Refactors ceph_cos_agent lib as a charm-lib

UtkarshBhatthere · UtkarshBhatthere · commit 18114b4231fc · 2025-03-03T21:21:28.000+05:30
Signed-off-by: utkarsh bhatt &lt;utkarsh.bhatt@canonical.com&gt;
diff --git a/ceph-mon/lib/charms/ceph_mon/v0/ceph_cos_agent.py b/ceph-mon/lib/charms/ceph_mon/v0/ceph_cos_agent.py
@@ -0,0 +1,124 @@
+# Copyright 2022 Canonical Ltd.
+# See LICENSE file for licensing details.
+
+"""
+Library for handling ceph observability integrations
+"""
+
+import logging
+import socket
+import tenacity
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    import charm
+
+from charms.grafana_agent.v0 import cos_agent
+from charms_ceph import utils as ceph_utils
+
+# The unique Charmhub library identifier, never change it
+LIBID = "ac526775f8ed42ebb46b231dc00519d3"
+
+# Increment this major API version when introducing breaking changes
+LIBAPI = 0
+
+# Increment this PATCH version before using `charmcraft publish-lib` or reset
+# to 0 if you are raising the major API version
+LIBPATCH = 1
+
+logger = logging.getLogger(__name__)
+
+
+class CephCOSAgentProvider(cos_agent.COSAgentProvider):
+
+    def __init__(self, charm):
+        super().__init__(
+            charm,
+            metrics_rules_dir="./files/prometheus_alert_rules",
+            dashboard_dirs=["./files/grafana_dashboards"],
+            scrape_configs=self._custom_scrape_configs,
+        )
+        events = self._charm.on[cos_agent.DEFAULT_RELATION_NAME]
+        self.framework.observe(
+            events.relation_departed, self._on_relation_departed
+        )
+
+    def _on_refresh(self, event):
+        """Enable prometheus on relation change"""
+        if not ceph_utils.is_bootstrapped():
+            logger.debug("not bootstrapped, defer _on_refresh: %s", event)
+            event.defer()
+            return
+        logger.debug("refreshing cos_agent relation")
+        if self._charm.unit.is_leader():
+            self.mgr_config_set_rbd_stats_pools()
+            ceph_utils.mgr_enable_module("prometheus")
+        super()._on_refresh(event)
+
+    def _on_relation_departed(self, event):
+        """Disable prometheus on depart of relation"""
+        if self._charm.unit.is_leader() and ceph_utils.is_bootstrapped():
+            logger.debug(
+                "is_leader and is_bootstrapped, running rel departed: %s",
+                event,
+            )
+            ceph_utils.mgr_disable_module("prometheus")
+            logger.debug("module_disabled")
+
+    def _custom_scrape_configs(self):
+        fqdn = socket.getfqdn()
+        fqdn_parts = fqdn.split('.')
+        domain = '.'.join(fqdn_parts[1:]) if len(fqdn_parts) > 1 else fqdn
+        return [
+            {
+                "metrics_path": "/metrics",
+                "static_configs": [{"targets": ["localhost:9283"]}],
+                "honor_labels": True,
+                "metric_relabel_configs": [
+                    {
+                        # localhost:9283 is the generic default instance label
+                        # added by grafana-agent which is kinda useless.
+                        # Replace it with a somewhat more meaningful label
+                        "source_labels": ["instance"],
+                        "regex": "^localhost:9283$",
+                        "target_label": "instance",
+                        "action": "replace",
+                        "replacement": "ceph_cluster",
+                    },
+                    {   # if we have a non-empty hostname label, use it as the
+                        # instance label
+                        "source_labels": ["hostname"],
+                        "regex": "(.+)",
+                        "target_label": "instance",
+                        "action": "replace",
+                        "replacement": "${1}",
+                    },
+                    {   # tack on the domain to the instance label to make it
+                        # conform to grafana-agent's node-exporter expectations
+                        "source_labels": ["instance"],
+                        "regex": "(.*)",
+                        "target_label": "instance",
+                        "action": "replace",
+                        "replacement": "${1}." + domain,
+                    },
+                ]
+            },
+
+        ]
+
+    # TODO: switch to charms_ceph sourced util function
+    @tenacity.retry(
+        wait=tenacity.wait_exponential(multiplier=1, max=10),
+        reraise=True,
+        stop=tenacity.stop_after_attempt(30))
+    def mgr_config_set_rbd_stats_pools(self):
+        """Update ceph mgr config with the value from rbd-status-pools config
+        """
+        if self._charm.unit.is_leader() and ceph_utils.is_bootstrapped():
+            rbd_stats_pools = self._charm.model.config.get('rbd-stats-pools')
+            if rbd_stats_pools:
+                ceph_utils.mgr_config_set(
+                    'mgr/prometheus/rbd_stats_pools',
+                    rbd_stats_pools
+                )
+
diff --git a/ceph-mon/src/ceph_metrics.py b/ceph-mon/src/ceph_metrics.py
@@ -5,11 +5,11 @@
 
