node packages availability
+      check_yum_update:
+        packages:
+          - "{{ openshift.common.service_type }}"
+          - "{{ openshift.common.service_type }}-node"
+          - "{{ openshift.common.service_type }}-sdn-ovs"
+      register: r
+
+    - set_fact:
+        oo_preflight_check_results: "{{ oo_preflight_check_results + [r|combine({'_task': 'main node packages availability'})] }}"
+
+    - name: other node packages availability
+      check_yum_update:
+        packages:
+          - docker
+          - PyYAML
+          - firewalld
+          - iptables
+          - iptables-services
+          - nfs-utils
+          - ntp
+          - yum-utils
+          - dnsmasq
+          - libselinux-python
+          - ceph-common
+          - glusterfs-fuse
+          - iscsi-initiator-utils
+          - pyparted
+          - python-httplib2
+          - openssl
+          - flannel
+          - bind
+      register: r
+
+    - set_fact:
+        oo_preflight_check_results: "{{ oo_preflight_check_results + [r|combine({'_task': 'other node packages availability'})] }}"
diff --git a/roles/openshift_preflight/verify_status/callback_plugins/zz_failure_summary.py b/roles/openshift_preflight/verify_status/callback_plugins/zz_failure_summary.py
new file mode 100644
index 000000000..180ed8d8f
--- /dev/null
+++ b/roles/openshift_preflight/verify_status/callback_plugins/zz_failure_summary.py
@@ -0,0 +1,96 @@
+# vim: expandtab:tabstop=4:shiftwidth=4
+'''
+Ansible callback plugin.
+'''
+
+from ansible.plugins.callback import CallbackBase
+from ansible import constants as C
+from ansible.utils.color import stringc
+
+
+class CallbackModule(CallbackBase):
+    '''
+    This callback plugin stores task results and summarizes failures.
+    The file name is prefixed with `zz_` to make this plugin be loaded last by
+    Ansible, thus making its output the last thing that users see.
+    '''
+
+    CALLBACK_VERSION = 2.0
+    CALLBACK_TYPE = 'aggregate'
+    CALLBACK_NAME = 'failure_summary'
+    CALLBACK_NEEDS_WHITELIST = False
+
+    def __init__(self):
+        super(CallbackModule, self).__init__()
+        self.__failures = []
+
+    def v2_runner_on_failed(self, result, ignore_errors=False):
+        super(CallbackModule, self).v2_runner_on_failed(result, ignore_errors)
+        self.__failures.append(dict(result=result, ignore_errors=ignore_errors))
+
+    def v2_playbook_on_stats(self, stats):
+        super(CallbackModule, self).v2_playbook_on_stats(stats)
+        # TODO: update condition to consider a host var or env var to
+        # enable/disable the summary, so that we can control the output from a
+        # play.
+        if self.__failures:
+            self._print_failure_summary()
+
+    def _print_failure_summary(self):
+        '''Print a summary of failed tasks (including ignored failures).'''
+        self._display.display(u'\nFailure summary:\n')
+
+        # TODO: group failures by host or by task. If grouped by host, it is
+        # easy to see all problems of a given host. If grouped by task, it is
+        # easy to see what hosts needs the same fix.
+
+        width = len(str(len(self.__failures)))
+        initial_indent_format = u'  {{:>{width}}}. '.format(width=width)
+        initial_indent_len = len(initial_indent_format.format(0))
+        subsequent_indent = u' ' * initial_indent_len
+        subsequent_extra_indent = u' ' * (initial_indent_len + 10)
+
+        for i, failure in enumerate(self.__failures, 1):
+            lines = _format_failure(failure)
+            self._display.display(u'\n{}{}'.format(initial_indent_format.format(i), lines[0]))
+            for line in lines[1:]:
+                line = line.replace(u'\n', u'\n' + subsequent_extra_indent)
+                indented = u'{}{}'.format(subsequent_indent, line)
+                self._display.display(indented)
+
+
+# Reason: disable pylint protected-access because we need to access _*
+#         attributes of a task result to implement this method.
+# Status: permanently disabled unless Ansible's API changes.
+# pylint: disable=protected-access
+def _format_failure(failure):
+    '''Return a list of pretty-formatted lines describing a failure, including
+    relevant information about it. Line separators are not included.'''
+    result = failure['result']
+    host = result._host.get_name()
+    play = _get_play(result._task)
+    if play:
+        play = play.get_name()
+    task = result._task.get_name()
+    msg = result._result.get('msg', u'???')
+    rows = (
+        (u'Host', host),
+        (u'Play', play),
+        (u'Task', task),
+        (u'Message', stringc(msg, C.COLOR_ERROR)),
+    )
+    row_format = '{:10}{}'
+    return [row_format.format(header + u':', body) for header, body in rows]
+
+
+# Reason: disable pylint protected-access because we need to access _*
+#         attributes of obj to implement this function.
+#         This is inspired by ansible.playbook.base.Base.dump_me.
+# Status: permanently disabled unless Ansible's API changes.
+# pylint: disable=protected-access
+def _get_play(obj):
+    '''Given a task or block, recursively tries to find its parent play.'''
+    if hasattr(obj, '_play'):
+        return obj._play
+    if getattr(obj, '_parent'):
+        return _get_play(obj._parent)
diff --git a/roles/openshift_preflight/verify_status/tasks/main.yml b/roles/openshift_preflight/verify_status/tasks/main.yml
new file mode 100644
index 000000000..36ccf648a
--- /dev/null
+++ b/roles/openshift_preflight/verify_status/tasks/main.yml
@@ -0,0 +1,8 @@
+---
+- name: find check failures
+  set_fact:
+    oo_preflight_check_failures: "{{ oo_preflight_check_results | select('failed', 'equalto', True) | list }}"
+
+- name: ensure all checks succeed
+  action: fail
+  when: oo_preflight_check_failures