Give threads less time to shut down gracefully. Attempt shutdown() rather than stop() on the telnet connections to see if that improves things

# Conflicts:
#	templates/add_spot.html
#	templates/alerts.html
#	templates/bands.html
#	templates/base.html
#	templates/conditions.html
#	templates/map.html
#	templates/spots.html
#	templates/status.html

core/cleanup.py

@@ -35,7 +35,7 @@ class CleanupTimer:
 
         self._stop_event.set()
         if self._thread:
-            self._thread.join(timeout=15)
+            self._thread.join(timeout=5)
             if self._thread.is_alive():
                 logger.warning("Cleanup worker thread did not exit on time and will be killed.")
 

core/constants.py

@@ -4,7 +4,7 @@ from data.band import Band
 from data.sig import SIG
 
 # General software
-SOFTWARE_VERSION = "2.1"
+SOFTWARE_VERSION = "2.1.2"
 
 # HTTP headers used for spot providers that use HTTP
 HTTP_HEADERS = {"User-Agent": f"Spothole v{SOFTWARE_VERSION} (operated by {SERVER_OWNER_CALLSIGN})"}

core/data_providers.py

@@ -91,7 +91,7 @@ class DataProviders:
         for t in threads:
             t.start()
 
-        deadline = time.monotonic() + 40
+        deadline = time.monotonic() + 15
         for t in threads:
             t.join(timeout=max(0.0, deadline - time.monotonic()))
         still_running = [t for t in threads if t.is_alive()]

core/live_data_cache.py

@@ -77,7 +77,12 @@ class LiveDataCache:
             logger.exception(f"Failed to write snapshot to {self._snapshot_dir}")
 
     def _load_snapshot(self):
+        try:
             data = self._disk_cache.get("snapshot")
+        except Exception:
+            logger.warning(f"Failed to load snapshot from {self._snapshot_dir}, clearing it.")
+            self._disk_cache.clear()
+            return
         if not data:
             return
 

core/single_object_data_cache.py

@@ -21,7 +21,13 @@ class SingleObjectDataCache:
         # This cache stores a single object, doesn't matter what it's called so "object" will do
         if "object" not in self._cache:
             self._cache.add("object", object_if_empty)
+        try:
             self._obj = self._cache.get("object")
+        except Exception:
+            logger.warning(f"Failed to load cache from {cache_dir}, clearing it.")
+            self._cache.clear()
+            self._cache.add("object", object_if_empty)
+            self._obj = object_if_empty
 
     def get(self):
         """Get the data object. This can then be manipulated as necessary across multiple threads. Any function

providers/alert/http_alert_provider.py

@@ -4,7 +4,7 @@ from threading import Event, Thread
 
 import pytz
 import requests
-from requests.exceptions import ConnectionError, ConnectTimeout, ReadTimeout
+from requests.exceptions import ConnectionError, ConnectTimeout, JSONDecodeError, ReadTimeout
 
 from core.constants import HTTP_HEADERS
 from providers.alert.alert_provider import AlertProvider
@@ -33,7 +33,7 @@ class HTTPAlertProvider(AlertProvider):
     def stop(self):
         self._stop_event.set()
         if self._thread:
-            self._thread.join(timeout=35)
+            self._thread.join(timeout=12)
             if self._thread.is_alive():
                 logger.warning(f"{self.name} alert worker thread did not exit on time and will be killed.")
 
@@ -64,9 +64,14 @@ class HTTPAlertProvider(AlertProvider):
                 logger.warning(f"HTTP {http_response.status_code} when calling {self.name} alerts API.")
 
         except ConnectionError:
+            self.status = "Error"
             logger.warning(f"Connection error when accessing {self.name} alerts API.")
         except (ConnectTimeout, ReadTimeout):
+            self.status = "Error"
             logger.warning(f"Timeout when accessing {self.name} alerts API.")
+        except JSONDecodeError:
+            self.status = "Error"
+            logger.warning(f"Invalid or empty JSON response from {self.name} alert API.")
         except Exception:
             self.status = "Error"
             logger.exception(f"Exception in HTTP JSON Alert Provider ({self.name})")

providers/callsigndata/file_download_callsign_data_provider.py

@@ -38,7 +38,7 @@ class FileDownloadCallsignDataProvider(CallsignDataProvider):
     def stop(self):
         self._stop_event.set()
         if self._thread:
-            self._thread.join(timeout=35)
+            self._thread.join(timeout=12)
             if self._thread.is_alive():
                 logger.warning(f"{self.name} callsign data worker thread did not exit on time and will be killed.")
 

