5305: Added num_retries to all execute() calls. Refactored
[arvados.git] / sdk / python / arvados / events.py
index df64b4e241ec6fdc9524e57599aa6e32ffd5a206..09f2a871a966522201f5808e09fb37a820b1baf5 100644 (file)
@@ -1,23 +1,34 @@
-from ws4py.client.threadedclient import WebSocketClient
-import threading
+import arvados
+import config
+import errors
+
+import logging
 import json
-import os
+import threading
 import time
-import ssl
+import os
 import re
-import config
-import logging
+import ssl
+from ws4py.client.threadedclient import WebSocketClient
 
 _logger = logging.getLogger('arvados.events')
 
 class EventClient(WebSocketClient):
     def __init__(self, url, filters, on_event):
-        ssl_options = None
-        if re.match(r'(?i)^(true|1|yes)$',
-                    config.get('ARVADOS_API_HOST_INSECURE', 'no')):
-            ssl_options={'cert_reqs': ssl.CERT_NONE}
+        # Prefer system's CA certificates (if available)
+        ssl_options = {}
+        certs_path = '/etc/ssl/certs/ca-certificates.crt'
+        if os.path.exists(certs_path):
+            ssl_options['ca_certs'] = certs_path
+        if config.flag_is_true('ARVADOS_API_HOST_INSECURE'):
+            ssl_options['cert_reqs'] = ssl.CERT_NONE
         else:
-            ssl_options={'cert_reqs': ssl.CERT_REQUIRED}
+            ssl_options['cert_reqs'] = ssl.CERT_REQUIRED
+
+        # Warning: If the host part of url resolves to both IPv6 and
+        # IPv4 addresses (common with "localhost"), only one of them
+        # will be attempted -- and it might not be the right one. See
+        # ws4py's WebSocketBaseClient.__init__.
         super(EventClient, self).__init__(url, ssl_options=ssl_options)
         self.filters = filters
         self.on_event = on_event
@@ -48,20 +59,27 @@ class PollClient(threading.Thread):
     def __init__(self, api, filters, on_event, poll_time):
         super(PollClient, self).__init__()
         self.api = api
-        self.filters = [filters]
-        self.on_event = on_event
-        items = self.api.logs().list(limit=1, order="id desc", filters=filters).execute()['items']
-        if len(items) > 0:
-            self.id = items[0]["id"]
+        if filters:
+            self.filters = [filters]
         else:
-            self.id = 0
+            self.filters = [[]]
+        self.on_event = on_event
         self.poll_time = poll_time
+        self.daemon = True
         self.stop = threading.Event()
-        self.on_event({'status': 200})
 
     def run(self):
+        self.id = 0
+        for f in self.filters:
+            items = self.api.logs().list(limit=1, order="id desc", filters=f).execute()['items']
+            if items:
+                if items[0]['id'] > self.id:
+                    self.id = items[0]['id']
+
+        self.on_event({'status': 200})
+
         while not self.stop.isSet():
-            max_id = 0
+            max_id = self.id
             for f in self.filters:
                 items = self.api.logs().list(order="id asc", filters=f+[["id", ">", str(self.id)]]).execute()['items']
                 for i in items:
@@ -71,9 +89,21 @@ class PollClient(threading.Thread):
             self.id = max_id
             self.stop.wait(self.poll_time)
 
+    def run_forever(self):
+        # Have to poll here, otherwise KeyboardInterrupt will never get processed.
+        while not self.stop.is_set():
+            self.stop.wait(1)
+
     def close(self):
         self.stop.set()
-        self.join()
+        try:
+            self.join()
+        except RuntimeError:
+            # "join() raises a RuntimeError if an attempt is made to join the
+            # current thread as that would cause a deadlock. It is also an
+            # error to join() a thread before it has been started and attempts
+            # to do so raises the same exception."
+            pass
 
     def subscribe(self, filters):
         self.on_event({'status': 200})
@@ -82,23 +112,42 @@ class PollClient(threading.Thread):
     def unsubscribe(self, filters):
         del self.filters[self.filters.index(filters)]
 
+
+def _subscribe_websocket(api, filters, on_event):
+    endpoint = api._rootDesc.get('websocketUrl', None)
+    if not endpoint:
+        raise errors.FeatureNotEnabledError(
+            "Server does not advertise a websocket endpoint")
+    uri_with_token = "{}?api_token={}".format(endpoint, api.api_token)
+    client = EventClient(uri_with_token, filters, on_event)
+    ok = False
+    try:
+        client.connect()
+        ok = True
+        return client
+    finally:
+        if not ok:
+            client.close_connection()
+
 def subscribe(api, filters, on_event, poll_fallback=15):
-    ws = None
-    if 'websocketUrl' in api._rootDesc:
-        try:
-            url = "{}?api_token={}".format(api._rootDesc['websocketUrl'], api.api_token)
-            ws = EventClient(url, filters, on_event)
-            ws.connect()
-            return ws
-        except Exception as e:
-            _logger.warn("Got exception %s trying to connect to web sockets at %s" % (e, api._rootDesc['websocketUrl']))
-            if ws:
-                ws.close_connection()
-    if poll_fallback:
-        _logger.warn("Web sockets not available, falling back to log table polling")
-        p = PollClient(api, filters, on_event, poll_fallback)
-        p.start()
-        return p
-    else:
-        _logger.error("Web sockets not available")
-        return None
+    """
+    :api:
+      a client object retrieved from arvados.api(). The caller should not use this client object for anything else after calling subscribe().
+    :filters:
+      Initial subscription filters.
+    :on_event:
+      The callback when a message is received.
+    :poll_fallback:
+      If websockets are not available, fall back to polling every N seconds.  If poll_fallback=False, this will return None if websockets are not available.
+    """
+
+    if not poll_fallback:
+        return _subscribe_websocket(api, filters, on_event)
+
+    try:
+        return _subscribe_websocket(api, filters, on_event)
+    except Exception as e:
+        _logger.warn("Falling back to polling after websocket error: %s" % e)
+    p = PollClient(api, filters, on_event, poll_fallback)
+    p.start()
+    return p