|
|
|
@@ -12,7 +12,7 @@ fi
|
|
|
|
|
WORKSPACE_DIR="${WORKSPACE_DIR:-/workspace}"
|
|
|
|
|
|
|
|
|
|
SERVER_DIR="$WORKSPACE_DIR/vast-pyworker"
|
|
|
|
|
ENV_PATH="${ENV_PATH:-$WORKSPACE_DIR/worker-env}"
|
|
|
|
|
ENV_PATH="$WORKSPACE_DIR/worker-env"
|
|
|
|
|
DEBUG_LOG="$WORKSPACE_DIR/debug.log"
|
|
|
|
|
PYWORKER_LOG="$WORKSPACE_DIR/pyworker.log"
|
|
|
|
|
|
|
|
|
@@ -54,38 +54,35 @@ JSON
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
function install_vastai_sdk() {
|
|
|
|
|
local uv_flags=()
|
|
|
|
|
if [ "${USE_SYSTEM_PYTHON:-}" = "true" ]; then
|
|
|
|
|
uv_flags+=(--system --break-system-packages)
|
|
|
|
|
fi
|
|
|
|
|
local force_flag=""
|
|
|
|
|
if [ "$FORCE_UPDATE" = true ]; then
|
|
|
|
|
uv_flags+=(--force-reinstall)
|
|
|
|
|
echo "Force reinstalling vastai"
|
|
|
|
|
force_flag="--force-reinstall"
|
|
|
|
|
echo "Force reinstalling vastai-sdk"
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
# If SDK_BRANCH is set, install vastai from the vast-cli repo at that branch/tag/commit.
|
|
|
|
|
# If SDK_BRANCH is set, install vastai-sdk from the vast-sdk repo at that branch/tag/commit.
|
|
|
|
|
if [ -n "${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 from https://github.com/vast-ai/vast-cli/ @ ${SDK_BRANCH}"
|
|
|
|
|
if ! uv pip install "${uv_flags[@]}" "vastai @ git+https://github.com/vast-ai/vast-cli.git@${SDK_BRANCH}"; then
|
|
|
|
|
report_error_and_exit "Failed to install vastai from vast-ai/vast-cli@${SDK_BRANCH}"
|
|
|
|
|
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
|
|
|
|
|
report_error_and_exit "Failed to install vastai-sdk from vast-ai/vast-sdk@${SDK_BRANCH}"
|
|
|
|
|
fi
|
|
|
|
|
return 0
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
if [ -n "${SDK_VERSION:-}" ]; then
|
|
|
|
|
echo "Installing vastai version ${SDK_VERSION}"
|
|
|
|
|
if ! uv pip install "${uv_flags[@]}" "vastai==${SDK_VERSION}"; then
|
|
|
|
|
report_error_and_exit "Failed to install vastai==${SDK_VERSION}"
|
|
|
|
|
echo "Installing vastai-sdk version ${SDK_VERSION}"
|
|
|
|
|
if ! uv pip install $force_flag "vastai-sdk==${SDK_VERSION}"; then
|
|
|
|
|
report_error_and_exit "Failed to install vastai-sdk==${SDK_VERSION}"
|
|
|
|
|
fi
|
|
|
|
|
return 0
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
echo "Installing default vastai"
|
|
|
|
|
if ! uv pip install "${uv_flags[@]}" vastai; then
|
|
|
|
|
report_error_and_exit "Failed to install vastai"
|
|
|
|
|
echo "Installing default vastai-sdk"
|
|
|
|
|
if ! uv pip install $force_flag vastai-sdk; then
|
|
|
|
|
report_error_and_exit "Failed to install vastai-sdk"
|
|
|
|
|
fi
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
@@ -128,21 +125,8 @@ if ! grep -q "VAST" /etc/environment; then
|
|
|
|
|
fi
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
