Merge pull request #5459 from zgalor/nfs_fix

Fix prometheus role nfs

12 files changed, 90 insertions, 166 deletions

diff --git a/roles/openshift_facts/library/openshift_facts.py b/roles/openshift_facts/library/openshift_facts.py
index ba1d8f29d..33028fea4 100755
--- a/roles/openshift_facts/library/openshift_facts.py
+++ b/roles/openshift_facts/library/openshift_facts.py
@@ -498,6 +498,20 @@ def set_selectors(facts):
         facts['hosted']['etcd'] = {}
     if 'selector' not in facts['hosted']['etcd'] or facts['hosted']['etcd']['selector'] in [None, 'None']:
         facts['hosted']['etcd']['selector'] = None
+    if 'prometheus' not in facts:
+        facts['prometheus'] = {}
+    if 'selector' not in facts['prometheus'] or facts['prometheus']['selector'] in [None, 'None']:
+        facts['prometheus']['selector'] = None
+    if 'alertmanager' not in facts['prometheus']:
+        facts['prometheus']['alertmanager'] = {}
+    # pylint: disable=line-too-long
+    if 'selector' not in facts['prometheus']['alertmanager'] or facts['prometheus']['alertmanager']['selector'] in [None, 'None']:
+        facts['prometheus']['alertmanager']['selector'] = None
+    if 'alertbuffer' not in facts['prometheus']:
+        facts['prometheus']['alertbuffer'] = {}
+    # pylint: disable=line-too-long
+    if 'selector' not in facts['prometheus']['alertbuffer'] or facts['prometheus']['alertbuffer']['selector'] in [None, 'None']:
+        facts['prometheus']['alertbuffer']['selector'] = None
 
     return facts
 
@@ -1779,7 +1793,8 @@ class OpenShiftFacts(object):
                    'node',
                    'logging',
                    'loggingops',
-                   'metrics']
+                   'metrics',
+                   'prometheus']
 
     # Disabling too-many-arguments, this should be cleaned up as a TODO item.
     # pylint: disable=too-many-arguments,no-value-for-parameter
@@ -2068,6 +2083,66 @@ class OpenShiftFacts(object):
                 )
             )
 
+            defaults['prometheus'] = dict(
+                storage=dict(
+                    kind=None,
+                    volume=dict(
+                        name='prometheus',
+                        size='10Gi'
+                    ),
+                    nfs=dict(
+                        directory='/exports',
+                        options='*(rw,root_squash)'
+                    ),
+                    host=None,
+                    access=dict(
+                        modes=['ReadWriteOnce']
+                    ),
+                    create_pv=True,
+                    create_pvc=False
+                )
+            )
+
+            defaults['prometheus']['alertmanager'] = dict(
+                storage=dict(
+                    kind=None,
+                    volume=dict(
+                        name='prometheus-alertmanager',
+                        size='10Gi'
+                    ),
+                    nfs=dict(
+                        directory='/exports',
+                        options='*(rw,root_squash)'
+                    ),
+                    host=None,
+                    access=dict(
+                        modes=['ReadWriteOnce']
+                    ),
+                    create_pv=True,
+                    create_pvc=False
+                )
+            )
+
+            defaults['prometheus']['alertbuffer'] = dict(
+                storage=dict(
+                    kind=None,
+                    volume=dict(
+                        name='prometheus-alertbuffer',
+                        size='10Gi'
+                    ),
+                    nfs=dict(
+                        directory='/exports',
+                        options='*(rw,root_squash)'
+                    ),
+                    host=None,
+                    access=dict(
+                        modes=['ReadWriteOnce']
+                    ),
+                    create_pv=True,
+                    create_pvc=False
+                )
+            )
+
         return defaults
 