providers/callsigndata/hamqth.py

@@ -124,8 +124,8 @@ class HamQTH(APIQueryCallsignDataProvider):
         lat = None
         lon = None
         if (
-            "latitude" in data
-            and "longitude" in data
+            data.get("latitude") is not None
+            and data.get("longitude") is not None
             and (float(data["latitude"]) != 0 or float(data["longitude"]) != 0)
             and -89.9 < float(data["latitude"]) < 89.9
         ):
@@ -134,7 +134,7 @@ class HamQTH(APIQueryCallsignDataProvider):
 
         # Check for sensible grids
         grid = None
-        if "grid" in data and not data["grid"].startswith("AA00"):
+        if data.get("grid") and not data["grid"].startswith("AA00"):
             grid = data["grid"]
 
         return Callsign(
@@ -147,8 +147,8 @@ class HamQTH(APIQueryCallsignDataProvider):
             latitude=lat,
             longitude=lon,
             grid=grid,
-            dxcc_id=int(data["adif"]) if "adif" in data else None,
-            cq_zone=int(data["cq"]) if "cq" in data else None,
-            itu_zone=int(data["itu"]) if "itu" in data else None,
+            dxcc_id=int(data["adif"]) if data.get("adif") is not None else None,
+            cq_zone=int(data["cq"]) if data.get("cq") is not None else None,
+            itu_zone=int(data["itu"]) if data.get("itu") is not None else None,
             location_source=LocationSourceForCallsign.HOME_QTH,
         )

providers/callsigndata/qrz.py

@@ -150,8 +150,8 @@ class QRZ(APIQueryCallsignDataProvider):
         lat = None
         lon = None
         if (
-            "latitude" in data
-            and "longitude" in data
+            data.get("latitude") is not None
+            and data.get("longitude") is not None
             and (float(data["latitude"]) != 0 or float(data["longitude"]) != 0)
             and -89.9 < float(data["latitude"]) < 89.9
         ):
@@ -160,7 +160,7 @@ class QRZ(APIQueryCallsignDataProvider):
 
         # Check for sensible grids
         grid = None
-        if "grid" in data and not data["grid"].startswith("AA00"):
+        if data.get("grid") and not data["grid"].startswith("AA00"):
             grid = data["grid"]
 
         return Callsign(
@@ -173,8 +173,8 @@ class QRZ(APIQueryCallsignDataProvider):
             latitude=lat,
             longitude=lon,
             grid=grid,
-            dxcc_id=int(data["adif"]) if "adif" in data else None,
-            cq_zone=int(data["cqzone"]) if "cqzone" in data else None,
-            itu_zone=int(data["ituzone"]) if "ituzone" in data else None,
+            dxcc_id=int(data["adif"]) if data.get("adif") is not None else None,
+            cq_zone=int(data["cqzone"]) if data.get("cqzone") is not None else None,
+            itu_zone=int(data["ituzone"]) if data.get("ituzone") is not None else None,
             location_source=LocationSourceForCallsign.HOME_QTH,
         )

providers/sigrefdata/file_download_sig_ref_data_provider.py

@@ -33,7 +33,7 @@ class FileDownloadSIGRefDataProvider(SIGRefDataProvider):
     def stop(self):
         super().stop()
         if self._thread:
-            self._thread.join(timeout=35)
+            self._thread.join(timeout=12)
             if self._thread.is_alive():
                 logger.warning(f"{self.sig_name} SIG ref data worker thread did not exit on time and will be killed.")
 

providers/sigrefdata/sig_ref_data_provider.py

@@ -36,8 +36,10 @@ class SIGRefDataProvider:
     def _add_data(self, new_data):
         """Add all the provided reference data objects to the data store."""
 