if [ "${USE_SYSTEM_PYTHON:-}" = "true" ]; then
|
|
|
|
|
echo "Using system Python: $(which python3)"
|
|
|
|
|
if ! which uv > /dev/null 2>&1; then
|
|
|
|
|
if ! curl -LsSf https://astral.sh/uv/install.sh | sh; then
|
|
|
|
|
report_error_and_exit "Failed to install uv package manager"
|
|
|
|
|
fi
|
|
|
|
|
if [[ -f ~/.local/bin/env ]]; then
|
|
|
|
|
if ! source ~/.local/bin/env; then
|
|
|
|
|
report_error_and_exit "Failed to source uv environment"
|
|
|
|
|
fi
|
|
|
|
|
fi
|
|
|
|
|
fi
|
|
|
|
|
install_vastai_sdk
|
|
|
|
|
touch ~/.no_auto_tmux
|
|
|
|
|
elif [ ! -d "$ENV_PATH" ]; then
|
|
|
|
|
if [ ! -d "$ENV_PATH" ]
|
|
|
|
|
then
|
|
|
|
|
echo "setting up venv"
|
|
|
|
|
if ! which uv; then
|
|
|
|
|
if ! curl -LsSf https://astral.sh/uv/install.sh | sh; then
|
|
|
|
@@ -170,7 +154,7 @@ elif [ ! -d "$ENV_PATH" ]; then
|
|
|
|
|
if [[ -n ${PYWORKER_REF:-} ]]; then
|
|
|
|
|
if [ "$FORCE_UPDATE" = true ]; then
|
|
|
|
|
echo "Force updating to pyworker reference: $PYWORKER_REF"
|
|
|
|
|
if ! (cd "$SERVER_DIR" && git checkout "$PYWORKER_REF" && git pull); then
|
|
|
|
|
if ! (cd "$SERVER_DIR" && git fetch --all && git checkout "$PYWORKER_REF" && git pull); then
|
|
|
|
|
report_error_and_exit "Failed to force update pyworker reference: $PYWORKER_REF"
|
|
|
|
|
fi
|
|
|
|
|
else
|
|
|
|
@@ -208,7 +192,7 @@ else
|
|
|
|
|
report_error_and_exit "Failed to source uv environment"
|
|
|
|
|
fi
|
|
|
|
|
fi
|
|
|
|
|
if ! source "$ENV_PATH/bin/activate"; then
|
|
|
|
|
if ! source "$WORKSPACE_DIR/worker-env/bin/activate"; then
|
|
|
|
|
report_error_and_exit "Failed to activate existing virtual environment"
|
|
|
|
|
fi
|
|
|
|
|
echo "environment activated"
|
|
|
|
@@ -218,6 +202,7 @@ else
|
|
|
|
|
if [ "$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
|
|
|
|
@@ -237,6 +222,7 @@ else
|
|
|
|
|
fi
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
# Force reinstall SDK
|
|
|
|
|
install_vastai_sdk
|
|
|
|
|
fi
|
|
|
|
|
fi
|
|
|
|
@@ -244,8 +230,11 @@ fi
|
|
|
|
|
# Remove force update flag after successful update
|
|
|
|
|
if [ "$FORCE_UPDATE" = true ]; then
|
|
|
|
|
echo "Removing force update flag"
|
|
|
|
|
rm -f "/.force_update"
|
|
|
|
|
echo "Force update completed successfully"
|
|
|
|
|
if ! rm -f "/.force_update"; then
|
|
|
|
|
echo "WARNING: Failed to remove /.force_update, continuing anyway"
|
|
|
|
|
else
|
|
|
|
|
echo "Force update completed successfully"
|
|
|
|
|
fi
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
if [ "$USE_SSL" = true ]; then
|
|
|
|
@@ -283,51 +272,16 @@ EOF
|
|
|
|
|
report_error_and_exit "Failed to generate SSL certificate request"
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
max_retries=5
|
|
|
|
|
retry_delay=2
|
|
|
|
|
for attempt in $(seq 1 "$max_retries"); do