 Configure prometheus scrape jobs via the metrics-endpoint relation.
 """
+
 import json
 import logging
 import os.path
 import pathlib
-import socket
 
 from typing import Optional, Union, List, TYPE_CHECKING
 
@@ -19,7 +19,6 @@
     import charm
 
 from charms.prometheus_k8s.v0 import prometheus_scrape
-from charms.grafana_agent.v0 import cos_agent
 from charms_ceph import utils as ceph_utils
 from ops.framework import BoundEvent
 from utils import mgr_config_set_rbd_stats_pools
@@ -151,81 +150,3 @@ def update_alert_rules(self):
                 self._charm._stored.alert_rule_errors = msg
                 return
             self._set_alert_rules(alert_rules_as_dict)
-
-
-class CephCOSAgentProvider(cos_agent.COSAgentProvider):
-
-    def __init__(self, charm):
-        super().__init__(
-            charm,
-            metrics_rules_dir="./files/prometheus_alert_rules",
-            dashboard_dirs=["./files/grafana_dashboards"],
-            scrape_configs=self._custom_scrape_configs,
-        )
-        events = self._charm.on[cos_agent.DEFAULT_RELATION_NAME]
-        self.framework.observe(
-            events.relation_departed, self._on_relation_departed
-        )
-
-    def _on_refresh(self, event):
-        """Enable prometheus on relation change"""
-        if not ceph_utils.is_bootstrapped():
-            logger.debug("not bootstrapped, defer _on_refresh: %s", event)
-            event.defer()
-            return
-        logger.debug("refreshing cos_agent relation")
-        if self._charm.unit.is_leader():
-            mgr_config_set_rbd_stats_pools()
-            ceph_utils.mgr_enable_module("prometheus")
-        super()._on_refresh(event)
-
-    def _on_relation_departed(self, event):
-        """Disable prometheus on depart of relation"""
-        if self._charm.unit.is_leader() and ceph_utils.is_bootstrapped():
-            logger.debug(
-                "is_leader and is_bootstrapped, running rel departed: %s",
-                event,
-            )
-            ceph_utils.mgr_disable_module("prometheus")
-            logger.debug("module_disabled")
-
-    def _custom_scrape_configs(self):
-        fqdn = socket.getfqdn()
-        fqdn_parts = fqdn.split('.')
-        domain = '.'.join(fqdn_parts[1:]) if len(fqdn_parts) > 1 else fqdn
-        return [
-            {
-                "metrics_path": "/metrics",
-                "static_configs": [{"targets": ["localhost:9283"]}],
-                "honor_labels": True,
-                "metric_relabel_configs": [
-                    {
-                        # localhost:9283 is the generic default instance label
-                        # added by grafana-agent which is kinda useless.
-                        # Replace it with a somewhat more meaningful label
-                        "source_labels": ["instance"],
-                        "regex": "^localhost:9283$",
-                        "target_label": "instance",
-                        "action": "replace",
-                        "replacement": "ceph_cluster",
-                    },
-                    {   # if we have a non-empty hostname label, use it as the
-                        # instance label
-                        "source_labels": ["hostname"],
-                        "regex": "(.+)",
-                        "target_label": "instance",
-                        "action": "replace",
-                        "replacement": "${1}",
-                    },
-                    {   # tack on the domain to the instance label to make it
-                        # conform to grafana-agent's node-exporter expectations
-                        "source_labels": ["instance"],
-                        "regex": "(.*)",
-                        "target_label": "instance",
-                        "action": "replace",
-                        "replacement": "${1}." + domain,
-                    },
-                ]
-            },
-
-        ]
diff --git a/ceph-mon/src/charm.py b/ceph-mon/src/charm.py
@@ -10,6 +10,7 @@
 
