Merge ~barryprice/landscape-charm/+git/landscape-charm:main into landscape-charm:main

[Barry Price]

WIP, DO NOT MERGE. First pass at LP:2012693

Unmerged commits

fb5aef2... by Barry Price on 2023-03-25

WIP, DO NOT MERGE. First pass at LP:2012693

diff --git a/charmcraft.yaml b/charmcraft.yaml
index a93272d..95623a0 100644
--- a/charmcraft.yaml
+++ b/charmcraft.yaml
@@ -18,3 +18,6 @@ bases:
       architectures:
       - amd64
       - arm64
+parts:
+  charm:
+    build-packages: [git]
diff --git a/requirements.txt b/requirements.txt
index 96faf88..ae051de 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1 +1,2 @@
 ops >= 1.4.0
+-e git+https://github.com/canonical/ops-lib-nrpe.git#egg=ops-lib-nrpe
diff --git a/src/charm.py b/src/charm.py
index d883217..8ba6a64 100755
--- a/src/charm.py
+++ b/src/charm.py
@@ -26,6 +26,8 @@ from charms.operator_libs_linux.v0.apt import (
 from charms.operator_libs_linux.v0.passwd import group_exists, user_exists
 from charms.operator_libs_linux.v0.systemd import service_reload
 
+from nrpe.client import NRPEClient, NRPEAvailable
+
 from ops.charm import (
     ActionEvent, CharmBase, InstallEvent, LeaderElectedEvent,
     LeaderSettingsChangedEvent, RelationChangedEvent, RelationJoinedEvent,
@@ -47,7 +49,6 @@ DPKG_RECONFIGURE = "/usr/sbin/dpkg-reconfigure"
 HAPROXY_CONFIG_FILE = os.path.join(os.path.dirname(__file__),
                                    "haproxy-config.yaml")
 LSCTL = "/usr/bin/lsctl"
-NRPE_D_DIR = "/etc/nagios/nrpe.d"
 POSTFIX_CF = "/etc/postfix/main.cf"
 SCHEMA_SCRIPT = "/usr/bin/landscape-schema"
 BOOTSTRAP_ACCOUNT_SCRIPT = "/opt/canonical/landscape/bootstrap-account"
@@ -112,8 +113,10 @@ class LandscapeServerCharm(CharmBase):
                                self._website_relation_joined)
         self.framework.observe(self.on.website_relation_changed,
                                self._website_relation_changed)
-        self.framework.observe(self.on.nrpe_external_master_relation_joined,
-                               self._nrpe_external_master_relation_joined)
+
+        # NRPE
+        self.nrpe_client = NRPEClient(self, 'nrpe-external-master')
+        self.framework.observe(self.nrpe_client.on.nrpe_available, self._on_nrpe_available)
 
         # Leadership/peering
         self.framework.observe(self.on.leader_elected, self._leader_elected)
@@ -605,11 +608,7 @@ class LandscapeServerCharm(CharmBase):
 
         self._update_ready_status()
 
-    def _nrpe_external_master_relation_joined(
-            self, event: RelationJoinedEvent) -> None:
-        self._update_nrpe_checks(event.relation)
-
-    def _update_nrpe_checks(self, relation: Relation):
+    def _on_nrpe_available(self, event: NRPEAvailable) -> None:
         logger.debug("Configuring NRPE checks")
 
         if self.unit.is_leader():
@@ -619,47 +618,19 @@ class LandscapeServerCharm(CharmBase):
             services_to_add = DEFAULT_SERVICES
             services_to_remove = LEADER_SERVICES
 
-        monitors = {
-            "monitors": {
-                "remote": {
-                    "nrpe": {
-                        s: {"command": f"check_{s}"} for s in services_to_add
-                    },
-                },
-            },
-        }
-
-        relation.data[self.unit].update({
-            "monitors": yaml.safe_dump(monitors),
-        })
-
-        if not os.path.exists(NRPE_D_DIR):
-            logger.debug("NRPE directories not ready")
-            return
-
         for service in services_to_add:
-            service_cfg = service.replace("-", "_")
-            cfg_filename = os.path.join(NRPE_D_DIR, f"check_{service_cfg}.cfg")
-
-            if os.path.exists(cfg_filename):
-                continue
-
-            with open(cfg_filename, "w") as cfg_fp:
-                cfg_fp.write(f"""# check {service}
-# The following header was added by the landscape-server charm
-# Modifying it will affect nagios monitoring and alerting
-# servicegroups: juju
-command[check_{service}]=/usr/local/lib/nagios/plugins/check_systemd.py {service}
-""")
+            self.nrpe_client.add_check(
+                command=[
+                    '/usr/local/lib/nagios/plugins/check_systemd.py',
+                    '{}'.format(service),
+                ],
+                name='check_{}'.format(service)
+            )
 
         for service in services_to_remove:
-            service_cfg = service.replace("-", "_")
-            cfg_filename = os.path.join(NRPE_D_DIR, f"check_{service_cfg}.cfg")
-
-            if not os.path.exists(cfg_filename):
-                continue
-
-            os.remove(cfg_filename)
+            self.nrpe_client.remote_check(
+                name='check_{}'.format(service),
+            )
 
     def _leader_elected(self, event: LeaderElectedEvent) -> None:
         # Just because we received this event does not mean we are
@@ -672,13 +643,15 @@ command[check_{service}]=/usr/local/lib/nagios/plugins/check_systemd.py {service
             ip = str(self.model.get_binding(peer_relation).network.bind_address)
             peer_relation.data[self.app].update({"leader-ip": ip})
 
-            update_service_conf({
-                "package-search": {
-                    "host": "localhost",
-                },
-            })
+            self._on_nrpe_available(event)
 
+        update_service_conf({
+            "package-search": {
+                "host": "localhost",
+            },
+        })
         self._leader_changed()
+        self._on_nrpe_available(event)
 
     def _leader_settings_changed(
             self, event: LeaderSettingsChangedEvent) -> None:
@@ -698,18 +671,13 @@ command[check_{service}]=/usr/local/lib/nagios/plugins/check_systemd.py {service
                 })
 
         self._leader_changed()
+        self._on_nrpe_available(event)
 
     def _leader_changed(self) -> None:
         """
         Generic updates that need to happen whenever leadership changes,
         in both leaders and non-leaders.
         """
-        # Update any nrpe checks.
-        nrpe_relations = self.model.relations.get("nrpe-external-master", [])
-
-        for relation in nrpe_relations:
-            self._update_nrpe_checks(relation)
-
         haproxy_relations = self.model.relations.get("website", [])
         for relation in haproxy_relations:
             self._update_haproxy_connection(relation)