|
|
|
|
|
http_code=$(curl -sS -o /etc/instance.crt -w '%{http_code}' \
|
|
|
|
|
--header 'Content-Type: application/octet-stream' \
|
|
|
|
|
--data-binary @/etc/instance.csr \
|
|
|
|
|
-X POST "https://console.vast.ai/api/v0/sign_cert/?instance_id=$CONTAINER_ID")
|
|
|
|
|
if [ "$http_code" -ge 200 ] && [ "$http_code" -lt 300 ]; then
|
|
|
|
|
break
|
|
|
|
|
fi
|
|
|
|
|
echo "SSL cert signing attempt $attempt/$max_retries failed (HTTP $http_code)"
|
|
|
|
|
if [ "$attempt" -eq "$max_retries" ]; then
|
|
|
|
|
report_error_and_exit "Failed to sign SSL certificate after $max_retries attempts (HTTP $http_code)"
|
|
|
|
|
fi
|
|
|
|
|
sleep "$retry_delay"
|
|
|
|
|
retry_delay=$((retry_delay * 2))
|
|
|
|
|
done
|
|
|
|
|
if ! curl --header 'Content-Type: application/octet-stream' \
|
|
|
|
|
--data-binary @/etc/instance.csr \
|
|
|
|
|
-X \
|
|
|
|
|
POST "https://console.vast.ai/api/v0/sign_cert/?instance_id=$CONTAINER_ID" > /etc/instance.crt; then
|
|
|
|
|
report_error_and_exit "Failed to sign SSL certificate"
|
|
|
|
|
fi
|
|
|
|
|
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.
|
|
|
|
|
|
|
|
|
|
# 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"
|
|
|
|
|
serve-vast-deployment
|
|
|
|
|
exit $?
|
|
|
|
|
fi
|
|
|
|
|
# ─── End SDK Deployment Mode ───────────────────────────────────────────
|
|
|
|
|
|
|
|
|
|
if ! cd "$SERVER_DIR"; then
|
|
|
|
|
report_error_and_exit "Failed to cd into SERVER_DIR: $SERVER_DIR"
|
|
|
|
|
fi
|
|
|
|
@@ -339,19 +293,19 @@ set +e
|
|
|
|
|
PY_STATUS=1
|
|
|
|
|
|
|
|
|
|
if [ -f "$SERVER_DIR/worker.py" ]; then
|
|
|
|
|
echo "Running worker.py"
|
|
|
|
|
echo "trying worker.py"
|
|
|
|
|
python3 -m "worker" |& tee -a "$PYWORKER_LOG"
|
|
|
|
|
PY_STATUS=${PIPESTATUS[0]}
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
if [ "${PY_STATUS}" -ne 0 ] && [ -f "$SERVER_DIR/workers/$BACKEND/worker.py" ]; then
|
|
|
|
|
echo "Running workers.${BACKEND}.worker"
|
|
|
|
|
echo "trying workers.${BACKEND}.worker"
|
|
|
|
|
python3 -m "workers.${BACKEND}.worker" |& tee -a "$PYWORKER_LOG"
|
|
|
|
|
PY_STATUS=${PIPESTATUS[0]}
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
if [ "${PY_STATUS}" -ne 0 ] && [ -f "$SERVER_DIR/workers/$BACKEND/server.py" ]; then
|
|
|
|
|
echo "Running workers.${BACKEND}.server"
|
|
|
|
|
echo "trying workers.${BACKEND}.server"
|
|
|
|
|
python3 -m "workers.${BACKEND}.server" |& tee -a "$PYWORKER_LOG"
|
|
|
|
|
PY_STATUS=${PIPESTATUS[0]}
|
|
|
|
|
fi
|
|
|
|
@@ -365,4 +319,4 @@ if [ "${PY_STATUS}" -ne 0 ]; then
|
|
|
|
|
report_error_and_exit "PyWorker exited with status ${PY_STATUS}"
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
echo "PyWorker bootstrap complete"
|
|
|
|
|
echo "launching PyWorker server done"
|
|
|
|
|
Lucas Armand