-        # with transact() batches all writes together to save making thousands of individual sqlite writes
-        with DATA_STORE.sigrefs.transact():
+        # with transact() batches all writes together to save making thousands of individual sqlite writes. However,
+        # that means that each provider holds the lock while it writes, and the default behaviour for other attempted
+        # transact()s is to fail if they can't get the lock (?!). This behaviour is fixed by retry=True.
+        with DATA_STORE.sigrefs.transact(retry=True):
             for d in new_data:
                 DATA_STORE.sigrefs.set(f"{self.sig_name}:{d.id}", d)
 

providers/solarconditions/giroionosonde.py

@@ -73,7 +73,7 @@ class GIROIonosonde(SolarConditionsProvider):
     def stop(self):
         self._stop_event.set()
         if self._thread:
-            self._thread.join(timeout=35)
+            self._thread.join(timeout=12)
             if self._thread.is_alive():
                 logger.warning("GIRO ionosonde worker thread did not exit on time and will be killed.")
 

providers/solarconditions/http_solar_conditions_provider.py

@@ -31,7 +31,7 @@ class HTTPSolarConditionsProvider(SolarConditionsProvider):
     def stop(self):
         self._stop_event.set()
         if self._thread:
-            self._thread.join(timeout=35)
+            self._thread.join(timeout=12)
             if self._thread.is_alive():
                 logger.warning(f"{self.name} solar conditions worker thread did not exit on time and will be killed.")
 

providers/solarconditions/kc2gprop.py

@@ -38,7 +38,7 @@ class KC2GProp(SolarConditionsProvider):
     def stop(self):
         self._stop_event.set()
         if self._thread:
-            self._thread.join(timeout=35)
+            self._thread.join(timeout=12)
             if self._thread.is_alive():
                 logger.warning("KC2G ionosonde worker thread did not exit on time and will be killed.")
 

providers/spot/aprsis.py

@@ -49,7 +49,7 @@ class APRSIS(SpotProvider):
         if self._aprsis:
             self._aprsis.close()
         if self._thread:
-            self._thread.join(timeout=15)
+            self._thread.join(timeout=5)
             if self._thread.is_alive():
                 logger.warning("APRS-IS worker thread did not exit on time and will be killed.")
 
@@ -69,8 +69,8 @@ class APRSIS(SpotProvider):
                 de_call=de_call,
                 de_ssid=de_ssid,
                 comment=str(data["comment"]) if "comment" in data else None,
-                dx_latitude=float(data["latitude"]) if "latitude" in data else None,
-                dx_longitude=float(data["longitude"]) if "longitude" in data else None,
+                dx_latitude=float(data["latitude"]) if data.get("latitude") is not None else None,
+                dx_longitude=float(data["longitude"]) if data.get("longitude") is not None else None,
                 time=datetime.now(pytz.UTC).timestamp(),
             )  # APRS-IS spots are live so we can assume spot time is "now"
 

providers/spot/dxcluster.py

@@ -1,7 +1,8 @@
 import logging
 import re
+import socket
 from datetime import datetime
-from threading import Event, Thread
+from threading import Event, Lock, Thread
 
 import pytz
 import telnetlib3
@@ -40,6 +41,7 @@ class DXCluster(SpotProvider):
             self._LINE_PATTERN_ALLOW_RBN if self._allow_rbn_spots else self._LINE_PATTERN_EXCLUDE_RBN
         )
         self._telnet = None
+        self._telnet_lock = Lock()
         self._thread = None
         self._stop_event = Event()
 
@@ -49,10 +51,15 @@ class DXCluster(SpotProvider):
 
     def stop(self):
         self._stop_event.set()
+        with self._telnet_lock:
             if self._telnet:
+                try:
+                    self._telnet.sock.shutdown(socket.SHUT_RDWR)
+                except (AttributeError, OSError):
+                    pass
                 self._telnet.close()
         if self._thread:
-            self._thread.join(timeout=15)
+            self._thread.join(timeout=5)
             if self._thread.is_alive():
                 logger.warning(f"DX Cluster {self._hostname} worker thread did not exit on time and will be killed.")
 
