From c70fa140b824c4a2136fd55734b9560f6b4176ca Mon Sep 17 00:00:00 2001
From: pavel <pavel.larkin@ikeja.co.za>
Date: Tue, 24 Feb 2026 17:56:04 +0200
Subject: [PATCH] 1756

---
 files/nats_registration_listener.py | 56 +++++++++++++++++++++++++++--
 1 file changed, 53 insertions(+), 3 deletions(-)

diff --git a/files/nats_registration_listener.py b/files/nats_registration_listener.py
index f8cd0b2..740aeb2 100644
--- a/files/nats_registration_listener.py
+++ b/files/nats_registration_listener.py
@@ -41,6 +41,10 @@ NB_URL = "http://netbox.gt-tiso.ikeja.co.za"  # Base URL
 NB_TOKEN = "7648e4f5ee370cda7834682e61b47c2ee8e95623"  # keep as provided
 NB_TIMEOUT = 3.0  # seconds per HTTP GET
 
+# Cache for MAC -> NetBox lookup result (seconds). Keeps NetBox load down under chatty devices.
+NB_LOOKUP_CACHE_TTL = float(os.environ.get("NB_LOOKUP_CACHE_TTL", "10.0"))
+NB_LOOKUP_CACHE = {}  # mac_norm -> (expires_monotonic, cached_tuple)
+
 # =========================
 # RabbitMQ hardcoded config
 # =========================
@@ -232,6 +236,22 @@ def nb_lookup_device_by_mac(mac: str, log_status) -> Tuple[
     if not mac_norm:
         return None, None, None, None, None, None, None, None, None
 
+    # Step 0: short TTL cache (avoid repeated NetBox GETs for chatty devices)
+    if NB_LOOKUP_CACHE_TTL > 0:
+        nowm = monotonic()
+        cached = NB_LOOKUP_CACHE.get(mac_norm)
+        if cached:
+            exp, val = cached
+            if exp > nowm:
+                host, iface_id, dev_id, status_val, tag_slugs, action_next, action_last, action_next_timestamp, action_state = val
+                if tag_slugs is not None:
+                    try:
+                        tag_slugs = set(tag_slugs)
+                    except Exception:
+                        pass
+                return host, iface_id, dev_id, status_val, tag_slugs, action_next, action_last, action_next_timestamp, action_state
+            NB_LOOKUP_CACHE.pop(mac_norm, None)
+
     base = NB_URL.rstrip("/")
     h = {
         "Accept": "application/json",
@@ -240,7 +260,13 @@ def nb_lookup_device_by_mac(mac: str, log_status) -> Tuple[
     }
 
     # Step 1: MAC lookup
-    data, code = http_get_json(f"{base}/api/dcim/mac-addresses/", params={"mac_address": mac_norm, "limit": "2"}, headers=h)
+    data, code = http_get_json(
+        f"{base}/api/dcim/mac-addresses/",
+        params={"mac_address": mac_norm, "limit": "2", "fields": "assigned_object_type,assigned_object_id"},
+        headers=h,
+    )
+    if code == 400:
+        data, code = http_get_json(f"{base}/api/dcim/mac-addresses/", params={"mac_address": mac_norm, "limit": "2"}, headers=h)
     if code != 200 or not data:
         asyncio.create_task(nb_problem(log_status, f"[{ts()}] nb: mac query http={code} mac={mac_norm}"))
         return None, None, None, None, None, None, None, None, None
@@ -265,7 +291,9 @@ def nb_lookup_device_by_mac(mac: str, log_status) -> Tuple[
         return None, aoid, None, None, None, None, None, None, None
 
     # Step 2: Interface -> Device (shallow)
-    iface, code2 = http_get_json(f"{base}/api/dcim/interfaces/{aoid}/", headers=h)
+    iface, code2 = http_get_json(f"{base}/api/dcim/interfaces/{aoid}/", params={"fields": "device"}, headers=h)
+    if code2 == 400:
+        iface, code2 = http_get_json(f"{base}/api/dcim/interfaces/{aoid}/", headers=h)
     if code2 != 200 or not iface:
         asyncio.create_task(nb_problem(log_status, f"[{ts()}] nb: iface fetch http={code2} iface_id={aoid}"))
         return None, aoid, None, None, None, None, None, None, None
@@ -278,7 +306,9 @@ def nb_lookup_device_by_mac(mac: str, log_status) -> Tuple[
         return None, aoid, None, None, None, None, None, None, None
 
     # Step 3: Device detail (single fetch for status, tags, custom_fields.*)
-    device, code3 = http_get_json(f"{base}/api/dcim/devices/{dev_id}/", headers=h)
+    device, code3 = http_get_json(f"{base}/api/dcim/devices/{dev_id}/", params={"fields": "status,tags,custom_fields"}, headers=h)
+    if code3 == 400:
+        device, code3 = http_get_json(f"{base}/api/dcim/devices/{dev_id}/", headers=h)
     if code3 != 200 or not device:
         # treat as "no extra info"
         return host, aoid, dev_id, None, None, None, None, None, None
@@ -297,6 +327,26 @@ def nb_lookup_device_by_mac(mac: str, log_status) -> Tuple[
     action_next_timestamp = cf.get("action_next_timestamp")
     action_state = cf.get("action_state")
 
+    # Step 4: populate cache (only on full success)
+    if NB_LOOKUP_CACHE_TTL > 0:
+        try:
+            NB_LOOKUP_CACHE[mac_norm] = (
+                monotonic() + NB_LOOKUP_CACHE_TTL,
+                (
+                    host,
+                    aoid,
+                    dev_id,
+                    (status_val if isinstance(status_val, str) else None),
+                    (tuple(tag_slugs) if tag_slugs is not None else None),
+                    action_next,
+                    action_last,
+                    action_next_timestamp,
+                    action_state,
+                ),
+            )
+        except Exception:
+            pass
+
     return host, aoid, dev_id, (status_val if isinstance(status_val, str) else None), tag_slugs, action_next, action_last, action_next_timestamp, action_state