Add interface for retrieving alerting config from charmed AlertManager (

#14)

This commit allows the charm to retrieve a configuration string from the
alertmanager charm. The configuration string is a JSON string that this
charm converts to a dict that is then inserted to another dict that
ultimately gets serialized into a YAML-formatted prometheus config file.

While this charm is agnostic to the contents of the JSON string, in
AlertManager's current iteration, it uses the `kube_sd_configs` method
so that Prometheus can quickly detect changes in the number of
AlertManager pods without waiting for any Juju events.

.gitignore

@@ -1,5 +1,6 @@
 .idea
 .tox
+.orig
 *.egg-info
 **/__pycache__
 .coverage

metadata.yaml

@@ -13,6 +13,9 @@ series:
 provides:
     http-api:
         interface: prometheus-http-api
+requires:
+    alertmanager:
+        interface: prometheus-alerting-config
 resources:
     prometheus-image:
         type: oci-image

src/adapters/framework.py

@@ -68,7 +68,7 @@ class FrameworkAdapter:
     def __init__(self, framework):
         self._framework = framework
 
-    def am_i_leader(self):
+    def unit_is_leader(self):
         return self._framework.model.unit.is_leader()
 
     def get_app_name(self):
@@ -86,9 +86,15 @@ def get_image_meta(self, image_name):
     def get_model_name(self):
         return self._framework.model.name
 
+    def get_relations(self, relation_name):
+        return self._framework.model.relations[relation_name]
+
     def get_resources_repo(self):
         return self._framework.model.resources
 
+    def get_unit(self):
+        return self._framework.model.unit
+
     def get_unit_name(self):
         return self._framework.model.unit.name
 

src/charm.py

@@ -1,4 +1,5 @@
 #!/usr/bin/env python3
+import json
 import logging
 
 logger = logging.getLogger()
@@ -24,6 +25,7 @@
 )
 from adapters import k8s
 from exceptions import CharmError
+from interface_alertmanager import AlertManagerInterface
 from interface_http import PrometheusInterface
 
 
@@ -46,13 +48,15 @@ def __init__(self, *args):
         # adapter and not directly with the framework.
         self.fw_adapter = FrameworkAdapter(self.framework)
         self.prometheus = PrometheusInterface(self, 'http-api')
-
+        self.alertmanager = AlertManagerInterface(self, 'alertmanager')
         # Bind event handlers to events
         event_handler_bindings = {
             self.on.start: self.on_start,
             self.on.config_changed: self.on_config_changed,
             self.on.upgrade_charm: self.on_upgrade,
             self.on.stop: self.on_stop,
+            self.alertmanager.on.new_relation:
+                self.on_new_alertmanager_relation
         }
         for event, handler in event_handler_bindings.items():
             self.fw_adapter.observe(event, handler)
@@ -73,6 +77,9 @@ def __init__(self, *args):
     def on_config_changed(self, event):
         on_config_changed_handler(event, self.fw_adapter, self._stored)
 
+    def on_new_alertmanager_relation(self, event):
+        on_new_alertmanager_relation_handler(event, self.fw_adapter)
+
     def on_start(self, event):
         on_start_handler(event, self.fw_adapter)
 
@@ -113,6 +120,11 @@ def on_config_changed_handler(event, fw_adapter, state):
         time.sleep(1)
 
 
+def on_new_alertmanager_relation_handler(event, fw_adapter):
+    alerting_config = json.loads(event.data.get('alerting_config', '{}'))
+    set_juju_pod_spec(fw_adapter, alerting_config)
+
+
 def on_start_handler(event, fw_adapter):
     set_juju_pod_spec(fw_adapter)
 
@@ -125,17 +137,24 @@ def on_stop_handler(event, fw_adapter):
     fw_adapter.set_unit_status(MaintenanceStatus("Pod is terminating"))
 
 
-def set_juju_pod_spec(fw_adapter):
-    if not fw_adapter.am_i_leader():
+def set_juju_pod_spec(fw_adapter, alerting_config={}):
+    if not fw_adapter.unit_is_leader():
         logging.debug("Unit is not a leader, skip pod spec configuration")
         return
 
