Compare commits

..

5 Commits

Author SHA1 Message Date
Lucas Armand 62cd96ea68 Allow pre-release 2026-03-24 13:09:04 -07:00
LucasArmandVast 50633c5003 Update deployments script with retries. (#81) 2026-03-23 14:58:32 -07:00
LucasArmandVast 2e8f18276f Add beta deployments script (#80) 2026-03-23 14:14:06 -07:00
Scott Darden eba9c480eb Merge pull request #79 from vast-ai/update-requirements
Updated requirements to only require vastai-sdk
2026-01-14 12:07:33 -08:00
LucasArmandVast f319db6bd5 flag for model log rotate (#78) 2026-01-12 17:03:18 -08:00
+79 -4
View File
@@ -53,7 +53,7 @@ function install_vastai_sdk() {
echo "WARNING: Both SDK_BRANCH and SDK_VERSION are set; using SDK_BRANCH=${SDK_BRANCH}" echo "WARNING: Both SDK_BRANCH and SDK_VERSION are set; using SDK_BRANCH=${SDK_BRANCH}"
fi fi
echo "Installing vastai-sdk from https://github.com/vast-ai/vast-sdk/ @ ${SDK_BRANCH}" echo "Installing vastai-sdk from https://github.com/vast-ai/vast-sdk/ @ ${SDK_BRANCH}"
if ! uv pip install "vastai-sdk @ git+https://github.com/vast-ai/vast-sdk.git@${SDK_BRANCH}"; then if ! uv pip install --prerelease=allow "vastai-sdk @ git+https://github.com/vast-ai/vast-sdk.git@${SDK_BRANCH}"; then
report_error_and_exit "Failed to install vastai-sdk from vast-ai/vast-sdk@${SDK_BRANCH}" report_error_and_exit "Failed to install vastai-sdk from vast-ai/vast-sdk@${SDK_BRANCH}"
fi fi
return 0 return 0
@@ -61,14 +61,14 @@ function install_vastai_sdk() {
if [ -n "${SDK_VERSION:-}" ]; then if [ -n "${SDK_VERSION:-}" ]; then
echo "Installing vastai-sdk version ${SDK_VERSION}" echo "Installing vastai-sdk version ${SDK_VERSION}"
if ! uv pip install "vastai-sdk==${SDK_VERSION}"; then if ! uv pip install --prerelease=allow "vastai-sdk==${SDK_VERSION}"; then
report_error_and_exit "Failed to install vastai-sdk==${SDK_VERSION}" report_error_and_exit "Failed to install vastai-sdk==${SDK_VERSION}"
fi fi
return 0 return 0
fi fi
echo "Installing default vastai-sdk" echo "Installing default vastai-sdk"
if ! uv pip install vastai-sdk; then if ! uv pip install --prerelease=allow vastai-sdk; then
report_error_and_exit "Failed to install vastai-sdk" report_error_and_exit "Failed to install vastai-sdk"
fi fi
} }
@@ -90,7 +90,8 @@ echo_var DEBUG_LOG
echo_var PYWORKER_LOG echo_var PYWORKER_LOG
echo_var MODEL_LOG echo_var MODEL_LOG
if [ -e "$MODEL_LOG" ]; then ROTATE_MODEL_LOG="${ROTATE_MODEL_LOG:-false}"
if [ "$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" >> "$MODEL_LOG.old"; then
report_error_and_exit "Failed to rotate model log" report_error_and_exit "Failed to rotate model log"
@@ -213,6 +214,80 @@ fi
export REPORT_ADDR WORKER_PORT USE_SSL UNSECURED export REPORT_ADDR WORKER_PORT USE_SSL UNSECURED
# ─── SDK Deployment Mode ───────────────────────────────────────────────
if [ "$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