Merge branch 'main' into add-hacky-deployments-script

start_server.sh

@@ -2,13 +2,6 @@
 
 set -e -o pipefail
 
-# Check for force update flag
-FORCE_UPDATE=false
-if [ -f "/.force_update" ]; then
-    echo "Force update flag detected at /.force_update"
-    FORCE_UPDATE=true
-fi
-
 WORKSPACE_DIR="${WORKSPACE_DIR:-/workspace}"
 
 SERVER_DIR="$WORKSPACE_DIR/vast-pyworker"
@@ -54,19 +47,13 @@ JSON
 }
 
 function install_vastai_sdk() {
-    local force_flag=""
-    if [ "$FORCE_UPDATE" = true ]; then
-        force_flag="--force-reinstall"
-        echo "Force reinstalling vastai-sdk"
-    fi
-
     # If SDK_BRANCH is set, install vastai-sdk from the vast-sdk repo at that branch/tag/commit.
     if [ -n "${SDK_BRANCH:-}" ]; then
         if [ -n "${SDK_VERSION:-}" ]; then
             echo "WARNING: Both SDK_BRANCH and SDK_VERSION are set; using SDK_BRANCH=${SDK_BRANCH}"
         fi
         echo "Installing vastai-sdk from https://github.com/vast-ai/vast-sdk/ @ ${SDK_BRANCH}"
-        if ! uv pip install $force_flag "vastai-sdk @ git+https://github.com/vast-ai/vast-sdk.git@${SDK_BRANCH}"; then
+        if ! uv pip install "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}"
         fi
         return 0
@@ -74,14 +61,14 @@ function install_vastai_sdk() {
 
     if [ -n "${SDK_VERSION:-}" ]; then
         echo "Installing vastai-sdk version ${SDK_VERSION}"
-        if ! uv pip install $force_flag "vastai-sdk==${SDK_VERSION}"; then
+        if ! uv pip install "vastai-sdk==${SDK_VERSION}"; then
             report_error_and_exit "Failed to install vastai-sdk==${SDK_VERSION}"
         fi
         return 0
     fi
 
     echo "Installing default vastai-sdk"
-    if ! uv pip install $force_flag vastai-sdk; then
+    if ! uv pip install vastai-sdk; then
         report_error_and_exit "Failed to install vastai-sdk"
     fi
 }
@@ -145,27 +132,10 @@ then
         if ! git clone "${PYWORKER_REPO:-https://github.com/vast-ai/pyworker}" "$SERVER_DIR"; then
             report_error_and_exit "Failed to clone pyworker repository"
         fi
-    elif [ "$FORCE_UPDATE" = true ]; then
-        echo "Force updating pyworker repository"
-        if ! (cd "$SERVER_DIR" && git fetch --all); then
-            report_error_and_exit "Failed to fetch pyworker repository updates"
-        fi
     fi
     if [[ -n ${PYWORKER_REF:-} ]]; then
-        if [ "$FORCE_UPDATE" = true ]; then
+        if ! (cd "$SERVER_DIR" && git checkout "$PYWORKER_REF"); then
-            echo "Force updating to pyworker reference: $PYWORKER_REF"
+            report_error_and_exit "Failed to checkout pyworker reference: $PYWORKER_REF"
-            if ! (cd "$SERVER_DIR" && git fetch --all && git checkout "$PYWORKER_REF" && git pull); then
-                report_error_and_exit "Failed to force update pyworker reference: $PYWORKER_REF"
-            fi
-        else
-            if ! (cd "$SERVER_DIR" && git checkout "$PYWORKER_REF"); then
-                report_error_and_exit "Failed to checkout pyworker reference: $PYWORKER_REF"
-            fi
-        fi
-    elif [ "$FORCE_UPDATE" = true ]; then
-        echo "Force updating pyworker to latest"
-        if ! (cd "$SERVER_DIR" && git pull); then
-            report_error_and_exit "Failed to pull latest pyworker changes"
         fi
     fi
 
@@ -197,44 +167,6 @@ else
     fi
     echo "environment activated"
     echo "venv: $VIRTUAL_ENV"
-
-    # Handle force update for existing environment
-    if [ "$FORCE_UPDATE" = true ]; then
-        echo "Performing force update on existing environment"
-
-        # Update pyworker repository
-        if [[ -d $SERVER_DIR ]]; then
-            echo "Force updating pyworker repository"
-            if ! (cd "$SERVER_DIR" && git fetch --all); then
-                report_error_and_exit "Failed to fetch pyworker repository updates"
-            fi
-
-            if [[ -n ${PYWORKER_REF:-} ]]; then
-                echo "Force updating to pyworker reference: $PYWORKER_REF"
-                if ! (cd "$SERVER_DIR" && git checkout "$PYWORKER_REF" && git pull); then
-                    report_error_and_exit "Failed to force update pyworker reference: $PYWORKER_REF"
-                fi
-            else
-                echo "Force updating pyworker to latest"
-                if ! (cd "$SERVER_DIR" && git pull); then
-                    report_error_and_exit "Failed to pull latest pyworker changes"
-                fi
-            fi
-        fi
-
-        # Force reinstall SDK
-        install_vastai_sdk
-    fi
-fi
-
-# Remove force update flag after successful update
-if [ "$FORCE_UPDATE" = true ]; then
-    echo "Removing force update flag"
-    if ! rm -f "/.force_update"; then
-        echo "WARNING: Failed to remove /.force_update, continuing anyway"
-    else
-        echo "Force update completed successfully"
-    fi
 fi
 
 if [ "$USE_SSL" = true ]; then
@@ -282,6 +214,80 @@ fi
 
 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
     report_error_and_exit "Failed to cd into SERVER_DIR: $SERVER_DIR"
 fi