+    if alerting_config:
+        logger.debug(
+            "Got alerting config: {} {}".format(type(alerting_config),
+                                                alerting_config)
+        )
+
     logging.debug("Building Juju pod spec")
     try:
         juju_pod_spec = build_juju_pod_spec(
             app_name=fw_adapter.get_app_name(),
             charm_config=fw_adapter.get_config(),
-            image_meta=fw_adapter.get_image_meta('prometheus-image')
+            image_meta=fw_adapter.get_image_meta('prometheus-image'),
+            alerting_config=alerting_config
         )
     except CharmError as e:
         fw_adapter.set_unit_status(

src/domain.py

@@ -1,13 +1,15 @@
 import copy
 import json
-import yaml
 import logging
+import yaml
 
 logger = logging.getLogger()
 
 import sys
 sys.path.append('lib')
 
+logger = logging.getLogger()
+
 from ops.model import (
     ActiveStatus,
     MaintenanceStatus,
@@ -87,10 +89,11 @@ class PrometheusConfigFile:
     https://prometheus.io/docs/prometheus/latest/configuration/configuration
     '''
 
-    def __init__(self, global_opts):
+    def __init__(self, global_opts, alerting={}):
         self._config_dict = {
             'global': global_opts,
             'scrape_configs': [],
+            'alerting': alerting
         }
 
     def add_scrape_config(self, scrape_config):
@@ -102,6 +105,9 @@ def add_scrape_config(self, scrape_config):
     def yaml_dump(self):
         return yaml.dump(self._config_dict)
 
+    def __repr__(self):
+        return str(self._config_dict)
+
 
 # DOMAIN SERVICES
 
@@ -189,7 +195,8 @@ def build_prometheus_cli_args(charm_config):
 
 def build_juju_pod_spec(app_name,
                         charm_config,
-                        image_meta):
+                        image_meta,
+                        alerting_config={}):
 
     # There is never ever a need to customize the advertised port of a
     # containerized Prometheus instance so we are removing that config
@@ -340,4 +347,5 @@ def build_prometheus_config(charm_config):
         for scrape_config in k8s_scrape_configs:
             prometheus_config.add_scrape_config(scrape_config)
 
+    logger.debug("Build prom config: {}".format(prometheus_config))
     return prometheus_config

src/interface_alertmanager.py

@@ -0,0 +1,55 @@
+import json
+import logging
+
+logger = logging.getLogger()
+
+from ops.framework import (
+    EventSource,
+    Object,
+    ObjectEvents,
+)
+from ops.framework import EventBase
+from adapters.framework import FrameworkAdapter
+
+
+class NewAlertManagerRelationEvent(EventBase):
+
+    def __init__(self, handle, remote_data):
+        super().__init__(handle)
+        self.data = dict(remote_data)
+
+    # The Operator Framework will serialize and deserialize this event object
+    # as it passes it to the charm. The following snapshot and restore methos
+    # ensure that our underlying data don't get lost along the way.
+
+    def snapshot(self):
+        return json.dumps(self.data)
+
+    def restore(self, snapshot):
+        self.data = json.loads(snapshot)
+
+
+class AlertManagerEvents(ObjectEvents):
+    new_relation = EventSource(NewAlertManagerRelationEvent)
+
+
+class AlertManagerInterface(Object):
+    on = AlertManagerEvents()
+
+    def __init__(self, charm, relation_name):
+        super().__init__(charm, relation_name)
+
+        self.fw_adapter = FrameworkAdapter(self.framework)
+        self.relation_name = relation_name
+
+        self.fw_adapter.observe(charm.on[relation_name].relation_changed,
+                                self.on_relation_changed)
+
+    def on_relation_changed(self, event):
+        remote_data = event.relation.data[event.unit]
+        logging.debug(
+            "Received remote_data: {}".format(dict(remote_data))
+        )
+
+        logger.debug("Emitting new_relation event")
+        self.on.new_relation.emit(remote_data)