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
|
||||||
|
|||||||
+16
-131
@@ -2,17 +2,10 @@
|
|||||||
|
|
||||||
set -e -o pipefail
|
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="${WORKSPACE_DIR:-/workspace}"
|
WORKSPACE_DIR="${WORKSPACE_DIR:-/workspace}"
|
||||||
|
|
||||||
SERVER_DIR="$WORKSPACE_DIR/vast-pyworker"
|
SERVER_DIR="$WORKSPACE_DIR/vast-pyworker"
|
||||||
ENV_PATH="${ENV_PATH:-$WORKSPACE_DIR/worker-env}"
|
ENV_PATH="$WORKSPACE_DIR/worker-env"
|
||||||
DEBUG_LOG="$WORKSPACE_DIR/debug.log"
|
DEBUG_LOG="$WORKSPACE_DIR/debug.log"
|
||||||
PYWORKER_LOG="$WORKSPACE_DIR/pyworker.log"
|
PYWORKER_LOG="$WORKSPACE_DIR/pyworker.log"
|
||||||
|
|
||||||
@@ -54,22 +47,13 @@ JSON
|
|||||||
}
|
}
|
||||||
|
|
||||||
function install_vastai_sdk() {
|
function install_vastai_sdk() {
|
||||||
local uv_flags=()
|
|
||||||
if [ "${USE_SYSTEM_PYTHON:-}" = "true" ]; then
|
|
||||||
uv_flags+=(--system --break-system-packages)
|
|
||||||
fi
|
|
||||||
if [ "$FORCE_UPDATE" = true ]; then
|
|
||||||
uv_flags+=(--force-reinstall)
|
|
||||||
echo "Force reinstalling vastai-sdk"
|
|
||||||
fi
|
|
||||||
|
|
||||||
# If SDK_BRANCH is set, install vastai-sdk from the vast-sdk repo at that branch/tag/commit.
|
# If SDK_BRANCH is set, install vastai-sdk from the vast-sdk repo at that branch/tag/commit.
|
||||||
if [ -n "${SDK_BRANCH:-}" ]; then
|
if [ -n "${SDK_BRANCH:-}" ]; then
|
||||||
if [ -n "${SDK_VERSION:-}" ]; then
|
if [ -n "${SDK_VERSION:-}" ]; then
|
||||||
echo "WARNING: Both SDK_BRANCH and SDK_VERSION are set; using SDK_BRANCH=${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/ @ ${SDK_BRANCH}"
|
||||||
if ! uv pip install "${uv_flags[@]}" "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@${SDK_BRANCH}"
|
report_error_and_exit "Failed to install vastai-sdk from vast-ai/vast-sdk@${SDK_BRANCH}"
|
||||||
fi
|
fi
|
||||||
return 0
|
return 0
|
||||||
@@ -77,14 +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 ${SDK_VERSION}"
|
||||||
if ! uv pip install "${uv_flags[@]}" "vastai-sdk==${SDK_VERSION}"; then
|
if ! uv pip install "vastai-sdk==${SDK_VERSION}"; then
|
||||||
report_error_and_exit "Failed to install vastai-sdk==${SDK_VERSION}"
|
report_error_and_exit "Failed to install vastai-sdk==${SDK_VERSION}"
|
||||||
fi
|
fi
|
||||||
return 0
|
return 0
|
||||||
fi
|
fi
|
||||||
|
|
||||||
echo "Installing default vastai-sdk"
|
echo "Installing default vastai-sdk"
|
||||||
if ! uv pip install "${uv_flags[@]}" vastai-sdk; then
|
if ! uv pip install vastai-sdk; then
|
||||||
report_error_and_exit "Failed to install vastai-sdk"
|
report_error_and_exit "Failed to install vastai-sdk"
|
||||||
fi
|
fi
|
||||||
}
|
}
|
||||||
@@ -106,8 +90,7 @@ echo_var DEBUG_LOG
|
|||||||
echo_var PYWORKER_LOG
|
echo_var PYWORKER_LOG
|
||||||
echo_var MODEL_LOG
|
echo_var MODEL_LOG
|
||||||
|
|
||||||
ROTATE_MODEL_LOG="${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" >> "$MODEL_LOG.old"; then
|
||||||
report_error_and_exit "Failed to rotate model log"
|
report_error_and_exit "Failed to rotate model log"
|
||||||
@@ -128,21 +111,8 @@ if ! grep -q "VAST" /etc/environment; then
|
|||||||
fi
|
fi
|
||||||
fi
|
fi
|
||||||
|
|
||||||
if [ "${USE_SYSTEM_PYTHON:-}" = "true" ]; then
|
if [ ! -d "$ENV_PATH" ]
|
||||||
echo "Using system Python: $(which python3)"
|
then
|
||||||
if ! which uv > /dev/null 2>&1; then
|
|
||||||
if ! curl -LsSf https://astral.sh/uv/install.sh | sh; then
|
|
||||||
report_error_and_exit "Failed to install uv package manager"
|
|
||||||
fi
|
|
||||||
if [[ -f ~/.local/bin/env ]]; then
|
|
||||||
if ! source ~/.local/bin/env; then
|
|
||||||
report_error_and_exit "Failed to source uv environment"
|
|
||||||
fi
|
|
||||||
fi
|
|
||||||
