Add read-only PostgreSQL promotion database preflight

2026-09-17 22:48:25 -04:00
parent 5fcd0bf931
commit 1e72dda238
4 changed files with 414 additions and 3 deletions
@@ -531,6 +531,191 @@ def notify_target(config, runtime_state, request_parameters):
    }


+
+def promote_database(config, runtime_state, request_parameters):
+    """
+    Read-only PostgreSQL promotion preflight.
+
+    Validate that Pacemaker selected this local node for promotion, that the
+    local PostgreSQL instance is still an eligible standby, and that no
+    reachable configured peer is currently writable.
+
+    This action never promotes PostgreSQL.  The OCF resource agent remains
+    authoritative for the actual pg_ctl promote operation and for local
+    filesystem state such as the rejoin-required marker.
+    """
+    del request_parameters
+
+    node_name = str(
+        config.get("PACEMAKER_NODE_NAME", "")
+    ).strip().lower()
+
+    if not node_name:
+        raise ValueError(
+            "PACEMAKER_NODE_NAME is required for promote.database"
+        )
+
+    notification = _postgres_runtime_state(
+        runtime_state
+    ).get("notification")
+
+    if not isinstance(notification, dict):
+        return {
+            "ocf_rc": 1,
+            "ready": False,
+            "mutation": False,
+            "reason": "notification_missing",
+            "node": node_name,
+        }
+
+    if notification.get("type") != "pre":
+        return {
+            "ocf_rc": 1,
+            "ready": False,
+            "mutation": False,
+            "reason": "notification_type_invalid",
+            "node": node_name,
+        }
+
+    if notification.get("operation") != "promote":
+        return {
+            "ocf_rc": 1,
+            "ready": False,
+            "mutation": False,
+            "reason": "notification_operation_invalid",
+            "node": node_name,
+        }
+
+    received_at = notification.get("received_at")
+
+    if not isinstance(received_at, (int, float)):
+        return {
+            "ocf_rc": 1,
+            "ready": False,
+            "mutation": False,
+            "reason": "notification_timestamp_invalid",
+            "node": node_name,
+        }
+
+    age = time.time() - received_at
+
+    if age < 0 or age > NOTIFY_TARGET_MAX_AGE_SECONDS:
+        return {
+            "ocf_rc": 1,
+            "ready": False,
+            "mutation": False,
+            "reason": "notification_stale",
+            "age_seconds": age,
+            "node": node_name,
+        }
+
+    target_node = str(
+        notification.get("promote_uname", "")
+    ).strip().lower()
+
+    previous_primary_node = str(
+        notification.get("demote_uname", "")
+    ).strip().lower()
+
+    if not target_node:
+        return {
+            "ocf_rc": 1,
+            "ready": False,
+            "mutation": False,
+            "reason": "promotion_target_missing",
+            "node": node_name,
+        }
+
+    if target_node != node_name:
+        return {
+            "ocf_rc": 1,
+            "ready": False,
+            "mutation": False,
+            "reason": "local_node_not_target",
+            "node": node_name,
+            "target_node": target_node,
+            "previous_primary_node": previous_primary_node,
+        }
+
+    local = monitor(config, runtime_state)
+    local_ha = local.get("ha") or {}
+
+    checks = {
+        "notification": True,
+        "local_target": True,
+        "local_monitor_success": (
+            local_ha.get("ocf_rc") == OCF_SUCCESS
+        ),
+        "local_standby": (
+            local.get("role") == "standby"
+            and local.get("in_recovery") is True
+        ),
+        "promotion_eligible": (
+            local_ha.get("promotion_eligible") is True
+        ),
+    }
+
+    if not all(checks.values()):
+        return {
+            "ocf_rc": 1,
+            "ready": False,
+            "mutation": False,
+            "reason": "local_not_promotion_eligible",
+            "node": node_name,
+            "target_node": target_node,
+            "previous_primary_node": previous_primary_node,
+            "local": local_ha,
+            "checks": checks,
+        }
+
+    try:
+        password = read_secret(config["PGPASSWORD_FILE"])
+        peer_state = probe_replication_peers(config, password)
+    except (KeyError, TypeError, ValueError, psycopg2.Error) as exc:
+        return {
+            "ocf_rc": 1,
+            "ready": False,
+            "mutation": False,
+            "reason": "peer_probe_failed",
+            "node": node_name,
+            "target_node": target_node,
+            "previous_primary_node": previous_primary_node,
+            "error": str(exc).splitlines()[0],
+            "checks": checks,
+        }
+
+    writable_primaries = [
+        str(name).strip().lower()
+        for name in peer_state.get("writable_primaries", [])
+    ]
+
+    checks["no_reachable_writable_primary"] = (
+        len(writable_primaries) == 0
+    )
+
+    ready = all(checks.values())
+
+    return {
+        "ocf_rc": OCF_SUCCESS if ready else OCF_ERR_GENERIC,
+        "ready": ready,
+        "mutation": False,
+        "reason": (
+            "promotion_preflight_ready"
+            if ready
+            else "reachable_writable_primary"
+        ),
+        "node": node_name,
+        "target_node": target_node,
+        "previous_primary_node": previous_primary_node,
+        "writable_primaries": writable_primaries,
+        "writable_primary_count": len(writable_primaries),
+        "peer_state": peer_state,
+        "checks": checks,
+        "age_seconds": age,
+    }
+
+
+
 def demote_prepare(config, runtime_state, request_parameters):
    """
    Read-only demotion preparation.