Proven working Postgresql monitor

2026-09-17 09:23:17 -04:00
commit 0123da0e12
7 changed files with 769 additions and 0 deletions
@@ -0,0 +1,139 @@
+#!/usr/bin/env python3
+
+import psycopg2
+import psycopg2.extras
+
+
+def read_secret(path):
+    with open(path, "r", encoding="utf-8") as f:
+        return f.read().strip()
+
+
+def monitor(config):
+    password = read_secret(
+        config["PGPASSWORD_FILE"]
+    )
+
+    conn = psycopg2.connect(
+        host=config["PGHOST"],
+        port=int(config["PGPORT"]),
+        user=config["PGUSER"],
+        dbname=config["PGDATABASE"],
+        password=password,
+        connect_timeout=5,
+        application_name="managed-containerd",
+    )
+
+    try:
+        conn.autocommit = True
+
+        with conn.cursor(
+            cursor_factory=psycopg2.extras.RealDictCursor
+        ) as cur:
+
+            cur.execute("""
+                SELECT pg_is_in_recovery() AS in_recovery
+            """)
+
+            in_recovery = bool(
+                cur.fetchone()["in_recovery"]
+            )
+
+            if not in_recovery:
+                cur.execute("""
+                    SELECT
+                        pg_current_wal_lsn()::text
+                            AS current_wal_lsn
+                """)
+
+                primary = cur.fetchone()
+
+            else:
+                primary = None
+
+            if in_recovery:
+                cur.execute("""
+                    SELECT
+                        pg_last_wal_receive_lsn()::text
+                            AS receive_lsn,
+                        pg_last_wal_replay_lsn()::text
+                            AS replay_lsn,
+                        pg_is_wal_replay_paused()
+                            AS replay_paused
+                """)
+
+                recovery = cur.fetchone()
+
+                cur.execute("""
+                    SELECT
+                        status,
+                        sender_host,
+                        sender_port,
+                        slot_name,
+                        latest_end_lsn::text
+                            AS latest_end_lsn
+                    FROM pg_stat_wal_receiver
+                    LIMIT 1
+                """)
+
+                receiver = cur.fetchone()
+
+            else:
+                recovery = {
+                    "receive_lsn": None,
+                    "replay_lsn": None,
+                    "replay_paused": False,
+                }
+
+                receiver = None
+
+            cur.execute("""
+                SELECT
+                    current_setting('server_version')
+                        AS server_version,
+                    current_setting('cluster_name', true)
+                        AS cluster_name,
+                    current_user AS current_user,
+                    current_database()
+                        AS current_database,
+                    inet_server_addr()::text
+                        AS server_address,
+                    inet_server_port()
+                        AS server_port
+            """)
+
+            server = cur.fetchone()
+
+        if in_recovery:
+            role = "standby"
+
+            streaming = bool(
+                receiver
+                and receiver.get("status")
+                    == "streaming"
+            )
+
+        else:
+            role = "primary"
+            streaming = False
+
+        return {
+            "role": role,
+            "in_recovery": in_recovery,
+            "streaming": streaming,
+            "wal_receiver": (
+                dict(receiver)
+                if receiver
+                else None
+            ),
+            "recovery": dict(recovery),
+            "primary": (
+                dict(primary)
+                if primary
+                else None
+            ),
+            "server": dict(server),
+        }
+
+    finally:
+        conn.close()