Compare commits
7 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| 5330039cbf | |||
| fe999dfd16 | |||
| 85707af107 | |||
| 82023f1cfb | |||
| 5f9580dde2 | |||
| 0b02f31aa8 | |||
| 3e8da87ce2 |
@@ -1 +1,13 @@
|
|||||||
|
aiohttp==3.10.1
|
||||||
|
aiodns~=3.6.0
|
||||||
|
pycares~=4.11.0
|
||||||
|
anyio~=4.4
|
||||||
|
lib~=4.0
|
||||||
|
nltk~=3.9
|
||||||
|
psutil~=6.0
|
||||||
|
pycryptodome~=3.20
|
||||||
|
Requests~=2.32
|
||||||
|
transformers~=4.52
|
||||||
|
utils==1.0.*
|
||||||
|
hf_transfer>=0.1.9
|
||||||
vastai-sdk>=0.3.0
|
vastai-sdk>=0.3.0
|
||||||
|
|||||||
+1
-76
@@ -90,8 +90,7 @@ echo_var DEBUG_LOG
|
|||||||
echo_var PYWORKER_LOG
|
echo_var PYWORKER_LOG
|
||||||
echo_var MODEL_LOG
|
echo_var MODEL_LOG
|
||||||
|
|
||||||
ROTATE_MODEL_LOG="${ROTATE_MODEL_LOG:-false}"
|
if [ -e "$MODEL_LOG" ]; then
|
||||||
if [ "$ROTATE_MODEL_LOG" = "true" ] && [ -e "$MODEL_LOG" ]; then
|
|
||||||
echo "Rotating model log at $MODEL_LOG to $MODEL_LOG.old"
|
echo "Rotating model log at $MODEL_LOG to $MODEL_LOG.old"
|
||||||
if ! cat "$MODEL_LOG" >> "$MODEL_LOG.old"; then
|
if ! cat "$MODEL_LOG" >> "$MODEL_LOG.old"; then
|
||||||
report_error_and_exit "Failed to rotate model log"
|
report_error_and_exit "Failed to rotate model log"
|
||||||
@@ -214,80 +213,6 @@ fi
|
|||||||
|
|
||||||
export REPORT_ADDR WORKER_PORT USE_SSL UNSECURED
|
export REPORT_ADDR WORKER_PORT USE_SSL UNSECURED
|
||||||
|
|
||||||
# ─── SDK Deployment Mode ───────────────────────────────────────────────
|
|
||||||
if [ "$IS_DEPLOYMENT" = "true" ]; then
|
|
||||||
echo "=== SDK Deployment Mode ==="
|
|
||||||
echo "DEPLOYMENT_ID: $DEPLOYMENT_ID"
|
|
||||||
|
|
||||||
DEPLOY_DIR="/workspace/deployment"
|
|
||||||
mkdir -p "$DEPLOY_DIR"
|
|
||||||
|
|
||||||
VAST_API_BASE="${VAST_API_BASE:-https://console.vast.ai}"
|
|
||||||
|
|
||||||
# Download deployment code, retrying until the blob is available on S3.
|
|
||||||
# The s3_key exists in the DB as soon as the deployment is created, but the
|
|
||||||
# actual upload may still be in flight from the client side.
|
|
||||||
echo "Downloading deployment code..."
|
|
||||||
RETRY=0
|
|
||||||
while true; do
|
|
||||||
DOWNLOAD_RESPONSE=$(curl -sS \
|
|
||||||
-H "Authorization: Bearer $CONTAINER_API_KEY" \
|
|
||||||
"${VAST_API_BASE}/api/v0/deployment/${DEPLOYMENT_ID}/download_url/")
|
|
||||||
DOWNLOAD_URL=$(python3 -c "
|
|
||||||
import sys, json
|
|
||||||
try:
|
|
||||||
d = json.load(sys.stdin)
|
|
||||||
print(d.get('download_url') or '')
|
|
||||||
except: print('')
|
|
||||||
" <<< "$DOWNLOAD_RESPONSE")
|
|
||||||
|
|
||||||
if [ -z "$DOWNLOAD_URL" ] || [ "$DOWNLOAD_URL" = "None" ]; then
|
|
||||||
RETRY=$((RETRY + 1))
|
|
||||||
echo "No download URL yet (attempt $RETRY), retrying in 10s... response: $DOWNLOAD_RESPONSE"
|
|
||||||
sleep 10
|
|
||||||
continue
|
|
||||||
fi
|
|
||||||
|
|
||||||
# Got a URL — try the actual S3 download
|
|
||||||
HTTP_CODE=$(curl -sS -L -o "$DEPLOY_DIR/deployment.tar.gz" -w "%{http_code}" "$DOWNLOAD_URL")
|
|
||||||
if [ "$HTTP_CODE" = "200" ]; then
|
|
||||||
break
|
|
||||||
fi
|
|
||||||
|
|
||||||
RETRY=$((RETRY + 1))
|
|
||||||
echo "S3 download returned HTTP $HTTP_CODE (attempt $RETRY), blob not yet uploaded. Retrying in 10s..."
|
|
||||||
rm -f "$DEPLOY_DIR/deployment.tar.gz"
|
|
||||||
sleep 10
|
|
||||||
done
|
|
||||||
|
|
||||||
cd "$DEPLOY_DIR" && tar xzf deployment.tar.gz
|
|
||||||
echo "Deployment code extracted."
|
|
||||||
|
|
||||||
# Source secrets if present
|
|
||||||
if [ -f "$DEPLOY_DIR/.secrets" ]; then
|
|
||||||
echo "Sourcing secrets..."
|
|
||||||
source "$DEPLOY_DIR/.secrets"
|
|
||||||
fi
|
|
||||||
|
|
||||||
# Run on_start.sh to completion if present
|
|
||||||
if [ -f "$DEPLOY_DIR/on_start.sh" ]; then
|
|
||||||
echo "Running on_start.sh..."
|
|
||||||
chmod +x "$DEPLOY_DIR/on_start.sh"
|
|
||||||
bash "$DEPLOY_DIR/on_start.sh"
|
|
||||||
echo "on_start.sh completed."
|
|
||||||
fi
|
|
||||||
|
|
||||||
# Install SDK (uses the install_vastai_sdk function which supports SDK_BRANCH/SDK_VERSION)
|
|
||||||
install_vastai_sdk
|
|
||||||
|
|
||||||
# Run deployment in serve mode
|
|
||||||
export VAST_DEPLOYMENT_MODE=serve
|
|
||||||
echo "Starting deployment: python3 $DEPLOY_DIR/deployment.py"
|
|
||||||
python3 "$DEPLOY_DIR/deployment.py"
|
|
||||||
exit $?
|
|
||||||
fi
|
|
||||||
# ─── End SDK Deployment Mode ───────────────────────────────────────────
|
|
||||||
|
|
||||||
if ! cd "$SERVER_DIR"; then
|
if ! cd "$SERVER_DIR"; then
|
||||||
report_error_and_exit "Failed to cd into SERVER_DIR: $SERVER_DIR"
|
report_error_and_exit "Failed to cd into SERVER_DIR: $SERVER_DIR"
|
||||||
fi
|
fi
|
||||||
|
|||||||
Reference in New Issue
Block a user