From a12523b1d29c8a1a8ee2e402c67c73992fcb8e22 Mon Sep 17 00:00:00 2001
From: Lucas Armand <lucas@vast.ai>
Date: Tue, 11 Nov 2025 17:41:12 -0800
Subject: [PATCH] Added bad code to tgi server to test

---
 lib/server.py         | 71 ++++++++++++++++++++++++++++---------------
 start_server.sh       | 45 ++++++++++++++++++++++++++-
 workers/tgi/server.py |  1 +
 3 files changed, 91 insertions(+), 26 deletions(-)

diff --git a/lib/server.py b/lib/server.py
index b21c880..25250ea 100644
--- a/lib/server.py
+++ b/lib/server.py
@@ -3,38 +3,59 @@ 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)
+            raise Exception("Oh no the SSL cert is gone!")
+            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.__send_metrics_and_reset() 
+                    await asyncio.sleep(10)
+            finally:
+                await metrics.aclose()
+
+        run(beacon())
diff --git a/start_server.sh b/start_server.sh
index edc16a4..87c7702 100755
--- a/start_server.sh
+++ b/start_server.sh
@@ -128,5 +128,48 @@ 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") &
+
+# Run the worker in foreground so we can detect non-zero exit and report it
+python3 -m "workers.$BACKEND.server" |& tee -a "$PYWORKER_LOG"
+STATUS=$?
+
+if [ $STATUS -ne 0 ]; then
+  echo "PyWorker exited with status $STATUS; notifying autoscaler..."
+
+  ERROR_MSG="PyWorker exited: code ${STATUS}"
+  MTOKEN="${MASTER_TOKEN:-}"
+  VERSION="${PYWORKER_VERSION:-0}"
+
+  # Comma-separated REPORT_ADDR is supported
+  IFS=',' read -r -a REPORT_ADDRS <<< "${REPORT_ADDR}"
+  for addr in "${REPORT_ADDRS[@]}"; do
+    # minimal, schema-compatible payload
+    curl -sS -X POST -H 'Content-Type: application/json' \
+      -d "$(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": ""
+}
+JSON
+)" "${addr%/}/worker_status/" || true
+  done
+  # Optional: exit non-zero to let the supervisor/container runtime handle restarts
+  exit $STATUS
+fi
+
 echo "launching PyWorker server done"
diff --git a/workers/tgi/server.py b/workers/tgi/server.py
index 99fc810..9ce8374 100644
--- a/workers/tgi/server.py
+++ b/workers/tgi/server.py
@@ -127,4 +127,5 @@ routes = [
 ]
 
 if __name__ == "__main__":
+    blips = blorps
     start_server(backend, routes)