remove test bugs

Use mtoken

lib/backend.py

@@ -30,7 +30,7 @@ from lib.data_types import (
     BenchmarkResult
 )
 
-VERSION = "0.1.0"
+VERSION = "0.2.0"
 
 MSG_HISTORY_LEN = 100
 log = logging.getLogger(__file__)
@@ -66,10 +66,17 @@ 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")
+    )
+    mtoken: str = dataclasses.field(
+        default_factory=lambda: os.environ.get("MASTER_TOKEN", "")
+    )
 
     def __post_init__(self):
         self.metrics = Metrics()
         self.metrics._set_version(self.version)
+        self.metrics._set_mtoken(self.mtoken)
         self._total_pubkey_fetch_errors = 0
         self._pubkey = self._fetch_pubkey()
         self.__start_healthcheck: bool = False
@@ -104,23 +111,19 @@ class Backend:
 
     #######################################Private#######################################
     def _fetch_pubkey(self):
-        command = ["curl", "-X", "GET", "https://run.vast.ai/pubkey/"]
-        result = subprocess.check_output(command, universal_newlines=True)
-        log.debug("public key:")
-        log.debug(result)
-        key = None
-        for _ in range(5):
-            try:
-                key = RSA.import_key(result)
-                break
-            except ValueError as e:
-                log.debug(f"Error downloading key: {e}")
-                time.sleep(15)
-        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
+        report_addr = self.report_addr.rstrip("/")
+        command = ["curl", "-X", "GET", f"{report_addr}/pubkey/"]
+        try:
+            result = subprocess.check_output(command, universal_newlines=True)
+            log.debug("public key:")
+            log.debug(result)
+            key = RSA.import_key(result)
+            if key is not None:
+                return key
+        except (ValueError , subprocess.CalledProcessError) as e:
+            log.debug(f"Error downloading key: {e}")
+        self.backend_errored("Failed to get autoscaler pubkey")
+       
 
     async def __handle_request(
         self,
@@ -286,7 +289,7 @@ 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(
@@ -296,7 +299,7 @@ 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:]
@@ -315,10 +318,10 @@ class Backend:
                 with open(BENCHMARK_INDICATOR_FILE, "r") as f:
                     log.debug("already ran benchmark")
                     # trigger model load
-                    payload = self.benchmark_handler.make_benchmark_payload()
-                    _ = await self.__call_api(
-                        handler=self.benchmark_handler, payload=payload
-                    )
+                    # payload = self.benchmark_handler.make_benchmark_payload()
+                    # _ = await self.__call_api(
+                    #     handler=self.benchmark_handler, payload=payload
+                    # )
                     return float(f.readline())
             except FileNotFoundError:
                 pass
@@ -393,7 +396,7 @@ class Backend:
                         )
                         # some backends need a few seconds after logging successful startup before
                         # they can begin accepting requests
-                        await sleep(5)
+                        # await sleep(5)
                         try:
                             max_throughput = await run_benchmark()
                             self.__start_healthcheck = True

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]):
@@ -286,6 +286,7 @@ class AutoScalerData:
     """Data that is reported to autoscaler"""
 
     id: int
+    mtoken: str
     version: str
     loadtime: float
     cur_load: float

lib/metrics.py

@@ -28,6 +28,7 @@ def get_url() -> str:
 @dataclass
 class Metrics:
     version: str = "0"
+    mtoken: str = ""
     last_metric_update: float = 0.0
     last_request_served: float = 0.0
     update_pending: bool = False
@@ -142,12 +143,16 @@ class Metrics:
     def _set_version(self, version: str) -> None:
         self.version = version
 
+    def _set_mtoken(self, mtoken: str) -> None:
+        self.mtoken = mtoken
+
     #######################################Private#######################################
 
     async def __send_delete_requests_and_reset(self):
         async def post(report_addr: str, idxs: list[int], success_flag: bool) -> bool:
             data = {
                 "worker_id": self.id,
+                "mtoken": self.mtoken,
                 "request_idxs": idxs,
                 "success": success_flag,
             }
@@ -180,6 +185,10 @@ 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
 
@@ -205,6 +214,7 @@ class Metrics:
         def compute_autoscaler_data() -> AutoScalerData:
             return AutoScalerData(
                 id=self.id,
+                mtoken=self.mtoken,
                 version=self.version,
                 loadtime=(loadtime_snapshot or 0.0), 
                 new_load=self.model_metrics.workload_processing,
@@ -224,17 +234,25 @@ class Metrics:
 
         async def send_data(report_addr: str) -> bool:
             data = compute_autoscaler_data()
-            full_path = report_addr.rstrip("/") + "/worker_status/"
+            log_data = asdict(data)
+            def obfuscate(secret: str) -> str:
+                if secret is None:
+                    return ""
+                return secret[:7] + "..." if len(secret) > 7 else ("*" * len(secret))
+            
+            log_data["mtoken"] = obfuscate(log_data.get("mtoken"))
             log.debug(
                 "\n".join(
                     [
                         "#" * 60,
                         f"sending data to autoscaler",
-                        f"{json.dumps((asdict(data)), indent=2)}",
+                        f"{json.dumps(log_data, indent=2)}",
                         "#" * 60,
                     ]
                 )
             )
+
+            full_path = report_addr.rstrip("/") + "/worker_status/"
             for attempt in range(1, 4):
                 try:
                     session = await self.http()

