feat(p2p): Relay all recently-seen peers for neighbors.

Author: msbrogli

hathor/p2p/manager.py

@@ -150,6 +150,9 @@ def __init__(self,
         # List of known peers.
         self.peer_storage = PeerStorage()  # dict[string (peer.id), PeerId]
 
+        # Maximum unseen time before removing a peer (seconds).
+        self.max_peer_unseen_dt: float = 30 * 60   # 30-minutes
+
         # A timer to try to reconnect to the disconnect known peers.
         self.lc_reconnect = LoopingCall(self.reconnect_to_all)
         self.lc_reconnect.clock = self.reactor
@@ -341,6 +344,7 @@ def on_peer_ready(self, protocol: HathorProtocol) -> None:
         assert protocol.peer is not None
         protocol.peer = self.peer_storage.add_or_merge(protocol.peer)
         assert protocol.peer.id is not None
+        protocol.peer.last_seen = self.reactor.seconds()
 
         self.handshaking_peers.remove(protocol)
         self.received_peer_storage.pop(protocol.peer.id, None)
@@ -378,7 +382,7 @@ def on_peer_ready(self, protocol: HathorProtocol) -> None:
             if conn != protocol:
                 assert conn.state is not None
                 assert isinstance(conn.state, ReadyState)
-                conn.state.send_peers([protocol])
+                conn.state.send_peers([protocol.my_peer])
 
     def on_peer_disconnect(self, protocol: HathorProtocol) -> None:
         """Called when a peer disconnect."""
@@ -439,12 +443,31 @@ def on_receive_peer(self, peer: PeerId, origin: Optional[ReadyState] = None) ->
         peer = self.received_peer_storage.add_or_merge(peer)
         self.connect_to_if_not_connected(peer, 0)
 
+    def peers_cleanup(self) -> None:
+        """Clean up aged peers."""
+        now = self.reactor.seconds()
+        to_be_removed: list[PeerId] = []
+        for peer in self.peer_storage.values():
+            assert peer.last_seen is not None
+            dt = now - peer.last_seen
+            if dt <= self.max_peer_unseen_dt:
+                continue
+            assert peer.id is not None
+            if self.is_peer_connected(peer.id):
+                self.log.warn('Aged peer is connected. It should never happen.', peer_id=peer.id)
+                continue
+            to_be_removed.append(peer)
+
+        for peer in to_be_removed:
+            self.peer_storage.remove(peer)
+
     def reconnect_to_all(self) -> None:
         """ It is called by the `lc_reconnect` timer and tries to connect to all known
         peers.
 
         TODO(epnichols): Should we always connect to *all*? Should there be a max #?
         """
+        self.peers_cleanup()
         # when we have no connected peers left, run the discovery process again
         assert self.manager is not None
         if len(self.connected_peers) < 1:

hathor/p2p/peer_id.py

@@ -15,6 +15,7 @@
 import base64
 import hashlib
 from enum import Enum
+from math import inf
 from typing import TYPE_CHECKING, Any, Generator, Optional, cast
 
 from cryptography import x509
@@ -61,6 +62,7 @@ class PeerId:
     retry_timestamp: int    # should only try connecting to this peer after this timestamp
     retry_interval: int     # how long to wait for next connection retry. It will double for each failure
     retry_attempts: int     # how many retries were made
+    last_seen: float        # last time this peer was seen
     flags: set[str]
 
     def __init__(self, auto_generate_keys: bool = True) -> None:
@@ -73,6 +75,7 @@ def __init__(self, auto_generate_keys: bool = True) -> None:
         self.retry_timestamp = 0
         self.retry_interval = 5
         self.retry_attempts = 0
+        self.last_seen = inf
         self.flags = set()
         self._certificate_options: Optional[CertificateOptions] = None
 

hathor/p2p/protocol.py

@@ -298,7 +298,10 @@ def recv_message(self, cmd: ProtocolMessages, payload: str) -> Optional[Deferred
         """
         assert self.state is not None
 
-        self.last_message = self.reactor.seconds()
+        now = self.reactor.seconds()
+        self.last_message = now
+        if self.peer is not None:
+            self.peer.last_seen = now
         self.reset_idle_timeout()
 
         if not self.ratelimit.add_hit(self.RateLimitKeys.GLOBAL):

hathor/p2p/resources/status.py

@@ -12,8 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import time
-
 import hathor
 from hathor.api_util import Resource, set_cors
 from hathor.cli.openapi_files.register import register_resource
@@ -34,11 +32,14 @@ class StatusResource(Resource):
     def __init__(self, manager):
         self._settings = get_settings()
         self.manager = manager
+        self.reactor = manager.reactor
 
     def render_GET(self, request):
         request.setHeader(b'content-type', b'application/json; charset=utf-8')
         set_cors(request, 'GET')
 
