remove redis pubsub from pyworker

Redis subscriber queue patch

lib/backend.py

@@ -91,17 +91,6 @@ class Backend:
         timeout = ClientTimeout(total=None)
         return ClientSession(self.model_server_url, timeout=timeout, connector=connector)
 
-    async def _worker(self):
-        while True:
-            handler, request, fut = await self.request_queue.get()
-            try:
-                res = await self.__process_request(handler, request)
-                fut.set_result(res)
-            except Exception as e:
-                fut.set_exception(e)
-            finally:
-                self.request_queue.task_done()
-
     def create_handler(
         self,
         handler: EndpointHandler[ApiPayload_T],
@@ -297,7 +286,7 @@ class Backend:
         message = {
             key: value
             for (key, value) in (dataclasses.asdict(auth_data).items())
-            if key != "signature"
+            if key != "signature" and key != "__request_id"
         }
         if auth_data.reqnum < (self.reqnum - MSG_HISTORY_LEN):
             log.debug(
@@ -307,7 +296,7 @@ class Backend:
         elif message in self.msg_history:
             log.debug(f"message: {message} already in message history")
             return False
-        elif verify_signature(json.dumps(message, indent=4), auth_data.signature):
+        elif verify_signature(json.dumps(message, indent=4, sort_keys=True), auth_data.signature):
             self.reqnum = max(auth_data.reqnum, self.reqnum)
             self.msg_history.append(message)
             self.msg_history = self.msg_history[-MSG_HISTORY_LEN:]

lib/data_types.py

@@ -65,12 +65,12 @@ class ApiPayload(ABC):
 class AuthData:
     """data used to authenticate requester"""
 
-    signature: str
     cost: str
     endpoint: str
     reqnum: int
-    url: str
     request_idx: int
+    signature: str
+    url: str
 
     @classmethod
     def from_json_msg(cls, json_msg: Dict[str, Any]):
@@ -190,10 +190,11 @@ class SystemMetrics:
         self.additional_disk_usage = disk_usage - self.last_disk_usage
         self.last_disk_usage = disk_usage
 
-    def reset(self):
+    def reset(self, expected: float | None) -> None:
         # autoscaler excepts model_loading_time to be populated only once, when the instance has
         # finished benchmarking and is ready to receive requests. This applies to restarted instances
         # as well: they should send model_loading_time once when they are done loading
+        if self.model_loading_time == expected:
             self.model_loading_time = None
 
 

lib/metrics.py

@@ -180,6 +180,10 @@ class Metrics:
             return  # nothing to do
 
         for report_addr in self.report_addr:
+            # TODO: Add a Redis subscriber queue for delete_requests
+            if report_addr == "https://cloud.vast.ai/api/v0":
+                # Patch: ignore the Redis API report_addr
+                continue
             sent_success = True
             sent_failed  = True
 
@@ -200,11 +204,13 @@ class Metrics:
 
     async def __send_metrics_and_reset(self):
 
+        loadtime_snapshot = self.system_metrics.model_loading_time
+
         def compute_autoscaler_data() -> AutoScalerData:
             return AutoScalerData(
                 id=self.id,
                 version=self.version,
-                loadtime=(self.system_metrics.model_loading_time or 0.0),
+                loadtime=(loadtime_snapshot or 0.0), 
                 new_load=self.model_metrics.workload_processing,
                 cur_load=self.model_metrics.cur_load,
                 rej_load=self.model_metrics.workload_rejected,
@@ -252,11 +258,15 @@ class Metrics:
 
         self.system_metrics.update_disk_usage()
 
+        sent = False
         for report_addr in self.report_addr:
-            success = await send_data(report_addr)
+            if await send_data(report_addr):
-            if success is True:
+                sent = True
                 break
+
+        if sent:
+            # clear the one-shot loadtime only if we actually sent *this* value
+            self.system_metrics.reset(expected=loadtime_snapshot)
             self.update_pending = False
             self.model_metrics.reset()
-        self.system_metrics.reset()
             self.last_metric_update = time.time()

start_server.sh

@@ -9,7 +9,7 @@ ENV_PATH="$WORKSPACE_DIR/worker-env"
 DEBUG_LOG="$WORKSPACE_DIR/debug.log"
 PYWORKER_LOG="$WORKSPACE_DIR/pyworker.log"
 
-REPORT_ADDR="${REPORT_ADDR:-https://cloud.vast.ai/api/v0,https://run.vast.ai}"
+REPORT_ADDR="${REPORT_ADDR:-https://run.vast.ai}"
 USE_SSL="${USE_SSL:-true}"
 WORKER_PORT="${WORKER_PORT:-3000}"
 mkdir -p "$WORKSPACE_DIR"