[DPE-4066] HA interface

config.yaml

@@ -16,6 +16,11 @@ options:
       The port on which the pgbouncer prometheus exporter serves metrics.
     type: int
 
+  vip:
+    description: |
+      Virtual IP to use to front pgbouncer units. Used only in case of external node connection.
+    type: string
+
   pool_mode:
     default: session
     description: |

lib/charms/grafana_agent/v0/cos_agent.py

@@ -213,14 +213,18 @@ def __init__(self, *args):
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, Callable, ClassVar, Dict, List, Optional, Set, Tuple, Union
 
-import pydantic
 from cosl import GrafanaDashboard, JujuTopology
 from cosl.rules import AlertRules
 from ops.charm import RelationChangedEvent
 from ops.framework import EventBase, EventSource, Object, ObjectEvents
 from ops.model import Relation
 from ops.testing import CharmType
 
+try:
+    import pydantic.v1 as pydantic
+except ImportError:
+    import pydantic
+
 if TYPE_CHECKING:
     try:
         from typing import TypedDict
@@ -234,9 +238,9 @@ class _MetricsEndpointDict(TypedDict):
 
 LIBID = "dc15fa84cef84ce58155fb84f6c6213a"
 LIBAPI = 0
-LIBPATCH = 8
+LIBPATCH = 9
 
-PYDEPS = ["cosl", "pydantic < 2"]
+PYDEPS = ["cosl", "pydantic"]
 
 DEFAULT_RELATION_NAME = "cos-agent"
 DEFAULT_PEER_RELATION_NAME = "peers"
@@ -721,8 +725,18 @@ def metrics_jobs(self) -> List[Dict]:
     @property
     def snap_log_endpoints(self) -> List[SnapEndpoint]:
         """Fetch logging endpoints exposed by related snaps."""
+        endpoints = []
+        endpoints_with_topology = self.snap_log_endpoints_with_topology
+        for endpoint, _ in endpoints_with_topology:
+            endpoints.append(endpoint)
+
+        return endpoints
+
+    @property
+    def snap_log_endpoints_with_topology(self) -> List[Tuple[SnapEndpoint, JujuTopology]]:
+        """Fetch logging endpoints and charm topology for each related snap."""
         plugs = []
-        for data, _ in self._remote_data:
+        for data, topology in self._remote_data:
             targets = data.log_slots
             if targets:
                 for target in targets:
@@ -733,15 +747,16 @@ def snap_log_endpoints(self) -> List[SnapEndpoint]:
                             "endpoints; this should not happen."
                         )
                     else:
-                        plugs.append(target)
+                        plugs.append((target, topology))
 
         endpoints = []
-        for plug in plugs:
+        for plug, topology in plugs:
             if ":" not in plug:
                 logger.error(f"invalid plug definition received: {plug}. Ignoring...")
             else:
                 endpoint = SnapEndpoint(*plug.split(":"))
-                endpoints.append(endpoint)
+                endpoints.append((endpoint, topology))
+
         return endpoints
 
     @property

lib/charms/tempo_k8s/v1/charm_tracing.py

@@ -172,14 +172,65 @@ def my_tracing_endpoint(self) -> Optional[str]:
 provide an *absolute* path to the certificate file instead.
 """
 