     def guess_host_provider(self):
diff --git a/roles/openshift_hosted_facts/tasks/main.yml b/roles/openshift_hosted_facts/tasks/main.yml
index 47dc9171d..8fc70cecb 100644
--- a/roles/openshift_hosted_facts/tasks/main.yml
+++ b/roles/openshift_hosted_facts/tasks/main.yml
@@ -16,4 +16,4 @@
                        | oo_openshift_env }}"
     openshift_env_structures:
     - 'openshift.hosted.router.*'
-  with_items: [hosted, logging, loggingops, metrics]
+  with_items: [hosted, logging, loggingops, metrics, prometheus]
diff --git a/roles/openshift_prometheus/defaults/main.yaml b/roles/openshift_prometheus/defaults/main.yaml
index 5aa8aecec..ab3d4e31e 100644
--- a/roles/openshift_prometheus/defaults/main.yaml
+++ b/roles/openshift_prometheus/defaults/main.yaml
@@ -10,50 +10,30 @@ openshift_prometheus_node_selector: {"region":"infra"}
 # images
 openshift_prometheus_image_proxy: "openshift/oauth-proxy:v1.0.0"
 openshift_prometheus_image_prometheus: "openshift/prometheus:v2.0.0-dev"
-openshift_prometheus_image_alertmanager: "openshift/prometheus-alertmanager:dev"
+openshift_prometheus_image_alertmanager: "openshift/prometheus-alertmanager:v0.9.1"
 openshift_prometheus_image_alertbuffer: "openshift/prometheus-alert-buffer:v0.0.1"
 
 # additional prometheus rules file
 openshift_prometheus_additional_rules_file: null
 
-# All the required exports
-openshift_prometheus_pv_exports:
-  - prometheus
-  - prometheus-alertmanager
-  - prometheus-alertbuffer
-# PV template files and their created object names
-openshift_prometheus_pv_data:
-  - pv_name: prometheus
-    pv_template: prom-pv-server.yml
-    pv_label: Prometheus Server PV
-  - pv_name: prometheus-alertmanager
-    pv_template: prom-pv-alertmanager.yml
-    pv_label: Prometheus Alertmanager PV
-  - pv_name: prometheus-alertbuffer
-    pv_template: prom-pv-alertbuffer.yml
-    pv_label: Prometheus Alert Buffer PV
-
-# Hostname/IP of the NFS server. Currently defaults to first master
-openshift_prometheus_nfs_server: "{{ groups.nfs.0 }}"
-
 # storage
 openshift_prometheus_storage_type: pvc
 openshift_prometheus_pvc_name: prometheus
-openshift_prometheus_pvc_size: 10G
+openshift_prometheus_pvc_size: "{{ openshift_prometheus_storage_volume_size | default('10Gi') }}"
 openshift_prometheus_pvc_access_modes: [ReadWriteOnce]
-openshift_prometheus_pvc_pv_selector: {}
+openshift_prometheus_pvc_pv_selector: "{{ openshift_prometheus_storage_labels | default('') }}"
 
 openshift_prometheus_alertmanager_storage_type: pvc
 openshift_prometheus_alertmanager_pvc_name: prometheus-alertmanager
-openshift_prometheus_alertmanager_pvc_size: 10G
+openshift_prometheus_alertmanager_pvc_size: "{{ openshift_prometheus_alertmanager_storage_volume_size | default('10Gi') }}"
 openshift_prometheus_alertmanager_pvc_access_modes: [ReadWriteOnce]
-openshift_prometheus_alertmanager_pvc_pv_selector: {}
+openshift_prometheus_alertmanager_pvc_pv_selector: "{{ openshift_prometheus_alertmanager_storage_labels | default('') }}"
 
 openshift_prometheus_alertbuffer_storage_type: pvc
 openshift_prometheus_alertbuffer_pvc_name: prometheus-alertbuffer
-openshift_prometheus_alertbuffer_pvc_size: 10G
+openshift_prometheus_alertbuffer_pvc_size: "{{ openshift_prometheus_alertbuffer_storage_volume_size | default('10Gi') }}"
 openshift_prometheus_alertbuffer_pvc_access_modes: [ReadWriteOnce]
-openshift_prometheus_alertbuffer_pvc_pv_selector: {}
+openshift_prometheus_alertbuffer_pvc_pv_selector: "{{ openshift_prometheus_alertbuffer_storage_labels | default('') }}"
 
