Compare commits

..

2 Commits

Author SHA1 Message Date
Abiola Akinnubi 74efc2cb42 bump up version minor number 2025-11-14 18:07:17 -08:00
Abiola Akinnubi db3096bbaf feat AUTO-695: add loaded_at attribute to AutoScalerData and Metrics classes 2025-11-14 17:07:06 -08:00
3 changed files with 32 additions and 94 deletions
+2 -2
View File
@@ -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):
@@ -216,7 +217,7 @@ class Metrics:
id=self.id,
mtoken=self.mtoken,
version=self.version,
loadtime=(loadtime_snapshot or 0.0),
loadtime=(loadtime_snapshot or 0.0),
new_load=self.model_metrics.workload_processing,
cur_load=self.model_metrics.cur_load,
rej_load=self.model_metrics.workload_rejected,
@@ -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()
+25 -45
View File
@@ -3,58 +3,38 @@ 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:
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
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())
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())
run(main())
+5 -47
View File
@@ -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 "$SERVER_DIR"
echo "launching PyWorker server"
set +e
python3 -m "workers.$BACKEND.server" |& tee -a "$PYWORKER_LOG"
PY_STATUS=${PIPESTATUS[0]}
set -e
# 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 $MODEL_LOG.old for debugging only
[ -e "$MODEL_LOG" ] && cat "$MODEL_LOG" >> "$MODEL_LOG.old" && : > "$MODEL_LOG"
if [ "${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"
(python3 -m "workers.$BACKEND.server" |& tee -a "$PYWORKER_LOG") &
echo "launching PyWorker server done"