+
+def _remove_stale_otel_sdk_packages():
+    """Hack to remove stale opentelemetry sdk packages from the charm's python venv.
+
+    See https://github.com/canonical/grafana-agent-operator/issues/146 and
+    https://bugs.launchpad.net/juju/+bug/2058335 for more context. This patch can be removed after
+    this juju issue is resolved and sufficient time has passed to expect most users of this library
+    have migrated to the patched version of juju.  When this patch is removed, un-ignore rule E402 for this file in the pyproject.toml (see setting
+    [tool.ruff.lint.per-file-ignores] in pyproject.toml).
+
+    This only has an effect if executed on an upgrade-charm event.
+    """
+    # all imports are local to keep this function standalone, side-effect-free, and easy to revert later
+    import os
+
+    if os.getenv("JUJU_DISPATCH_PATH") != "hooks/upgrade-charm":
+        return
+
+    import logging
+    import shutil
+    from collections import defaultdict
+
+    from importlib_metadata import distributions
+
+    otel_logger = logging.getLogger("charm_tracing_otel_patcher")
+    otel_logger.debug("Applying _remove_stale_otel_sdk_packages patch on charm upgrade")
+    # group by name all distributions starting with "opentelemetry_"
+    otel_distributions = defaultdict(list)
+    for distribution in distributions():
+        name = distribution._normalized_name  # type: ignore
+        if name.startswith("opentelemetry_"):
+            otel_distributions[name].append(distribution)
+
+    otel_logger.debug(f"Found {len(otel_distributions)} opentelemetry distributions")
+
+    # If we have multiple distributions with the same name, remove any that have 0 associated files
+    for name, distributions_ in otel_distributions.items():
+        if len(distributions_) <= 1:
+            continue
+
+        otel_logger.debug(f"Package {name} has multiple ({len(distributions_)}) distributions.")
+        for distribution in distributions_:
+            if not distribution.files:  # Not None or empty list
+                path = distribution._path  # type: ignore
+                otel_logger.info(f"Removing empty distribution of {name} at {path}.")
+                shutil.rmtree(path)
+
+    otel_logger.debug("Successfully applied _remove_stale_otel_sdk_packages patch. ")
+
+
+_remove_stale_otel_sdk_packages()
+
+
 import functools
 import inspect
 import logging
 import os
-import shutil
 from contextlib import contextmanager
 from contextvars import Context, ContextVar, copy_context
-from importlib.metadata import distributions
 from pathlib import Path
 from typing import (
     Any,
@@ -199,14 +250,15 @@ def my_tracing_endpoint(self) -> Optional[str]:
 from opentelemetry.sdk.resources import Resource
 from opentelemetry.sdk.trace import Span, TracerProvider
 from opentelemetry.sdk.trace.export import BatchSpanProcessor
-from opentelemetry.trace import INVALID_SPAN, Tracer
-from opentelemetry.trace import get_current_span as otlp_get_current_span
 from opentelemetry.trace import (
+    INVALID_SPAN,
+    Tracer,
     get_tracer,
     get_tracer_provider,
     set_span_in_context,
     set_tracer_provider,
 )
+from opentelemetry.trace import get_current_span as otlp_get_current_span
 from ops.charm import CharmBase
 from ops.framework import Framework
 
@@ -219,7 +271,7 @@ def my_tracing_endpoint(self) -> Optional[str]:
 # Increment this PATCH version before using `charmcraft publish-lib` or reset
 # to 0 if you are raising the major API version
 
-LIBPATCH = 13
+LIBPATCH = 14
 
 PYDEPS = ["opentelemetry-exporter-otlp-proto-http==1.21.0"]
 
@@ -361,30 +413,6 @@ def _get_server_cert(
     return server_cert
 
 
-def _remove_stale_otel_sdk_packages():
-    """Hack to remove stale opentelemetry sdk packages from the charm's python venv.
-
-    See https://github.com/canonical/grafana-agent-operator/issues/146 and
-    https://bugs.launchpad.net/juju/+bug/2058335 for more context. This patch can be removed after
-    this juju issue is resolved and sufficient time has passed to expect most users of this library
-    have migrated to the patched version of juju.
-
-    This only does something if executed on an upgrade-charm event.
-    """
-    if os.getenv("JUJU_DISPATCH_PATH") == "hooks/upgrade-charm":
-        logger.debug("Executing _remove_stale_otel_sdk_packages patch on charm upgrade")
-        # Find any opentelemetry_sdk distributions
-        otel_sdk_distributions = list(distributions(name="opentelemetry_sdk"))
-        # If there is more than 1, inspect each and if it has 0 entrypoints, infer that it is stale
-        if len(otel_sdk_distributions) > 1:
-            for distribution in otel_sdk_distributions:
-                if len(distribution.entry_points) == 0:
-                    # Distribution appears to be empty. Remove it
-                    path = distribution._path  # type: ignore
-                    logger.debug(f"Removing empty opentelemetry_sdk distribution at: {path}")
-                    shutil.rmtree(path)
-
-
 def _setup_root_span_initializer(
     charm_type: _CharmType,
     tracing_endpoint_attr: str,
@@ -420,7 +448,6 @@ def wrap_init(self: CharmBase, framework: Framework, *args, **kwargs):
         # apply hacky patch to remove stale opentelemetry sdk packages on upgrade-charm.
         # it could be trouble if someone ever decides to implement their own tracer parallel to
         # ours and before the charm has inited. We assume they won't.
-        _remove_stale_otel_sdk_packages()
         resource = Resource.create(
             attributes={
                 "service.name": _service_name,

metadata.yaml

@@ -71,3 +71,8 @@ requires:
     interface: tracing
     limit: 1
     optional: true
+
+  ha:
+    interface: hacluster
+    limit: 1
+    optional: true

src/charm.py

@@ -24,17 +24,18 @@
 from charms.tempo_k8s.v1.charm_tracing import trace_charm
 from charms.tempo_k8s.v2.tracing import TracingEndpointRequirer
 from jinja2 import Template
-from ops import JujuVersion
-from ops.charm import CharmBase, StartEvent
-from ops.main import main
-from ops.model import (
+from ops import (
     ActiveStatus,
     BlockedStatus,
+    CharmBase,
+    JujuVersion,
     MaintenanceStatus,
     ModelError,
     Relation,
+    StartEvent,
     WaitingStatus,
 )
+from ops.main import main
 
