increase wait time for llm backends

Remove sleeps and delays

lib/backend.py

@@ -66,6 +66,9 @@ class Backend:
     unsecured: bool = dataclasses.field(
         default_factory=lambda: bool(strtobool(os.environ.get("UNSECURED", "false"))),
     )
+    report_addr: str = dataclasses.field(
+        default_factory=lambda: os.environ.get("REPORT_ADDR", "https://run.vast.ai")
+    )
 
     def __post_init__(self):
         self.metrics = Metrics()
@@ -104,23 +107,19 @@ class Backend:
 
     #######################################Private#######################################
     def _fetch_pubkey(self):
-        command = ["curl", "-X", "GET", "https://run.vast.ai/pubkey/"]
+        report_addr = self.report_addr.rstrip("/")
-        result = subprocess.check_output(command, universal_newlines=True)
+        command = ["curl", "-X", "GET", f"{report_addr}/pubkey/"]
-        log.debug("public key:")
+        try:
-        log.debug(result)
+            result = subprocess.check_output(command, universal_newlines=True)
-        key = None
+            log.debug("public key:")
-        for _ in range(5):
+            log.debug(result)
-            try:
+            key = RSA.import_key(result)
-                key = RSA.import_key(result)
+            if key is not None:
-                break
+                return key
-            except ValueError as e:
+        except (ValueError , subprocess.CalledProcessError) as e:
-                log.debug(f"Error downloading key: {e}")
+            log.debug(f"Error downloading key: {e}")
-                time.sleep(15)
+        self.backend_errored("Failed to get autoscaler pubkey")
-        if key is None:
+       
-            self._total_pubkey_fetch_errors += 1
-            if self._total_pubkey_fetch_errors >= MAX_PUBKEY_FETCH_ATTEMPTS:
-                self.backend_errored("Failed to get autoscaler pubkey")
-        return key
 
     async def __handle_request(
         self,

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"

workers/comfyui-json/client.py

@@ -98,6 +98,7 @@ def call_text2image_workflow(
         endpoint=route_response["endpoint"],
         reqnum=route_response["reqnum"],
         url=route_response["url"],
+        request_idx=route_response["request_idx"],
     )
     
     # Build the payload for the worker request

workers/comfyui/client.py

@@ -82,6 +82,7 @@ def call_custom_workflow_for_sd3(
         endpoint=message["endpoint"],
         reqnum=message["reqnum"],
         url=message["url"],
+        request_idx=message["request_idx"],
     )
     workflow = {
         "3": {

workers/openai/server.py

@@ -43,6 +43,7 @@ backend = Backend(
             for error_msg in MODEL_SERVER_ERROR_LOG_MSGS
         ],
     ],
+    max_wait_time=600
 )
 
 

workers/tgi/server.py

@@ -113,6 +113,7 @@ backend = Backend(
             for error_msg in MODEL_SERVER_ERROR_LOG_MSGS
         ],
     ],
+    max_wait_time=600
 )