fi
|
|
||||||
install_vastai_sdk
|
|
||||||
touch ~/.no_auto_tmux
|
|
||||||
elif [ ! -d "$ENV_PATH" ]; then
|
|
||||||
echo "setting up venv"
|
echo "setting up venv"
|
||||||
if ! which uv; then
|
if ! which uv; then
|
||||||
if ! curl -LsSf https://astral.sh/uv/install.sh | sh; then
|
if ! curl -LsSf https://astral.sh/uv/install.sh | sh; then
|
||||||
@@ -161,27 +131,10 @@ elif [ ! -d "$ENV_PATH" ]; then
|
|||||||
if ! git clone "${PYWORKER_REPO:-https://github.com/vast-ai/pyworker}" "$SERVER_DIR"; then
|
if ! git clone "${PYWORKER_REPO:-https://github.com/vast-ai/pyworker}" "$SERVER_DIR"; then
|
||||||
report_error_and_exit "Failed to clone pyworker repository"
|
report_error_and_exit "Failed to clone pyworker repository"
|
||||||
fi
|
fi
|
||||||
elif [ "$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
|
fi
|
||||||
if [[ -n ${PYWORKER_REF:-} ]]; then
|
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 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 "$SERVER_DIR" && git pull); then
|
|
||||||
report_error_and_exit "Failed to pull latest pyworker changes"
|
|
||||||
fi
|
fi
|
||||||
fi
|
fi
|
||||||
|
|
||||||
@@ -208,44 +161,11 @@ else
|
|||||||
report_error_and_exit "Failed to source uv environment"
|
report_error_and_exit "Failed to source uv environment"
|
||||||
fi
|
fi
|
||||||
fi
|
fi
|
||||||
if ! source "$ENV_PATH/bin/activate"; then
|
if ! source "$WORKSPACE_DIR/worker-env/bin/activate"; then
|
||||||
report_error_and_exit "Failed to activate existing virtual environment"
|
report_error_and_exit "Failed to activate existing virtual environment"
|
||||||
fi
|
fi
|
||||||
echo "environment activated"
|
echo "environment activated"
|
||||||
echo "venv: $VIRTUAL_ENV"
|
echo "venv: $VIRTUAL_ENV"
|
||||||
|
|
||||||
# Handle force update for existing environment
|
|
||||||
if [ "$FORCE_UPDATE" = true ]; then
|
|
||||||
echo "Performing force update on existing environment"
|
|
||||||
|
|
||||||
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
|
|
||||||
|
|
||||||
install_vastai_sdk
|
|
||||||
fi
|
|
||||||
fi
|
|
||||||
|
|
||||||
# Remove force update flag after successful update
|
|
||||||
if [ "$FORCE_UPDATE" = true ]; then
|
|
||||||
echo "Removing force update flag"
|
|
||||||
rm -f "/.force_update"
|
|
||||||
echo "Force update completed successfully"
|
|
||||||
fi
|
fi
|
||||||
|
|
||||||
if [ "$USE_SSL" = true ]; then
|
if [ "$USE_SSL" = true ]; then
|
||||||
@@ -283,51 +203,16 @@ EOF
|
|||||||
report_error_and_exit "Failed to generate SSL certificate request"
|
report_error_and_exit "Failed to generate SSL certificate request"
|
||||||
fi
|
fi
|
||||||
|
|
||||||
max_retries=5
|
if ! curl --header 'Content-Type: application/octet-stream' \
|
||||||
retry_delay=2
|
--data-binary @/etc/instance.csr \
|
||||||
for attempt in $(seq 1 "$max_retries"); do
|
-X \
|
||||||
http_code=$(curl -sS -o /etc/instance.crt -w '%{http_code}' \
|
POST "https://console.vast.ai/api/v0/sign_cert/?instance_id=$CONTAINER_ID" > /etc/instance.crt; then
|
||||||
--header 'Content-Type: application/octet-stream' \
|
report_error_and_exit "Failed to sign SSL certificate"
|
||||||
--data-binary @/etc/instance.csr \
|
fi
|
||||||
-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=$((retry_delay * 2))
|
|
||||||
done
|
|
||||||
fi
|
fi
|
||||||
|
|
||||||
export REPORT_ADDR WORKER_PORT USE_SSL UNSECURED
|
export REPORT_ADDR WORKER_PORT USE_SSL UNSECURED
|
||||||
|
|
||||||
# ─── SDK Deployment Mode ───────────────────────────────────────────────
|
|
||||||
if [ "$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.
|
|
||||||
|
|
||||||
# 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
|
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