 from constants import (
     APP_SCOPE,
@@ -65,6 +66,7 @@
 )
 from relations.backend_database import BackendDatabaseRequires
 from relations.db import DbProvides
+from relations.hacluster import HaCluster
 from relations.peers import Peers
 from relations.pgbouncer_provider import PgBouncerProvider
 from upgrade import PgbouncerUpgrade, get_pgbouncer_dependencies_model
@@ -120,6 +122,7 @@ def __init__(self, *args):
         self.legacy_db_relation = DbProvides(self, admin=False)
         self.legacy_db_admin_relation = DbProvides(self, admin=True)
         self.tls = PostgreSQLTLS(self, PEER_RELATION_NAME)
+        self.hacluster = HaCluster(self)
 
         self.service_ids = list(range(self.instances_count))
         self.pgb_services = [
@@ -507,8 +510,24 @@ def update_status(self):
             self.unit.status = BlockedStatus("backend database relation not ready")
             return
 
+        if self.hacluster.relation and not self._is_exposed:
+            self.unit.status = BlockedStatus("ha integration used without data-intgrator")
+            return
+
+        vip = self.config.get("vip")
+        if self.hacluster.relation and not vip:
+            self.unit.status = BlockedStatus("ha integration used without vip configuration")
+            return
+
+        if vip and not self._is_exposed:
+            self.unit.status = BlockedStatus("vip configuration without data-intgrator")
+            return
+
         if self.check_pgb_running():
-            self.unit.status = ActiveStatus()
+            if self.unit.is_leader() and vip:
+                self.unit.status = ActiveStatus(f"VIP: {vip}")
+            else:
+                self.unit.status = ActiveStatus()
 
     def _on_config_changed(self, event) -> None:
         """Config changed handler.
@@ -521,19 +540,28 @@ def _on_config_changed(self, event) -> None:
             event.defer()
             return
 
+        old_vip = self.peers.app_databag.get("current_vip", "")
+        vip = self.config.get("vip", "")
+        vip_changed = old_vip != vip
+        if vip_changed and self._is_exposed:
+            self.hacluster.set_vip(self.config.get("vip"))
+
         old_port = self.peers.app_databag.get("current_port")
         port_changed = old_port != str(self.config["listen_port"])
         if port_changed and self._is_exposed:
-            if self.unit.is_leader():
-                self.peers.app_databag["current_port"] = str(self.config["listen_port"])
             # Open port
             try:
-                if old_port:
-                    self.unit.close_port("tcp", old_port)
-                self.unit.open_port("tcp", self.config["listen_port"])
+                self.unit.set_ports(self.config["listen_port"])
             except ModelError:
                 logger.exception("failed to open port")
 
+        if self.unit.is_leader():
+            self.peers.app_databag["current_port"] = str(self.config["listen_port"])
+            if vip:
+                self.peers.app_databag["current_vip"] = str(vip)
+            else:
+                self.peers.app_databag.pop("current_vip", None)
+
         # TODO hitting upgrade errors here due to secrets labels failing to set on non-leaders.
         # deferring until the leader manages to set the label
         try:
@@ -546,6 +574,9 @@ def _on_config_changed(self, event) -> None:
         if self.backend.postgres:
             self.render_prometheus_service()
 
+        if port_changed or vip_changed:
+            self.update_client_connection_info()
+
     def check_pgb_running(self):
         """Checks that pgbouncer service is running, and updates status accordingly."""
         prom_service = f"{PGB}-{self.app.name}-prometheus"
@@ -910,14 +941,13 @@ def unit_ip(self) -> str:
     #  Relation Utilities
     # =====================
 
-    def update_client_connection_info(self, port: Optional[str] = None):
+    def update_client_connection_info(self):
         """Update ports in backend relations to match updated pgbouncer port."""
         # Skip updates if backend.postgres doesn't exist yet.
         if not self.backend.postgres or not self.unit.is_leader():
             return
 
-        if port is None:
-            port = self.config["listen_port"]
+        port = self.config["listen_port"]
 
         for relation in self.model.relations.get("db", []):
             self.legacy_db_relation.update_connection_info(relation, port)

src/constants.py

@@ -36,6 +36,7 @@
 BACKEND_RELATION_NAME = "backend-database"
 PEER_RELATION_NAME = "pgb-peers"
 CLIENT_RELATION_NAME = "database"
+HACLUSTER_RELATION_NAME = "ha"
 
 TLS_KEY_FILE = "key.pem"
 TLS_CA_FILE = "ca.pem"