GlideKV 4.6.9.3

GlideKV with TensorFlow custom operations

GlideKV with TensorFlow Custom Operations

A high-performance GlideKV library featuring custom TensorFlow operations, written in C++ and Python, and seamlessly integrated with Bazel for reproducible builds.

Features

• 🚀 Custom TensorFlow Ops: High-performance C++ ops for advanced GlideKV tasks
• 🛠️ Bazel Build: Reproducible, robust, and scalable build system
• 🔄 Automatic Environment Detection: Automatically finds and links your TensorFlow installation
• 📦 Wheel Packaging: Easily build and distribute as a Python wheel
• 🔧 TensorFlow Serving: Production-ready model serving capabilities

TensorFlow Serving Deployment

Docker Deployment

GlideKV supports deployment via TensorFlow Serving with Docker. The following command starts a TensorFlow Serving container with GlideKV custom operations:

docker run -itd \
  -p 8080:8080 \  # Prometheus metrics port
  -p 8500:8500 \  # gRPC API port
  -p 8501:8501 \  # REST API port
  -e GLIDEKV_METRICS_ENABLED=1 \
  -v /data:/data \
  --privileged=true \
  --network=host \
  registry.cn-beijing.aliyuncs.com/bidops/bidops:GlideKV-v1 \
  --file_system_poll_wait_seconds=10 \
  --model_base_path=/data/model/dnn_winr_v1/export_dir/dense_model/

Configuration Parameters

Parameter	Description	Default
-p 8080:8080	Prometheus metrics port	8080
-p 8500:8500	gRPC API port	8500
-p 8501:8501	REST API port	8501
-e GLIDEKV_METRICS_ENABLED=1	Enable GlideKV metrics collection	0
-v /data:/data	Mount local data directory	-
--privileged=true	Enable privileged mode	false
--network=host	Use host network mode	bridge
--file_system_poll_wait_seconds=10	Model polling interval	10
--model_base_path	Path to the TensorFlow model	-

Environment Variables

• GLIDEKV_METRICS_ENABLED=1: Enables Prometheus metrics collection for monitoring

Model Requirements

The TensorFlow model must include GlideKV custom operations:

• HashTableOfTensors: For vector lookup operations
• LookupFind: For key-value retrieval

Health Check

After deployment, you can check the service status:

# Check HTTP REST API
curl http://localhost:8501/v1/models

# Check gRPC API (requires grpcurl installation)
grpcurl -plaintext localhost:8500 list

# Check Prometheus metrics
curl http://localhost:8080/metrics

Install grpcurl (Optional)

If you need to test gRPC API, install grpcurl:

# Ubuntu/Debian
sudo apt-get update && sudo apt-get install -y grpcurl

# CentOS/RHEL
sudo yum install -y grpcurl

# Or download from GitHub releases
wget https://github.com/fullstorydev/grpcurl/releases/download/v1.8.7/grpcurl_1.8.7_linux_x86_64.tar.gz
tar -xzf grpcurl_1.8.7_linux_x86_64.tar.gz
sudo mv grpcurl /usr/local/bin/
rm grpcurl_1.8.7_linux_x86_64.tar.gz

开发与构建说明

• Bazel 构建：所有 C++/TF 自定义算子通过 Bazel 构建，确保与 Python 环境 TensorFlow 版本一致。
• 环境变量：需设置 TF_SO_PATH 指向当前 Python 环境下的 TensorFlow 路径（python configure.py 可自动完成）。
• .gitignore 已优化，避免无关文件入库。

贡献指南

欢迎提交 Issue 和 PR！请确保：

• 遵循 PEP8 代码风格
• 新增功能需附带测试
• 详细描述你的更改

License

MIT