Compare commits
16 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| 4e951f4912 | |||
| f636012685 | |||
| ddb986d561 | |||
| 99a3319e66 | |||
| 83c31e25a9 | |||
| fbe1dca6fa | |||
| 4c3120dbc5 | |||
| d7d9b915f6 | |||
| 4660b337fb | |||
| 7506ecb6b5 | |||
| 50633c5003 | |||
| 2e8f18276f | |||
| eba9c480eb | |||
| aaca1c9645 | |||
| f319db6bd5 | |||
| 4d786b4d17 |
@@ -9,7 +9,7 @@ This repository contains **example PyWorkers** used by Vast.ai’s default Serve
|
|||||||
- Optionally supports FIFO queueing when the backend cannot process concurrent requests
|
- Optionally supports FIFO queueing when the backend cannot process concurrent requests
|
||||||
- Detects readiness/failure from model logs and runs a benchmark to estimate throughput
|
- Detects readiness/failure from model logs and runs a benchmark to estimate throughput
|
||||||
|
|
||||||
> Important: The **core PyWorker framework** (Worker, WorkerConfig, HandlerConfig, BenchmarkConfig, LogActionConfig) is provided by the **`vastai` / `vastai-sdk`** Python package (https://github.com/vast-ai/vast-sdk). This repo focuses on *worker implementations and examples*, not the framework internals.
|
> Important: The **core PyWorker framework** (Worker, WorkerConfig, HandlerConfig, BenchmarkConfig, LogActionConfig) is provided by the **`vastai`** Python package (https://github.com/vast-ai/vast-cli). This repo focuses on *worker implementations and examples*, not the framework internals.
|
||||||
|
|
||||||
## Repository Purpose
|
## Repository Purpose
|
||||||
|
|
||||||
|
|||||||
@@ -1,16 +1,16 @@
|
|||||||
# Where did the PyWorker code go?
|
# Where did the PyWorker code go?
|
||||||
We have moved the PyWorker source code into the `vastai-sdk` Python SDK.
|
We have moved the PyWorker source code into the `vastai` Python package.
|
||||||
You can install it with
|
You can install it with
|
||||||
```
|
```
|
||||||
pip install vastai-sdk
|
pip install vastai
|
||||||
```
|
```
|
||||||
|
|
||||||
All of the source code can be found here:
|
All of the source code can be found here:
|
||||||
https://github.com/vast-ai/vast-sdk
|
https://github.com/vast-ai/vast-cli
|
||||||
|
|
||||||
And can be imported from vastai.serverless.server.lib
|
And can be imported from vastai.serverless.server.lib
|
||||||
|
|
||||||
Serverless instances automatically run the start_server.sh script, which installs the vastai-sdk.
|
Serverless instances automatically run the start_server.sh script, which installs the vastai package.
|
||||||
This is how the PyWorker source code makes it onto your serverless instances.
|
This is how the PyWorker source code makes it onto your serverless instances.
|
||||||
You provide a worker.py file in your PYWORKER_REPO, and the start_server.sh will
|
You provide a worker.py file in your PYWORKER_REPO, and the start_server.sh will
|
||||||
create and run a PyWorker according to your configuration defined in the file.
|
create and run a PyWorker according to your configuration defined in the file.
|
||||||
|
|||||||
+1
-13
@@ -1,13 +1 @@
|
|||||||
aiohttp==3.10.1
|
vastai>=0.3.0
|
||||||
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
|
|
||||||
|
|||||||
+151
-24
@@ -2,10 +2,17 @@
|
|||||||
|
|
||||||
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="$WORKSPACE_DIR/worker-env"
|
ENV_PATH="${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="$WORKSPACE_DIR/pyworker.log"
|
||||||
|
|
||||||
@@ -46,21 +53,42 @@ JSON
|
|||||||
exit 1
|
exit 1
|
||||||
}
|
}
|
||||||
|
|
||||||
function install_vastai_sdk() {
|
function install_vastai() {
|
||||||
|
local uv_flags=()
|
||||||
|
if [ "${USE_SYSTEM_PYTHON:-}" = "true" ]; then
|
||||||
|
uv_flags+=(--system --break-system-packages)
|
||||||
|
fi
|
||||||
|
if [ "$FORCE_UPDATE" = true ]; then
|
||||||
|
uv_flags+=(--force-reinstall)
|
||||||
|
echo "Force reinstalling vastai"
