[opensourcedex]

loongcollector is a free, open source monitoring & observability project written in C++ and released under Apache-2.0. It has 2,213 GitHub stars, 456 forks and 166 open issues, and was last pushed 25 hours ago. On this registry it ranks #165 of 270 tracked projects in Monitoring & Observability, with 5 head-to-head comparisons available. It gained 13 stars over the last 30 tracked days.

What is loongcollector?

LoongCollector is an Apache-2.0, C++-based observability data collector from Alibaba that runs as a universal node agent for logs, metrics, traces, events and profiles, and it is built for platform, SRE and security teams operating Kubernetes clusters and high-volume production fleets.

What it is

LoongCollector is the node agent of LoongSuite, Alibaba's unified observability data collection suite. LoongSuite also ships process-level agents for Python, Go and Java, and LoongCollector is the component that runs on the host or in the cluster and gathers the telemetry. It uses a pluggable architecture with more than 100 built-in plugins written in C++ and Go, a SPL engine for data processing, native Kubernetes support, and eBPF-based collection for network monitoring and security events. It is licensed under Apache-2.0, written primarily in C++, and holds 2,196 stars, 450 forks and 166 open issues on GitHub, with the most recent push in September 2026.

The concrete problem it solves is agent sprawl and per-node resource cost. Teams that run FluentBit, Vector or Filebeat for logs usually add separate agents for metrics, traces and network data, then pay for each of them in CPU, memory and operational attention. LoongCollector replaces that set with one agent covering logs, metrics, traces, events and profiles. Alibaba reports 15 years of production use, tens of millions of deployments and hundreds of petabytes collected daily, and the README's benchmark table claims 10x higher throughput at roughly 80% lower resource usage than open-source alternatives.

Key capabilities

  • Collects logs, metrics, traces, events and profiles through a single agent instead of one agent per signal.
  • Gathers network traffic and security events using eBPF, alongside the standard log and metric paths.
  • Provides a pluggable architecture with 100+ built-in plugins, developed in C++ and Go.
  • Includes the SPL engine for flexible in-agent data processing and parsing.
  • Supports remote configuration management through the SLS console, an SDK, or a Kubernetes Operator.
  • Offers self-monitoring, flow control, resource control, alarms and statistics collection.
  • Handles native Kubernetes deployments, so the agent can run as a cluster-level collector.

Who uses it and how

  • Kubernetes platform teams deploy it as a node agent for cluster-wide log, metric and event collection.
  • High-volume log pipelines, where the README's benchmarks put single-line throughput at 546 MB/s against 36 MB/s for FluentBit, 38 MB/s for Vector and 9 MB/s for Filebeat.
  • Security and network visibility teams use the eBPF path to gather network traffic and security events without sidecars or application changes.
  • Polyglot application fleets pair it with the LoongSuite Python, Go and Java process agents for trace and profile instrumentation.
  • Fleets large enough that per-node resource overhead matters, which is the scale the project states it was built for.

Getting started

Start from the Quick Start section of the README, which covers installation and first run, with deployment options that include a Kubernetes Operator and remote configuration from the SLS console. The excerpt provided does not name a package manager, Docker image or compose file.

How it compares

The alternatives named in the README are FluentBit, Vector and Filebeat, and the comparison is a throughput and resource one: the project's own benchmark table places LoongCollector ahead on single-line, multi-line and regex-parsing workloads, with Filebeat listed as not supporting regex parsing in that table. All three alternatives and LoongCollector are self-hosted agents, so the practical difference for an operator is cost per node and whether one agent can cover five signal types. The benchmark numbers are published by the project itself rather than by an independent tester.

When to use it — and when not to

A self-hoster must run and manage the agent fleet, and for the remote management features must stay within the SLS console, SDK or Kubernetes Operator path the project provides. Teams that need only single-signal log shipping, or that run neither Kubernetes nor an Aliyun-adjacent stack, may find a simpler agent sufficient and the plugin surface unnecessary. Two honest caveats: the README excerpt is truncated mid-sentence in its benchmark discussion, and the repository carries 166 open issues, so review the issue tracker before committing.

project readme (upstream, from github) — read inline

LoongCollector - High-Performance Observability Data Collector

LoongCollector Logo

Fast, Lightweight, and Battle-Tested Observability Agent

GitHub contributors GitHub stars GitHub issues GitHub license Coverity Scan Build Status Coverage Status Go Report Card

Quick Start • Documentation • Performance Benchmarks • Contributing

Why LoongCollector?

LoongCollector is a high-performance, lightweight observability data collector designed for modern cloud-native environments. Born from Alibaba's 15-year journey handling exponential traffic growth and powering tens of millions of deployments, LoongCollector delivers 10x higher throughput with 80% lower resource usage compared to open-source alternatives.

LoongCollector is a key component of LoongSuite(Alibaba's unified observability data collection suite). LoongSuite includes the following key components:

  • LoongCollector: universal node agent, which prodivdes log collection, prometheus metric collection, and network and security collection capabilities based on eBPF.
  • LoongSuite Python Agent: a process agent providing instrumentaion for python applications.
  • LoongSuite Go Agent: a process agent for golang with compile time instrumentation.
  • LoongSuite Java Agent: a process agent for Java applications.
  • Other upcoming language agent.

The LoongSuite open-source community hosts monthly online sharing meetings. The meeting access link for each meeting will be posted to GitHub Discussions and shared in the DingTalk and WeChat groups within one week prior to the event. Recordings of past meetings are available on Bilibili.