@@ -63,7 +70,14 @@ class DXCluster(SpotProvider):
                 try:
                     self.status = "Connecting"
                     logger.info(f"DX Cluster {self._hostname} connecting...")
-                    self._telnet = telnetlib3.Telnet(self._hostname, self._port)
+                    new_telnet = telnetlib3.Telnet(self._hostname, self._port)
+                    with self._telnet_lock:
+                        self._telnet = new_telnet
+                        if self._stop_event.is_set():
+                            # stop() was called while we were connecting, close the connection rather than trying to
+                            # read when we know it won't work
+                            new_telnet.close()
+                            break
                     self._telnet.read_until(self._login_prompt.encode("latin-1"))
                     self._telnet.write(f"{self._login_callsign}\n".encode("latin-1"))
                     connected = True

providers/spot/http_spot_provider.py

@@ -4,7 +4,7 @@ from threading import Event, Thread
 
 import pytz
 import requests
-from requests.exceptions import ConnectionError, ConnectTimeout, ReadTimeout
+from requests.exceptions import ConnectionError, ConnectTimeout, JSONDecodeError, ReadTimeout
 
 from core.constants import HTTP_HEADERS
 from providers.spot.spot_provider import SpotProvider
@@ -35,7 +35,7 @@ class HTTPSpotProvider(SpotProvider):
         self._stop_event.set()
         self._wakeup_event.set()
         if self._thread:
-            self._thread.join(timeout=35)
+            self._thread.join(timeout=12)
             if self._thread.is_alive():
                 logger.warning(f"{self.name} spot worker thread did not exit on time and will be killed.")
 
@@ -73,9 +73,14 @@ class HTTPSpotProvider(SpotProvider):
                 logger.warning(f"HTTP {http_response.status_code} when calling {self.name} spot API.")
 
         except ConnectionError:
+            self.status = "Error"
             logger.warning(f"Connection error when accessing {self.name} spots API.")
         except (ConnectTimeout, ReadTimeout):
+            self.status = "Error"
             logger.warning(f"Timeout when accessing {self.name} spots API.")
+        except JSONDecodeError:
+            self.status = "Error"
+            logger.warning(f"Invalid or empty JSON response from {self.name} spots API.")
         except Exception:
             self.status = "Error"
             logger.exception(f"Exception in HTTP Spot Provider ({self.name})")

providers/spot/rbn.py

@@ -1,7 +1,8 @@
 import logging
 import re
+import socket
 from datetime import datetime
-from threading import Event, Thread
+from threading import Event, Lock, Thread
 
 import pytz
 import telnetlib3
@@ -29,6 +30,7 @@ class RBN(SpotProvider):
         super().__init__(name, provider_config)
         self._port = provider_config["port"]
         self._telnet = None
+        self._telnet_lock = Lock()
         self._thread = None
         self._stop_event = Event()
 
@@ -38,10 +40,15 @@ class RBN(SpotProvider):
 
     def stop(self):
         self._stop_event.set()
+        with self._telnet_lock:
             if self._telnet:
+                try:
+                    self._telnet.sock.shutdown(socket.SHUT_RDWR)
+                except (AttributeError, OSError):
+                    pass
                 self._telnet.close()
         if self._thread:
-            self._thread.join(timeout=15)
+            self._thread.join(timeout=5)
             if self._thread.is_alive():
                 logger.warning(f"RBN (port {self._port!s}) worker thread did not exit on time and will be killed.")
 
