DynamicTp: Real‑Time, Zero‑Intrusion ThreadPoolExecutor Tuning for Java Services

Background

In Java, ThreadPoolExecutor is the primary tool for improving system throughput, but determining optimal core size, maximum size, and queue capacity is often trial‑and‑error. Static configuration requires code changes and service restarts, and the runtime state of a thread pool is usually invisible until a problem occurs.

Modern microservice architectures rely on distributed configuration centers (e.g., Nacos, Apollo, Zookeeper, Consul, Etcd) to push configuration changes without downtime. By integrating a thread‑pool library with such a center, pool parameters can be adjusted dynamically at runtime.

DynamicTp Overview

DynamicTp is a SpringBoot‑based framework that extends ThreadPoolExecutor to enable real‑time modification of thread‑pool parameters, continuous monitoring, and alerting. The latest release (v1.0.9) provides:

Zero Intrusion : All configuration resides in external configuration centers; no business‑code changes are required.

Lightweight Integration : A SpringBoot starter ( starter) allows full setup in four steps.

High Extensibility : Core functions expose SPI interfaces for custom configuration sources, notification channels, metric collectors, and task wrappers.

Monitoring & Alerting : Metrics can be exported as JSON logs, via Micrometer for Prometheus/Grafana, or through a custom HTTP endpoint ( dynamic-tp). Alerts support DingTalk, WeChat Work, Feishu, email, and custom SPI extensions.

Task Wrappers : Implementations such as MdcTaskWrapper, TtlTaskWrapper, and SwTraceTaskWrapper propagate context across thread‑pool tasks.

Compatibility : Works with standard JUC thread pools and Spring’s ThreadPoolTaskExecutor via the @DynamicTp annotation.

Middleware Support : Manages thread pools of Tomcat, Jetty, Undertow, Dubbo, RocketMQ, Hystrix, and gRPC.

Multi‑Center Support : Native adapters for Nacos, Apollo, Zookeeper, Consul, and Etcd; additional providers can be added via SPI.

Architecture Design

The framework is organized into five primary modules:

Configuration Change Listener : Listens to configuration files from supported centers, parses YAML, properties, or JSON, and notifies the pool manager of updates.

Internal Thread‑Pool Management : Creates ThreadPoolExecutor instances at startup, registers them in an internal registry and the Spring context, and refreshes parameters on configuration changes. Pools can be obtained via Spring injection or DtpRegistry.getDtpExecutor(name).

Third‑Party Component Management : Detects and wraps thread pools of external middleware (Tomcat, Jetty, Undertow, Dubbo, RocketMQ, Hystrix, gRPC) to provide unified monitoring and dynamic adjustment.

Monitoring Module : Collects pool metrics and exports them through three default exporters: JSON log files, Micrometer (for Prometheus/Grafana), and a custom HTTP endpoint ( dynamic-tp).

Notification & Alarm Module : Sends alerts to enterprise messaging platforms and supports alarm types such as parameter changes, queue‑capacity thresholds, pool‑activity thresholds, rejection‑policy triggers, and task execution/queue timeouts.

Key Module Details

Configuration Change Listener

Monitors specific configuration files from Nacos, Apollo, Zookeeper, Consul, Etcd, etc., using SPI for extensibility.

Parses YAML, properties, or JSON formats; additional parsers can be added via SPI.

Triggers a refresh event to the internal thread‑pool manager.

Internal Thread‑Pool Management

At service startup, pulls configuration from the center, creates ThreadPoolExecutor instances, and registers them.

Handles refresh events from the listener to update pool parameters without restarting the service.

Provides access via Spring injection or DtpRegistry.getDtpExecutor("poolName").

Third‑Party Component Management

Detects and wraps thread pools of supported middleware, enabling unified monitoring, dynamic parameter refresh, metric collection, and alarm handling.

Monitoring Module

Exports metrics through:

JSON log output for custom parsing.

Micrometer integration for Prometheus/Grafana dashboards.

Custom HTTP endpoint ( dynamic-tp) for real‑time metric access.

Notification & Alarm Module

Supports alert channels such as DingTalk, WeChat Work, Feishu, and email. Alarm types include:

Parameter changes.

Queue‑capacity threshold breaches.

Pool‑activity threshold breaches.

Rejection‑policy triggers.

Task execution timeout.

Task waiting timeout.

Open‑Source Repository

Source code and documentation are available at:

https://github.com/dromara/dynamic-tp