 import charms.operator_libs_linux.v0.apt as apt
 import charms.operator_libs_linux.v1.systemd as systemd
+from charms.ceph_mon.v0 import ceph_cos_agent
 
 from ops.charm import CharmEvents
 from ops.framework import EventBase, EventSource
@@ -41,7 +42,7 @@ class CephCharmEvents(CharmEvents):
 
 class CephMonCharm(ops_openstack.core.OSBaseCharm):
 
-    release = 'quincy'
+    release = 'squid'
 
     PACKAGES = [
         'ceph', 'gdisk',
@@ -216,7 +217,7 @@ def __init__(self, *args):
 
         self.clients = ceph_client.CephClientProvides(self)
         self.metrics_endpoint = ceph_metrics.CephMetricsEndpointProvider(self)
-        self.cos_agent = ceph_metrics.CephCOSAgentProvider(self)
+        self.cos_agent = ceph_cos_agent.CephCOSAgentProvider(self)
         self.ceph_status = ceph_status.StatusAssessor(self)
         self.mds = ceph_mds.CephMdsProvides(self)
 
diff --git a/ceph-mon/unit_tests/test_ceph_metrics.py b/ceph-mon/unit_tests/test_ceph_metrics.py
@@ -12,7 +12,6 @@
 
 from ops.testing import Harness
 
-import ceph_metrics  # noqa: avoid circ. import
 import charm
 
 
@@ -155,17 +154,21 @@ def test_init(self):
             "cos-agent",
         )
 
-    @patch("ceph_metrics.mgr_config_set_rbd_stats_pools", lambda: None)
-    @patch("ceph_metrics.ceph_utils.is_bootstrapped", return_value=True)
-    @patch("ceph_metrics.ceph_utils.is_mgr_module_enabled", return_value=False)
-    @patch("ceph_metrics.ceph_utils.mgr_enable_module")
-    @patch("ceph_metrics.ceph_utils.mgr_disable_module")
+    @patch("charms.ceph_mon.v0.ceph_cos_agent."
+           "CephCOSAgentProvider.mgr_config_set_rbd_stats_pools")
+    @patch("charms.ceph_mon.v0.ceph_cos_agent."
+           "ceph_utils.is_bootstrapped", return_value=True)
+    @patch("charms.ceph_mon.v0.ceph_cos_agent."
+           "ceph_utils.is_mgr_module_enabled", return_value=False)
+    @patch("charms.ceph_mon.v0.ceph_cos_agent.ceph_utils.mgr_enable_module")
+    @patch("charms.ceph_mon.v0.ceph_cos_agent.ceph_utils.mgr_disable_module")
     def test_add_remove_rel(
         self,
         mgr_disable_module,
         mgr_enable_module,
         _is_mgr_module_enable,
         _is_bootstrapped,
+        mgr_config_set_rbd_stats_pools,
     ):
         rel_id = self.harness.add_relation("cos-agent", "grafana-agent")
         self.harness.add_relation_unit(rel_id, "grafana-agent/0")
@@ -181,6 +184,7 @@ def test_add_remove_rel(
         self.assertTrue("metrics_alert_rules" in data)
         self.assertTrue("groups" in data["metrics_alert_rules"])
         mgr_enable_module.assert_called_once()
+        mgr_config_set_rbd_stats_pools.assert_called_once()
 
         self.harness.remove_relation(rel_id)
         mgr_disable_module.assert_called_once()
