Compare commits
6 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| a47c9d1ed0 | |||
| 0b14562a63 | |||
| de9b50abb9 | |||
| c510801723 | |||
| a12523b1d2 | |||
| 7db54f3bd7 |
+21
-1
@@ -3,15 +3,17 @@ import logging
|
|||||||
from typing import List
|
from typing import List
|
||||||
import ssl
|
import ssl
|
||||||
from asyncio import run, gather
|
from asyncio import run, gather
|
||||||
|
import asyncio
|
||||||
|
|
||||||
from lib.backend import Backend
|
from lib.backend import Backend
|
||||||
|
from lib.metrics import Metrics
|
||||||
from aiohttp import web
|
from aiohttp import web
|
||||||
|
|
||||||
log = logging.getLogger(__file__)
|
log = logging.getLogger(__file__)
|
||||||
|
|
||||||
|
|
||||||
def start_server(backend: Backend, routes: List[web.RouteDef], **kwargs):
|
def start_server(backend: Backend, routes: List[web.RouteDef], **kwargs):
|
||||||
|
try:
|
||||||
log.debug("getting certificate...")
|
log.debug("getting certificate...")
|
||||||
use_ssl = os.environ.get("USE_SSL", "false") == "true"
|
use_ssl = os.environ.get("USE_SSL", "false") == "true"
|
||||||
if use_ssl is True:
|
if use_ssl is True:
|
||||||
@@ -38,3 +40,21 @@ def start_server(backend: Backend, routes: List[web.RouteDef], **kwargs):
|
|||||||
await gather(site.start(), backend._start_tracking())
|
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())
|
||||||
|
|||||||
+40
-1
@@ -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
|
# 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" && : > "$MODEL_LOG"
|
[ -e "$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"
|
echo "launching PyWorker server done"
|
||||||
Reference in New Issue
Block a user