@@ -52,7 +59,14 @@ class RBN(SpotProvider):
                 try:
                     self.status = "Connecting"
                     logger.info(f"RBN port {self._port!s} connecting...")
-                    self._telnet = telnetlib3.Telnet("telnet.reversebeacon.net", self._port)
+                    new_telnet = telnetlib3.Telnet("telnet.reversebeacon.net", self._port)
+                    with self._telnet_lock:
+                        self._telnet = new_telnet
+                        if self._stop_event.is_set():
+                            # stop() was called while we were connecting, close the connection rather than trying to
+                            # read when we know it won't work
+                            new_telnet.close()
+                            break
                     self._telnet.read_until("Please enter your call: ".encode("latin-1"))
                     self._telnet.write(f"{SERVER_OWNER_CALLSIGN}\n".encode("latin-1"))
                     connected = True

providers/spot/sse_spot_provider.py

@@ -42,7 +42,7 @@ class SSESpotProvider(SpotProvider):
                 logger.exception(f"Exception closing SSE connection for {self.name} during stop()")
 
         if self._thread:
-            self._thread.join(timeout=15)
+            self._thread.join(timeout=5)
             if self._thread.is_alive():
                 logger.warning(f"{self.name} SSE worker thread did not exit on time and will be killed.")
 

providers/spot/ukpacketnet.py

@@ -42,7 +42,7 @@ class UKPacketNet(HTTPSpotProvider):
                                 )
                                 comment = (
                                     f"{comment} {listed_port['baud']!s} baud"
-                                    if "baud" in listed_port and listed_port["baud"] > 0
+                                    if listed_port.get("baud") and listed_port["baud"] > 0
                                     else comment
                                 )
 
@@ -50,7 +50,7 @@ class UKPacketNet(HTTPSpotProvider):
                                 # very hacky but a lot of node comments contain their frequency as the first or second
                                 # word of their comment, but not in the proper data structure field.
                                 freq = (
-                                    listed_port["freq"] if "freq" in listed_port and listed_port["freq"] > 0 else None
+                                    listed_port["freq"] if listed_port.get("freq") and listed_port["freq"] > 0 else None
                                 )
                                 if not freq and comment:
                                     possible_freq = comment.split(" ")[0].upper().replace("MHZ", "")

providers/spot/websocket_spot_provider.py

@@ -34,7 +34,7 @@ class WebsocketSpotProvider(SpotProvider):
         if self._ws:
             self._ws.close()
         if self._thread:
-            self._thread.join(timeout=15)
+            self._thread.join(timeout=5)
             if self._thread.is_alive():
                 logger.warning(f"{self.name} websocket worker thread did not exit on time and will be killed.")
 

providers/spot/wwbota.py

@@ -36,7 +36,7 @@ class WWBOTA(SSESpotProvider):
             dx_call=source_spot["call"].upper(),
             de_call=source_spot["spotter"].upper(),
             freq=float(source_spot["freq"]) * 1000000,
-            mode=Mode.from_name(source_spot["mode"].upper()) if "mode" in source_spot else None,
+            mode=Mode.from_name(source_spot["mode"].upper()) if source_spot.get("mode") else None,
             comment=source_spot["comment"],
             sig="WWBOTA",
             sig_refs=refs,

providers/staticdata/file_download_static_data_provider.py

@@ -35,7 +35,7 @@ class FileDownloadStaticDataProvider(StaticDataProvider):
     def stop(self):
         self._stop_event.set()
         if self._thread:
-            self._thread.join(timeout=35)
+            self._thread.join(timeout=12)
             if self._thread.is_alive():
                 logger.warning(f"{self.name} static data worker thread did not exit on time and will be killed.")
 

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "spothole"
-version = "2.1"
+version = "2.1.2"
 authors = [
     { name = "Ian Renton", email = "ian@ianrenton.com" },
 ]

static/css/style.css

@@ -414,8 +414,8 @@ div.band-spot:hover span.band-spot-info {
 
     /* Make map stretch to horizontal screen edges */
     div#map, div#table-container, div#bands-container {
-        margin-left: -1em;
-        margin-right: -1em;
+        margin-left: -0.75rem;
+        margin-right: -0.75rem;
     }
 