+        now = self.reactor.seconds()
+
         connecting_peers = []
         # TODO: refactor as not to use a private item
         for endpoint, deferred in self.manager.connections.connecting_peers.items():
@@ -53,7 +54,7 @@ def render_GET(self, request):
             handshaking_peers.append({
                 'address': '{}:{}'.format(remote.host, remote.port),
                 'state': conn.state.state_name,
-                'uptime': time.time() - conn.connection_time,
+                'uptime': now - conn.connection_time,
                 'app_version': conn.app_version,
             })
 
@@ -65,12 +66,13 @@ def render_GET(self, request):
             connected_peers.append({
                 'id': conn.peer.id,
                 'app_version': conn.app_version,
-                'uptime': time.time() - conn.connection_time,
+                'current_time': now,
+                'uptime': now - conn.connection_time,
                 'address': '{}:{}'.format(remote.host, remote.port),
                 'state': conn.state.state_name,
                 # 'received_bytes': conn.received_bytes,
                 'rtt': list(conn.state.rtt_window),
-                'last_message': time.time() - conn.last_message,
+                'last_message': now - conn.last_message,
                 'plugins': status,
                 'warning_flags': [flag.value for flag in conn.warning_flags],
                 'protocol_version': str(conn.sync_version),
@@ -82,6 +84,7 @@ def render_GET(self, request):
             known_peers.append({
                 'id': peer.id,
                 'entrypoints': peer.entrypoints,
+                'last_seen': now - peer.last_seen,
                 'flags': [flag.value for flag in peer.flags],
             })
 
@@ -103,7 +106,7 @@ def render_GET(self, request):
                 'app_version': app,
                 'state': self.manager.state.value,
                 'network': self.manager.network,
-                'uptime': time.time() - self.manager.start_time,
+                'uptime': now - self.manager.start_time,
                 'entrypoints': self.manager.connections.my_peer.entrypoints,
             },
             'peers_whitelist': self.manager.peers_whitelist,

hathor/p2p/states/ready.py

@@ -48,6 +48,11 @@ def __init__(self, protocol: 'HathorProtocol') -> None:
         self.lc_ping = LoopingCall(self.send_ping_if_necessary)
         self.lc_ping.clock = self.reactor
 
+        # LC to send GET_PEERS every once in a while.
+        self.lc_get_peers = LoopingCall(self.send_get_peers)
+        self.lc_get_peers.clock = self.reactor
+        self.get_peers_interval: int = 5   # Once every 5 minutes.
+
         # Minimum interval between PING messages (in seconds).
         self.ping_interval: int = 3
 
@@ -108,6 +113,8 @@ def on_enter(self) -> None:
             self.protocol.on_peer_ready()
 
         self.lc_ping.start(1, now=False)
+
+        self.lc_get_peers.start(self.get_peers_interval, now=False)
         self.send_get_peers()
 
         if self.lc_get_best_blockchain is not None:
@@ -119,6 +126,9 @@ def on_exit(self) -> None:
         if self.lc_ping.running:
             self.lc_ping.stop()
 
+        if self.lc_get_peers.running:
+            self.lc_get_peers.stop()
+
         if self.lc_get_best_blockchain is not None and self.lc_get_best_blockchain.running:
             self.lc_get_best_blockchain.stop()
 
@@ -144,22 +154,20 @@ def handle_get_peers(self, payload: str) -> None:
         """ Executed when a GET-PEERS command is received. It just responds with
         a list of all known peers.
         """
-        if self.protocol.connections:
-            self.send_peers(self.protocol.connections.iter_ready_connections())
+        for peer in self.protocol.connections.peer_storage.values():
+            self.send_peers([peer])
 
-    def send_peers(self, connections: Iterable['HathorProtocol']) -> None:
-        """ Send a PEERS command with a list of all connected peers.
+    def send_peers(self, peer_list: Iterable['PeerId']) -> None:
+        """ Send a PEERS command with a list of peers.
         """
-        peers = []
-        for conn in connections:
-            assert conn.peer is not None
-            peers.append({
-                'id': conn.peer.id,
-                'entrypoints': conn.peer.entrypoints,
-                'last_message': conn.last_message,
+        data = []
+        for peer in peer_list:
+            data.append({
+                'id': peer.id,
+                'entrypoints': peer.entrypoints,
             })
-        self.send_message(ProtocolMessages.PEERS, json_dumps(peers))
-        self.log.debug('send peers', peers=peers)
+        self.send_message(ProtocolMessages.PEERS, json_dumps(data))
+        self.log.debug('send peers', peers=data)
 
     def handle_peers(self, payload: str) -> None:
         """ Executed when a PEERS command is received. It updates the list
@@ -203,6 +211,8 @@ def handle_ping(self, payload: str) -> None:
 
     def handle_pong(self, payload: str) -> None:
         """Executed when a PONG message is received."""
+        assert self.protocol.peer is not None
+        self.protocol.peer.last_seen = self.reactor.seconds()
         if self.ping_start_time is None:
             # This should never happen.
             return