AnoMod: A Dataset for Anomaly Detection and Root Cause Analysis in Microservice Systems
The paper introduces AnoMod, a comprehensive multimodal dataset for microservice systems that addresses existing gaps by providing diverse anomaly categories and five distinct monitoring modalities to enable advanced research in cross-modal anomaly detection and fine-grained root cause analysis.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are the manager of a massive, bustling train station (a Microservice System). Instead of one big building, the station is made of hundreds of tiny, specialized booths: one sells tickets, another checks IDs, a third handles luggage, and a fourth manages the train schedules. They all talk to each other constantly to get a passenger from point A to point B.
When something goes wrong—a ticket booth freezes, the luggage scanner lags, or the train schedule gets corrupted—it's a nightmare to figure out why. Is it the network? A bad piece of code? A full database?
This is the problem the paper AnoMod tries to solve. Here is the breakdown in simple terms:
1. The Problem: The "Blind Spot" in Current Tools
The authors say that right now, researchers trying to build "smart detectives" (AI tools) to fix these train stations are working with bad maps.
- The Old Maps: Existing datasets (collections of data) mostly only look at performance. They tell you if the station is "slow" or "busy," like checking if the line at the ticket booth is long.
- The Missing Pieces: They don't tell you why the line is long. Is it because the ticket machine broke? Did the cashier make a math error? Did the database run out of ink? Also, most old datasets only looked at two or three types of clues (like logs and speed charts), missing the bigger picture.
2. The Solution: A New "Super-Dataset" Called AnoMod
The team created a new, massive dataset called AnoMod. Think of this as a "crash test" where they intentionally broke two real-world train stations (SocialNetwork and TrainTicket) in very specific, realistic ways, and then recorded everything that happened.
They didn't just break things randomly; they organized the "breaks" into four distinct categories, like a mechanic's checklist:
- Performance Level: Making the engine run hot or the network slow (like a traffic jam).
- Service Level: Making one booth refuse to talk to another (like a broken phone line between the ticket seller and the gate).
- Database Level: Filling up the storage room so no new items can be stored (like a connection pool running dry).
- Code Level: Introducing logic errors, like a cashier who accidentally gives free tickets because of a typo in their instructions.
3. The "Five Senses" Approach
The most unique part of this dataset is that they didn't just record the "speed" of the station. They recorded five different types of data (modalities) simultaneously, giving a 360-degree view:
- Logs: The written diary of every action the booths took.
- Metrics: The speedometers and fuel gauges (CPU, memory usage).
- Traces: A map showing exactly which path a passenger took through every single booth.
- API Responses: What the passenger actually saw (Did they get a ticket? Did they get an error message?). This is the "user experience."
- Code Coverage Reports: A special highlighter that shows exactly which lines of the cashier's instruction manual were actually read and used. This helps pinpoint if the error was in the code itself.
4. How They Did It (The Experiment)
To build this, they used a robot tester called EvoMaster.
- Phase 1: The robot acted like thousands of customers, trying to buy tickets and use the system to make sure everything was working normally.
- Phase 2: The team injected the "breaks" (anomalies) into the system. They didn't just pick random booths; they targeted the most critical ones to see how the whole station reacted.
- Phase 3: They recorded all five data types while the robot kept working, capturing the chaos in real-time.
5. Why This Matters
The authors claim this dataset is a game-changer because it allows researchers to:
- Connect the dots: See how a tiny code error (a typo in the manual) eventually leads to a user seeing a "System Error" message.
- Find the root cause: Instead of just knowing "the system is slow," they can pinpoint "the database connection pool is exhausted because of a specific code path."
- Test better AI: It gives AI researchers a much richer, more realistic playground to train their "smart detectives" to fix these complex systems automatically.
In short: The paper introduces a new, highly detailed "crime scene" dataset where they broke a microservice system in four different ways and recorded it with five different cameras. This helps future AI tools learn not just that something is wrong, but exactly what is wrong and where to fix it.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.