bump up version minor number

lib/metrics.py

@@ -146,6 +146,7 @@ class Metrics:
     def _set_mtoken(self, mtoken: str) -> None:
         self.mtoken = mtoken
 
+
     #######################################Private#######################################
 
     async def __send_delete_requests_and_reset(self):
@@ -280,7 +281,6 @@ class Metrics:
 
         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.last_metric_update = time.time()

lib/server.py

@@ -3,17 +3,15 @@ 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):
-    try:
     log.debug("getting certificate...")
     use_ssl = os.environ.get("USE_SSL", "false") == "true"
     if use_ssl is True:
@@ -40,21 +38,3 @@ def start_server(backend: Backend, routes: List[web.RouteDef], **kwargs):
         await gather(site.start(), backend._start_tracking())
 
     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

@@ -41,14 +41,6 @@ echo_var DEBUG_LOG
 echo_var PYWORKER_LOG
 echo_var MODEL_LOG
 
-# if instance is rebooted, we want to clear out the log file so pyworker doesn't read lines
-# from the run prior to reboot. past logs are saved in $MODEL_LOG.old for debugging only
-if [ -e "$MODEL_LOG" ]; then
-    echo "Rotating model log at $MODEL_LOG to $MODEL_LOG.old"
-    cat "$MODEL_LOG" >> "$MODEL_LOG.old" 
-    : > "$MODEL_LOG"
-fi
-
 # Populate /etc/environment with quoted values
 if ! grep -q "VAST" /etc/environment; then
     env -0 | grep -zEv "^(HOME=|SHLVL=)|CONDA" | while IFS= read -r -d '' line; do
@@ -132,43 +124,9 @@ cd "$SERVER_DIR"
 
 echo "launching PyWorker server"
 
-set +e
+# if instance is rebooted, we want to clear out the log file so pyworker doesn't read lines
-python3 -m "workers.$BACKEND.server" |& tee -a "$PYWORKER_LOG"
+# from the run prior to reboot. past logs are saved in $MODEL_LOG.old for debugging only
-PY_STATUS=${PIPESTATUS[0]}
+[ -e "$MODEL_LOG" ] && cat "$MODEL_LOG" >> "$MODEL_LOG.old" && : > "$MODEL_LOG"
-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
 
+(python3 -m "workers.$BACKEND.server" |& tee -a "$PYWORKER_LOG") &
 echo "launching PyWorker server done"

workers/openai/server.py

@@ -11,7 +11,6 @@ MODEL_SERVER_START_LOG_MSG = [
     "llama runner started",  # Ollama
     '"message":"Connected","target":"text_generation_router"',  # TGI
     '"message":"Connected","target":"text_generation_router::server"',  # TGI
-    "main: model loaded" # llama.cpp
 ]
 
 MODEL_SERVER_ERROR_LOG_MSGS = [
@@ -35,7 +34,6 @@ backend = Backend(
     model_server_url=os.environ["MODEL_SERVER_URL"],
     model_log_file=os.environ["MODEL_LOG"],
     allow_parallel_requests=True,
-    max_wait_time=600.0,
     benchmark_handler=CompletionsHandler(benchmark_runs=3, benchmark_words=256),
     log_actions=[
         *[(LogAction.ModelLoaded, info_msg) for info_msg in MODEL_SERVER_START_LOG_MSG],