🚀 Core Advantages

  • ⚡ High Performance

    • 10x higher throughput with 80% less resource usage compared to competitors
    • Linear scaling with predictable performance growth

  • 🛡️ Production-Ready

    • Battle-tested in Alibaba's production environment for 15+ years
    • Powers tens of millions of deployments, collecting hundreds of petabytes daily

  • 🔧 All-in-One Collection

    • Unified agent for logs, metrics, traces, events, and profiles
    • Native Kubernetes support
    • eBPF-powered network monitoring and security event collection

  • 🎯 Pluggable Architecture

    • 100+ built-in plugins with multi-language development (C++, Go)
    • Powerful SPL engine for flexible data processing

  • ⚙️ Advanced Management

    • Remote configuration management via SLS console, SDK, K8s Operator
    • Self-monitoring, flow control, resource control, alarms, and statistics collection.

📊 Performance Benchmarks

Maximum Throughput Comparison

Log Type LoongCollector FluentBit Vector Filebeat
Single Line 546 MB/s 36 MB/s 38 MB/s 9 MB/s
Multi-line 238 MB/s 24 MB/s 22 MB/s 6 MB/s
Regex Parsing 68 MB/s 19 MB/s 12 MB/s Not Supported

📈 Breaking Point Analysis: While competitors hit CPU saturation at ~40 MB/s, LoongCollector maintains linear scaling up to 546 MB/s on a single processing thread.

Resource Efficiency at 10 MB/s Processing Load

Scenario LoongCollector FluentBit Vector Filebeat
Simple Line (512B) 3.40% CPU
29.01 MB RAM		 12.29% CPU (+261%)
46.84 MB RAM (+61%)		 35.80% CPU (+952%)
83.24 MB RAM (+186%)		 Performance Insufficient
Multi-line (512B) 5.82% CPU
29.39 MB RAM		 28.35% CPU (+387%)
46.39 MB RAM (+57%)		 55.99% CPU (+862%)
85.17 MB RAM (+189%)		 Performance Insufficient
Regex (512B) 14.20% CPU
34.02 MB RAM		 37.32% CPU (+162%)
46.44 MB RAM (+36%)		 43.90% CPU (+209%)
90.51 MB RAM (+166%)		 Not Supported

🏗️ Performance And Reliability Architecture Highlights

1. Memory Arena: Zero-Copy Design

  • Shared memory pool (SourceBuffer) stores all string data once per event group
  • String_view references point to original data segments instead of copying

2. Lock-Free Event Pool

  • Thread-aware allocation strategies eliminate lock contention
  • Same-thread pools for direct reuse, double-buffer pools for cross-thread scenarios

3. Zero-Copy Serialization: Direct Network Output

  • Bypasses intermediate Protobuf objects, serializes directly to network format

4. Multi-Tenant Pipeline Isolation

  • High-low watermark feedback queues prevent pipeline interference
  • Independent resource allocation per pipeline with automatic back-pressure control
  • Ensures one pipeline failure doesn't affect others

5. Fair Resource Allocation

  • Priority-aware round-robin scheduling ensures fairness while respecting business priorities
  • Higher priority pipelines always processed first, fair distribution within same priority level
  • Automatic resource yielding when constraints occur

6. Self-Healing Network Resilience

  • Adaptive concurrency limiting per destination using AIMD (Additive Increase, Multiplicative Decrease)
  • Fast failure detection and gradual recovery to prevent network jitter
  • Zero data loss guarantee with intelligent back-pressure control

🏭 Production Validation: Battle-Tested at Scale

LoongCollector has been battle-tested in some of the world's most demanding production environments:

  • Alibaba Group: Powers the entire Alibaba ecosystem including Double 11 shopping festival
  • Alibaba Cloud: Serves tens of thousands of enterprise customers
  • Ant Group: Handles financial transaction observability at massive scale
  • Daily Data Volume: Hundreds of petabytes of observability data
  • Deployment Scale: Tens of millions of active deployments

🚀 Quick Start

Prerequisites

  • Docker (for building from source)
  • Go 1.25.9+ (for building from source)

Build and Run

# Clone the repository
git clone https://github.com/alibaba/loongcollector.git
cd loongcollector
git submodule update --init

# Build LoongCollector
make all
cd output

# Start LoongCollector
nohup ./loongcollector > stdout.log 2> stderr.log &

LoongCollector is now running.

Docker Quick Start

# Build the Docker image alibaba/loongcollector:0.0.1
make dist
make docker

# Run with default configuration
docker run -d --name loongcollector \
  -v /:/logtail_host:ro \
  -v /var/run:/var/run \
  alibaba/loongcollector:0.0.1

📚 Documentation

🤝 Contributing

We welcome contributions from the community! Here are some ways you can help:

📞 Contact Us

Community Channels

Community QR Code

Other LoongSuite components's Community contact information

We are looking forward to your feedback and suggestions. You can scan the QR code below to engage with us.

LoongSuite Python SIG LoongSuite Go SIG LoongSuite Java SIG

📄 License

LoongCollector is licensed under the Apache 2.0 License.

Built with ❤️ by the Alibaba Cloud Observability Team

Empowering developers to build better observability solutions

Frequently asked questions

Is loongcollector free to use?

loongcollector is open source under the Apache-2.0 licence. There is no licence fee and no seat count — you can self-host it or, where the project offers one, pay a vendor for a managed version instead.

What does loongcollector do?

Fast and Lightweight Observability Data Collector

What is loongcollector written in?

loongcollector is primarily written in C++. Its source is publicly available at https://github.com/alibaba/loongcollector, and it has 2,213 GitHub stars.