     /* Avoid map page filters panel being larger than the map itself */

templates/add_spot.html

@@ -77,7 +77,7 @@
 
 </div>
 
-<script src="/static/js/add-spot.js?v=1789731849"></script>
+<script src="/static/js/add-spot.js?v=1789826851"></script>
 <script>$(document).ready(function () {
     $("#nav-link-add-spot").addClass("active");
 }); <!-- highlight active page in nav --></script>

templates/alerts.html

@@ -83,7 +83,7 @@
 
 </div>
 
-<script src="/static/js/alerts.js?v=1789731849"></script>
+<script src="/static/js/alerts.js?v=1789826852"></script>
 <script>$(document).ready(function () {
     $("#nav-link-alerts").addClass("active");
 }); <!-- highlight active page in nav --></script>

templates/bands.html

@@ -76,8 +76,8 @@
 
 </div>
 
-<script src="/static/js/spotsbandsandmap.js?v=1789731849"></script>
-<script src="/static/js/bands.js?v=1789731849"></script>
+<script src="/static/js/spotsbandsandmap.js?v=1789826851"></script>
+<script src="/static/js/bands.js?v=1789826851"></script>
 <script>$(document).ready(function () {
     $("#nav-link-bands").addClass("active");
 }); <!-- highlight active page in nav --></script>

templates/base.html

@@ -1,6 +1,6 @@
 {% extends "skeleton.html" %}
 {% block head_extra %}
-<link rel="stylesheet" href="/static/css/style.css?v=1789731849" type="text/css">
+<link rel="stylesheet" href="/static/css/style.css?v=1789826851" type="text/css">
 <link href="/static/vendor/css/bootstrap-5.3.8.min.css" rel="stylesheet">
 <link href="/static/vendor/css/fontawesome-6.7.2.min.css" rel="stylesheet">
 <link href="/static/vendor/css/solid-6.7.2.min.css" rel="stylesheet">
@@ -16,10 +16,10 @@
     window.fetchEventSource = fetchEventSource;
 </script>
 
-<script src="/static/js/utils.js?v=1789731849"></script>
-<script src="/static/js/ui-ham.js?v=1789731849"></script>
-<script src="/static/js/geo.js?v=1789731849"></script>
-<script src="/static/js/common.js?v=1789731849"></script>
+<script src="/static/js/utils.js?v=1789826851"></script>
+<script src="/static/js/ui-ham.js?v=1789826851"></script>
+<script src="/static/js/geo.js?v=1789826851"></script>
+<script src="/static/js/common.js?v=1789826851"></script>
 {% end %}
 {% block body %}
 <div class="container">

templates/conditions.html

@@ -284,7 +284,7 @@
 </div>
 
 <script src="/static/vendor/js/chart-4.4.9.umd.min.js"></script>
-<script src="/static/js/conditions.js?v=1789731849"></script>
+<script src="/static/js/conditions.js?v=1789826851"></script>
 <script>$(document).ready(function () {
     $("#nav-link-conditions").addClass("active");
 }); <!-- highlight active page in nav --></script>

templates/map.html

@@ -113,8 +113,8 @@
     const CARTODB_API_KEY = "{{ web_ui_options.get('cartodb_api_key', '') }}";
 </script>
 
-<script src="/static/js/spotsbandsandmap.js?v=1789731849"></script>
-<script src="/static/js/map.js?v=1789731849"></script>
+<script src="/static/js/spotsbandsandmap.js?v=1789826852"></script>
+<script src="/static/js/map.js?v=1789826852"></script>
 <script>$(document).ready(function () {
     $("#nav-link-map").addClass("active");
 }); <!-- highlight active page in nav --></script>

templates/spots.html

@@ -125,8 +125,8 @@
 
 </div>
 
-<script src="/static/js/spotsbandsandmap.js?v=1789731849"></script>
-<script src="/static/js/spots.js?v=1789731849"></script>
+<script src="/static/js/spotsbandsandmap.js?v=1789826851"></script>
+<script src="/static/js/spots.js?v=1789826851"></script>
 <script>$(document).ready(function () {
     $("#nav-link-spots").addClass("active");
 }); <!-- highlight active page in nav --></script>

templates/status.html

@@ -96,7 +96,7 @@
     </div>
 </div>
 
-<script src="/static/js/status.js?v=1789731849"></script>
+<script src="/static/js/status.js?v=1789826851"></script>
 <script>
     $(document).ready(function () {
         $("#nav-link-status").addClass("active");