 # container resources
 openshift_prometheus_cpu_limit: null
diff --git a/roles/openshift_prometheus/files/openshift_prometheus.exports b/roles/openshift_prometheus/files/openshift_prometheus.exports
deleted file mode 100644
index 3ccedb1fd..000000000
--- a/roles/openshift_prometheus/files/openshift_prometheus.exports
+++ /dev/null
@@ -1,3 +0,0 @@
-/exports/prometheus *(rw,no_root_squash,no_wdelay)
-/exports/prometheus-alertmanager *(rw,no_root_squash,no_wdelay)
-/exports/prometheus-alertbuffer *(rw,no_root_squash,no_wdelay)
diff --git a/roles/openshift_prometheus/tasks/create_pvs.yaml b/roles/openshift_prometheus/tasks/create_pvs.yaml
deleted file mode 100644
index 4e79da05f..000000000
--- a/roles/openshift_prometheus/tasks/create_pvs.yaml
+++ /dev/null
@@ -1,36 +0,0 @@
----
-# Check for existance and then conditionally:
-# - evaluate templates
-# - PVs
-#
-# These tasks idempotently create required Prometheus PV objects. Do not
-# call this file directly. This file is intended to be ran as an
-# include that has a 'with_items' attached to it. Hence the use below
-# of variables like "{{ item.pv_label }}"
-
-- name: "Check if the {{ item.pv_label }} template has been created already"
-  oc_obj:
-    namespace: "{{ openshift_prometheus_namespace }}"
-    state: list
-    kind: pv
-    name: "{{ item.pv_name }}"
-  register: prom_pv_check
-
-# Skip all of this if the PV already exists
-- block:
-    - name: "Ensure the {{ item.pv_label }} template is evaluated"
-      template:
-        src: "{{ item.pv_template }}.j2"
-        dest: "{{ tempdir }}/templates/{{ item.pv_template }}"
-
-    - name: "Ensure {{ item.pv_label }} is created"
-      oc_obj:
-        namespace: "{{ openshift_prometheus_namespace }}"
-        kind: pv
-        name: "{{ item.pv_name }}"
-        state: present
-        delete_after: True
-        files:
-          - "{{ tempdir }}/templates/{{ item.pv_template }}"
-  when:
-    - not prom_pv_check.results.results.0
diff --git a/roles/openshift_prometheus/tasks/install_prometheus.yaml b/roles/openshift_prometheus/tasks/install_prometheus.yaml
index a9bce2fb1..cb75eedca 100644
--- a/roles/openshift_prometheus/tasks/install_prometheus.yaml
+++ b/roles/openshift_prometheus/tasks/install_prometheus.yaml
@@ -54,15 +54,6 @@
     resource_name: cluster-reader
     user: "system:serviceaccount:{{ openshift_prometheus_namespace }}:prometheus"
 