|
||||||
|
fi
|
||||||
|
|
||||||
|
# If SDK_BRANCH is set, install vastai from the vast-cli repo at that branch/tag/commit.
|
||||||
|
if [ -n "${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}"
|
||||||
|
fi
|
||||||
|
return 0
|
||||||
|
fi
|
||||||
|
|
||||||
if [ -n "${SDK_VERSION:-}" ]; then
|
if [ -n "${SDK_VERSION:-}" ]; then
|
||||||
echo "Installing vastai-sdk version ${SDK_VERSION}"
|
echo "Installing vastai version ${SDK_VERSION}"
|
||||||
if ! uv pip install "vastai-sdk==${SDK_VERSION}"; then
|
if ! uv pip install --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple/ "${uv_flags[@]}" "vastai-sdk-vast==${SDK_VERSION}"; then
|
||||||
report_error_and_exit "Failed to install vastai-sdk==${SDK_VERSION}"
|
report_error_and_exit "Failed to install vastai-vast==${SDK_VERSION}"
|
||||||
fi
|
|
||||||
else
|
|
||||||
echo "Installing default vastai-sdk"
|
|
||||||
if ! uv pip install vastai-sdk; then
|
|
||||||
report_error_and_exit "Failed to install vastai-sdk"
|
|
||||||
fi
|
fi
|
||||||
|
return 0
|
||||||
|
fi
|
||||||
|
|
||||||
|
echo "Installing default vastai"
|
||||||
|
if ! uv pip install "${uv_flags[@]}" vastai; then
|
||||||
|
report_error_and_exit "Failed to install vastai"
|
||||||
fi
|
fi
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
[ -n "$BACKEND" ] && [ -z "$HF_TOKEN" ] && report_error_and_exit "HF_TOKEN must be set when BACKEND is set!"
|
[ -n "$BACKEND" ] && [ -z "$HF_TOKEN" ] && report_error_and_exit "HF_TOKEN must be set when BACKEND is set!"
|
||||||
[ -z "$CONTAINER_ID" ] && report_error_and_exit "CONTAINER_ID must be set!"
|
[ -z "$CONTAINER_ID" ] && report_error_and_exit "CONTAINER_ID must be set!"
|
||||||
[ "$BACKEND" = "comfyui" ] && [ -z "$COMFY_MODEL" ] && report_error_and_exit "For comfyui backends, COMFY_MODEL must be set!"
|
[ "$BACKEND" = "comfyui" ] && [ -z "$COMFY_MODEL" ] && report_error_and_exit "For comfyui backends, COMFY_MODEL must be set!"