lib/server.py

@@ -3,38 +3,58 @@ import logging
 from typing import List
 import ssl
 from asyncio import run, gather
-
+import asyncio
 
 from lib.backend import Backend
+from lib.metrics import Metrics
 from aiohttp import web
 
 log = logging.getLogger(__file__)
 
 
 def start_server(backend: Backend, routes: List[web.RouteDef], **kwargs):
-    log.debug("getting certificate...")
-    use_ssl = os.environ.get("USE_SSL", "false") == "true"
-    if use_ssl is True:
-        ssl_context = ssl.create_default_context(ssl.Purpose.CLIENT_AUTH)
-        ssl_context.load_cert_chain(
-            certfile="/etc/instance.crt",
-            keyfile="/etc/instance.key",
-        )
-    else:
-        ssl_context = None
+    try:
+        log.debug("getting certificate...")
+        use_ssl = os.environ.get("USE_SSL", "false") == "true"
+        if use_ssl is True:
+            ssl_context = ssl.create_default_context(ssl.Purpose.CLIENT_AUTH)
+            ssl_context.load_cert_chain(
+                certfile="/etc/instance.crt",
+                keyfile="/etc/instance.key",
+            )
+        else:
+            ssl_context = None
 
-    async def main():
-        log.debug("starting server...")
-        app = web.Application()
-        app.add_routes(routes)
-        runner = web.AppRunner(app)
-        await runner.setup()
-        site = web.TCPSite(
-            runner,
-            ssl_context=ssl_context,
-            port=int(os.environ["WORKER_PORT"]),
-            **kwargs
-        )
-        await gather(site.start(), backend._start_tracking())
+        async def main():
+            log.debug("starting server...")
+            app = web.Application()
+            app.add_routes(routes)
+            runner = web.AppRunner(app)
+            await runner.setup()
+            site = web.TCPSite(
+                runner,
+                ssl_context=ssl_context,
+                port=int(os.environ["WORKER_PORT"]),
+                **kwargs
+            )
+            await gather(site.start(), backend._start_tracking())
 
-    run(main())
+        run(main())
+
+    except Exception as e:
+        err_msg = f"PyWorker failed to launch: {e}"
+        log.error(err_msg)
+
+        async def beacon():
+            metrics = Metrics()
+            metrics._set_version(getattr(backend, "version", "0"))
+            metrics._set_mtoken(getattr(backend, "mtoken", ""))
+            try:
+                while True:
+                    metrics._model_errored(err_msg)
+                    await metrics._Metrics__send_metrics_and_reset()
+                    await asyncio.sleep(10)
+            finally:
+                await metrics.aclose()
+
+        run(beacon())

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"
@@ -128,5 +128,44 @@ echo "launching PyWorker server"
 # from the run prior to reboot. past logs are saved in $MODEL_LOG.old for debugging only
 [ -e "$MODEL_LOG" ] && cat "$MODEL_LOG" >> "$MODEL_LOG.old" && : > "$MODEL_LOG"
 
-(python3 -m "workers.$BACKEND.server" |& tee -a "$PYWORKER_LOG") &
+
+set +e
+python3 -m "workers.$BACKEND.server" |& tee -a "$PYWORKER_LOG"
+PY_STATUS=${PIPESTATUS[0]}
+set -e
+
+if [ "${PY_STATUS}" -ne 0 ]; then
+  echo "PyWorker exited with status ${PY_STATUS}; notifying autoscaler..."
+  ERROR_MSG="PyWorker exited: code ${PY_STATUS}"
+  MTOKEN="${MASTER_TOKEN:-}"
+  VERSION="${PYWORKER_VERSION:-0}"
+
+  IFS=',' read -r -a REPORT_ADDRS <<< "${REPORT_ADDR}"
+  for addr in "${REPORT_ADDRS[@]}"; do
+    curl -sS -X POST -H 'Content-Type: application/json' \
+      -d "$(cat <<JSON
+{
+  "id": ${CONTAINER_ID:-0},
+  "mtoken": "${MTOKEN}",
+  "version": "${VERSION}",
+  "loadtime": 0,
+  "new_load": 0,
+  "cur_load": 0,
+  "rej_load": 0,
+  "max_perf": 0,
+  "cur_perf": 0,
+  "error_msg": "${ERROR_MSG}",
+  "num_requests_working": 0,
+  "num_requests_recieved": 0,
+  "additional_disk_usage": 0,
+  "working_request_idxs": [],
+  "cur_capacity": 0,
+  "max_capacity": 0,
+  "url": "${URL}"
+}
+JSON
+)" "${addr%/}/worker_status/" || true
+  done
+fi
+
 echo "launching PyWorker server done"

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": {