-
-######################################################################
-# NFS
-# In the case that we are not running on a cloud provider, volumes must be statically provisioned
-
-- include: nfs.yaml
-  when: not (openshift_cloudprovider_kind is defined and (openshift_cloudprovider_kind == 'aws' or openshift_cloudprovider_kind == 'gce'))
-
-
 # create prometheus and alerts services
 # TODO join into 1 task with loop
 - name: Create prometheus service
diff --git a/roles/openshift_prometheus/tasks/nfs.yaml b/roles/openshift_prometheus/tasks/nfs.yaml
deleted file mode 100644
index 0b45f2cee..000000000
--- a/roles/openshift_prometheus/tasks/nfs.yaml
+++ /dev/null
@@ -1,44 +0,0 @@
----
-# Tasks to statically provision NFS volumes
-# Include if not using dynamic volume provisioning
-- name: Ensure the /exports/ directory exists
-  file:
-    path: /exports/
-    state: directory
-    mode: 0755
-    owner: root
-    group: root
-
-- name: Ensure the prom-pv0X export directories exist
-  file:
-    path: "/exports/{{ item }}"
-    state: directory
-    mode: 0777
-    owner: nfsnobody
-    group: nfsnobody
-  with_items: "{{ openshift_prometheus_pv_exports }}"
-
-- name: Ensure the NFS exports for Prometheus PVs exist
-  copy:
-    src: openshift_prometheus.exports
-    dest: /etc/exports.d/openshift_prometheus.exports
-  register: nfs_exports_updated
-
-- name: Ensure the NFS export table is refreshed if exports were added
-  command: exportfs -ar
-  when:
-    - nfs_exports_updated.changed
-
-
-######################################################################
-# Create the required Prometheus PVs. Check out these online docs if you
-# need a refresher on includes looping with items:
-# * http://docs.ansible.com/ansible/playbooks_loops.html#loops-and-includes-in-2-0
-# * http://stackoverflow.com/a/35128533
-#
-# TODO: Handle the case where a PV template is updated in
-# openshift-ansible and the change needs to be landed on the managed
-# cluster.
-
-- include: create_pvs.yaml
-  with_items: "{{ openshift_prometheus_pv_data }}"
diff --git a/roles/openshift_prometheus/templates/prom-pv-alertbuffer.yml.j2 b/roles/openshift_prometheus/templates/prom-pv-alertbuffer.yml.j2
deleted file mode 100644
index 55a5e19c3..000000000
--- a/roles/openshift_prometheus/templates/prom-pv-alertbuffer.yml.j2
+++ /dev/null
@@ -1,15 +0,0 @@
-apiVersion: v1
-kind: PersistentVolume
-metadata:
-  name: prometheus-alertbuffer
-  labels:
-    storage: prometheus-alertbuffer
-spec:
-  capacity:
-    storage: 15Gi
-  accessModes:
-    - ReadWriteOnce
-  nfs:
-    path: /exports/prometheus-alertbuffer
-    server: {{ openshift_prometheus_nfs_server }}
-  persistentVolumeReclaimPolicy: Retain
diff --git a/roles/openshift_prometheus/templates/prom-pv-alertmanager.yml.j2 b/roles/openshift_prometheus/templates/prom-pv-alertmanager.yml.j2
deleted file mode 100644
index 4ee518735..000000000
--- a/roles/openshift_prometheus/templates/prom-pv-alertmanager.yml.j2
+++ /dev/null
@@ -1,15 +0,0 @@
-apiVersion: v1
-kind: PersistentVolume
-metadata:
-  name: prometheus-alertmanager
-  labels:
-    storage: prometheus-alertmanager
-spec:
-  capacity:
-    storage: 15Gi
-  accessModes:
-    - ReadWriteOnce
-  nfs:
-    path: /exports/prometheus-alertmanager
-    server: {{ openshift_prometheus_nfs_server }}
-  persistentVolumeReclaimPolicy: Retain
diff --git a/roles/openshift_prometheus/templates/prom-pv-server.yml.j2 b/roles/openshift_prometheus/templates/prom-pv-server.yml.j2
deleted file mode 100644
index 933bf0f60..000000000
--- a/roles/openshift_prometheus/templates/prom-pv-server.yml.j2
+++ /dev/null
@@ -1,15 +0,0 @@
-apiVersion: v1
-kind: PersistentVolume
-metadata:
-  name: prometheus
-  labels:
-    storage: prometheus
-spec:
-  capacity:
-    storage: 15Gi
-  accessModes:
-    - ReadWriteOnce
-  nfs:
-    path: /exports/prometheus
-    server: {{ openshift_prometheus_nfs_server }}
-  persistentVolumeReclaimPolicy: Retain
diff --git a/roles/openshift_storage_nfs/tasks/main.yml b/roles/openshift_storage_nfs/tasks/main.yml
index 3047fbaf9..c4e023c1e 100644
--- a/roles/openshift_storage_nfs/tasks/main.yml
+++ b/roles/openshift_storage_nfs/tasks/main.yml
@@ -35,6 +35,9 @@
     - "{{ openshift.logging }}"
     - "{{ openshift.loggingops }}"
     - "{{ openshift.hosted.etcd }}"
+    - "{{ openshift.prometheus }}"
+    - "{{ openshift.prometheus.alertmanager }}"
+    - "{{ openshift.prometheus.alertbuffer }}"
 
 - name: Configure exports
   template:
diff --git a/roles/openshift_storage_nfs/templates/exports.j2 b/roles/openshift_storage_nfs/templates/exports.j2
index 0141e0d25..c2a741035 100644
--- a/roles/openshift_storage_nfs/templates/exports.j2
+++ b/roles/openshift_storage_nfs/templates/exports.j2
@@ -3,3 +3,6 @@
 {{ openshift.logging.storage.nfs.directory }}/{{ openshift.logging.storage.volume.name }} {{ openshift.logging.storage.nfs.options }}
 {{ openshift.loggingops.storage.nfs.directory }}/{{ openshift.loggingops.storage.volume.name }} {{ openshift.loggingops.storage.nfs.options }}
 {{ openshift.hosted.etcd.storage.nfs.directory }}/{{ openshift.hosted.etcd.storage.volume.name }} {{ openshift.hosted.etcd.storage.nfs.options }}
+{{ openshift.prometheus.storage.nfs.directory }}/{{ openshift.prometheus.storage.volume.name }} {{ openshift.prometheus.storage.nfs.options }}
+{{ openshift.prometheus.alertmanager.storage.nfs.directory }}/{{ openshift.prometheus.alertmanager.storage.volume.name }} {{ openshift.prometheus.alertmanager.storage.nfs.options }}
+{{ openshift.prometheus.alertbuffer.storage.nfs.directory }}/{{ openshift.prometheus.alertbuffer.storage.volume.name }} {{ openshift.prometheus.alertbuffer.storage.nfs.options }}