|
||||||
@@ -78,7 +106,8 @@ echo_var DEBUG_LOG
|
|||||||
echo_var PYWORKER_LOG
|
echo_var PYWORKER_LOG
|
||||||
echo_var MODEL_LOG
|
echo_var MODEL_LOG
|
||||||
|
|
||||||
if [ -e "$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" >> "$MODEL_LOG.old"; then
|
||||||
report_error_and_exit "Failed to rotate model log"
|
report_error_and_exit "Failed to rotate model log"
|
||||||
@@ -99,8 +128,21 @@ if ! grep -q "VAST" /etc/environment; then
|
|||||||
fi
|
fi
|
||||||
fi
|
fi
|
||||||
|
|
||||||
if [ ! -d "$ENV_PATH" ]
|
if [ "${USE_SYSTEM_PYTHON:-}" = "true" ]; then
|
||||||
then
|
echo "Using system Python: $(which python3)"
|
||||||
|
if ! which uv > /dev/null 2>&1; then
|
||||||
|
if ! curl -LsSf https://astral.sh/uv/install.sh | sh; then
|
||||||
|
report_error_and_exit "Failed to install uv package manager"
|
||||||
|
fi
|
||||||
|
if [[ -f ~/.local/bin/env ]]; then
|
||||||
|
if ! source ~/.local/bin/env; then
|
||||||
|
report_error_and_exit "Failed to source uv environment"
|
||||||
|
fi
|
||||||
|
fi
|
||||||
|
fi
|
||||||
|
install_vastai
|
||||||
|
touch ~/.no_auto_tmux
|
||||||
|
elif [ ! -d "$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
|
||||||
@@ -119,10 +161,27 @@ 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 ${PYWORKER_REF:-} ]]; then
|
||||||
if ! (cd "$SERVER_DIR" && git checkout "$PYWORKER_REF"); then
|
if [ "$FORCE_UPDATE" = true ]; then
|
||||||
report_error_and_exit "Failed to checkout pyworker reference: $PYWORKER_REF"
|
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: $PYWORKER_REF"
|
||||||
|
fi
|
||||||
|
else
|
||||||
|
if ! (cd "$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 "$SERVER_DIR" && git pull); then
|
||||||
|
report_error_and_exit "Failed to pull latest pyworker changes"
|
||||||
fi
|
fi
|
||||||
fi
|
fi
|
||||||
|
|
||||||
@@ -138,7 +197,7 @@ then
|
|||||||
report_error_and_exit "Failed to install Python requirements"
|
report_error_and_exit "Failed to install Python requirements"
|
||||||
fi
|
fi
|
||||||
|
|
||||||
install_vastai_sdk
|
install_vastai
|
||||||
|
|
||||||
if ! touch ~/.no_auto_tmux; then
|
if ! touch ~/.no_auto_tmux; then
|
||||||
report_error_and_exit "Failed to create ~/.no_auto_tmux"
|
report_error_and_exit "Failed to create ~/.no_auto_tmux"
|
||||||
@@ -149,11 +208,44 @@ else
|
|||||||
report_error_and_exit "Failed to source uv environment"
|
report_error_and_exit "Failed to source uv environment"
|
||||||
fi
|
fi
|
||||||
fi
|
fi
|
||||||
if ! source "$WORKSPACE_DIR/worker-env/bin/activate"; then
|
if ! source "$ENV_PATH/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
|
||||||
|
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
|
||||||
@@ -191,16 +283,51 @@ EOF
|
|||||||
report_error_and_exit "Failed to generate SSL certificate request"
|
report_error_and_exit "Failed to generate SSL certificate request"
|
||||||
fi
|
fi
|
||||||
|
|
||||||
if ! curl --header 'Content-Type: application/octet-stream' \
|
max_retries=5
|
||||||
--data-binary @/etc/instance.csr \
|
retry_delay=2
|
||||||
-X \
|
for attempt in $(seq 1 "$max_retries"); do
|
||||||
POST "https://console.vast.ai/api/v0/sign_cert/?instance_id=$CONTAINER_ID" > /etc/instance.crt; then
|
http_code=$(curl -sS -o /etc/instance.crt -w '%{http_code}' \
|
||||||
report_error_and_exit "Failed to sign SSL certificate"
|
--header 'Content-Type: application/octet-stream' \
|
||||||
fi
|
--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=$((retry_delay * 2))
|
||||||
|
done
|
||||||
fi
|
fi
|
||||||
|
|
||||||
export REPORT_ADDR WORKER_PORT USE_SSL UNSECURED
|
export REPORT_ADDR WORKER_PORT USE_SSL UNSECURED
|
||||||
|
|
||||||
|
# ─── SDK Deployment Mode ───────────────────────────────────────────────
|
||||||
|
if [ "$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 function which supports SDK_BRANCH/SDK_VERSION)
|
||||||
|
install_vastai
|
||||||
|
# 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