anon-ops/ops-lite

Dataset

1

stars

29

commits

1

linked in READMEs

May 13, 2026

updated

aiops
benchmark
causal-graph
chaos-engineering
microservices
observability
rca
root-cause-analysis

README

ops-lite

A curated 500-case root-cause-analysis (RCA) evaluation set for microservice systems, with manifest-driven causal-graph ground truth.

Each case bundles:

  • a chaos-injection ground truth (injection.json)
  • a causal service graph derived from the injection's fault contract (causal_graph.json)
  • the runtime environment snapshot (env.json, result.json, label.txt)
  • 12 parquet metric tables per case, split into the abnormal window (during fault) and the normal window (baseline)

The corpus spans three open-source microservice testbeds:

systemndescription
ts (Train-Ticket)320Java/Spring Cloud microservice system, 44 application services
hs (Hotel Reservation / DeathStarBench)142Go/gRPC microservice system, 9 application services
otel-demo38OpenTelemetry Demo e-commerce app, 15 polyglot application services

Intended use

Benchmarking RCA algorithms on microservice fault propagation. Each case provides the ground-truth root-cause service(s) and a manifest-derived causal graph against which an algorithm's predicted ranking or path can be scored.

Pipeline

  1. Generation — chaos faults are injected and observed by AegisLab, an open-source RCA benchmarking platform. A detector confirms each case is observable end-to-end before it enters the candidate pool.
  2. Annotation — for every observable case, a manifest-driven causal graph reasoner (rcabench-platform v3 internal/reasoning) enumerates the fault propagation layer-by-layer from the registered fault manifest, producing the causal_graph.json ground truth.
  3. Curation — a 1464-case raw pool is reduced to 500 via a greedy selector with hard filters (cyclic graphs, lp ≤ 1, frontend-only injections) and soft caps on system / chaos-family / root-service.

This repo card summarizes the released artifact itself. The full generation and selection pipeline will be documented in the associated paper / artifact release.

Data layout

ops-lite/
├── manifest.jsonl            # 500 lines, one JSON record per case
├── README.md
├── croissant.json            # core Croissant + RAI fields
└── cases/
    └── <case-name>/
        ├── injection.json
        ├── causal_graph.json
        ├── env.json
        ├── result.json
        ├── label.txt
        └── *.parquet         # abnormal_* + normal_*

manifest.jsonl schema:

{
  "name": "ts0-ts-order-service-exception-l2bqm5",
  "system": "ts|hs|otel-demo",
  "longest_path": 7,
  "n_svc": 13,
  "n_edge": 21,
  "n_alarm_svc": 2,
  "root_services": ["ts-order-service"],
  "chaos_family": "JVM*|HTTP*|Network*|Pod*|*Stress|DNS|Time|hybrid_clean|hybrid_kill",
  "primary_kind": "<chaos_type or 'hybrid'>",
  "subtypes": ["..."],
  "hybrid": false,
  "has_kill_leg": false
}

Composition

metricvalue
total cases500
mean longest_path3.18
mean n_edge4.06
mean n_svc3.97
chaos families8
systems3

License

Apache-2.0. The underlying microservice testbeds (Train-Ticket, Hotel Reservation / DeathStarBench, OpenTelemetry Demo) retain their own upstream licenses.

Citation

A formal citation will be added on publication.

Contributors

anon-ops

29 commits

anon-ops/ops-lite

Dataset

1

stars

29

commits

1

linked in READMEs

May 13, 2026

updated

aiops
benchmark
causal-graph
chaos-engineering
microservices
observability
rca
root-cause-analysis

README

ops-lite

A curated 500-case root-cause-analysis (RCA) evaluation set for microservice systems, with manifest-driven causal-graph ground truth.

Each case bundles:

  • a chaos-injection ground truth (injection.json)
  • a causal service graph derived from the injection's fault contract (causal_graph.json)
  • the runtime environment snapshot (env.json, result.json, label.txt)
  • 12 parquet metric tables per case, split into the abnormal window (during fault) and the normal window (baseline)

The corpus spans three open-source microservice testbeds:

systemndescription
ts (Train-Ticket)320Java/Spring Cloud microservice system, 44 application services
hs (Hotel Reservation / DeathStarBench)142Go/gRPC microservice system, 9 application services
otel-demo38OpenTelemetry Demo e-commerce app, 15 polyglot application services

Intended use

Benchmarking RCA algorithms on microservice fault propagation. Each case provides the ground-truth root-cause service(s) and a manifest-derived causal graph against which an algorithm's predicted ranking or path can be scored.

Pipeline

  1. Generation — chaos faults are injected and observed by AegisLab, an open-source RCA benchmarking platform. A detector confirms each case is observable end-to-end before it enters the candidate pool.
  2. Annotation — for every observable case, a manifest-driven causal graph reasoner (rcabench-platform v3 internal/reasoning) enumerates the fault propagation layer-by-layer from the registered fault manifest, producing the causal_graph.json ground truth.
  3. Curation — a 1464-case raw pool is reduced to 500 via a greedy selector with hard filters (cyclic graphs, lp ≤ 1, frontend-only injections) and soft caps on system / chaos-family / root-service.

This repo card summarizes the released artifact itself. The full generation and selection pipeline will be documented in the associated paper / artifact release.

Data layout

ops-lite/
├── manifest.jsonl            # 500 lines, one JSON record per case
├── README.md
├── croissant.json            # core Croissant + RAI fields
└── cases/
    └── <case-name>/
        ├── injection.json
        ├── causal_graph.json
        ├── env.json
        ├── result.json
        ├── label.txt
        └── *.parquet         # abnormal_* + normal_*

manifest.jsonl schema:

{
  "name": "ts0-ts-order-service-exception-l2bqm5",
  "system": "ts|hs|otel-demo",
  "longest_path": 7,
  "n_svc": 13,
  "n_edge": 21,
  "n_alarm_svc": 2,
  "root_services": ["ts-order-service"],
  "chaos_family": "JVM*|HTTP*|Network*|Pod*|*Stress|DNS|Time|hybrid_clean|hybrid_kill",
  "primary_kind": "<chaos_type or 'hybrid'>",
  "subtypes": ["..."],
  "hybrid": false,
  "has_kill_leg": false
}

Composition

metricvalue
total cases500
mean longest_path3.18
mean n_edge4.06
mean n_svc3.97
chaos families8
systems3

License

Apache-2.0. The underlying microservice testbeds (Train-Ticket, Hotel Reservation / DeathStarBench, OpenTelemetry Demo) retain their own upstream licenses.

Citation

A formal citation will be added on publication.

Contributors

anon-ops

29 commits