Skip to content
Other

ARMS Kafka Workload Classifier

Apache Kafka cluster workload classifier

2 min readupdated

A lightweight model built on LogisticRegression from the scikit-learn library (paired with StandardScaler for feature normalization) that identifies an Apache Kafka cluster's current operating mode using four system metrics. It classifies workloads into real-time streaming (REAL_TIME_EVENT_DRIVEN) and resource-heavy batch processing (BATCH_DATA_INTENSIVE). Created by Nwaoghor Praise Chukunweiken as part of a master's thesis, it serves as the decision module for the ARMS system to select the right recovery strategy upon broker failure instead of relying on a single rigid configuration. Useful for DevOps engineers and distributed systems developers experimenting with adaptive message broker management.

Key features

  • Operates on just four Prometheus input metrics: request rate variance, incoming message variance, disk write rate, and available memory.
  • Inference takes less than 0.1 ms on CPU, and the model ships as two compact joblib/pickle files.
  • Paired with ARMS logic, it selects aggressive rebalancing (QUICK_REBALANCE) for real-time workloads or a conservative mode (RESOURCE_OPTIMIZED) for batch workloads at 90%+ confidence, falling back to gradual CONTROLLED_GRADUAL in borderline cases.
  • The public repository includes a ready-to-use Docker stack with Kafka, Prometheus, and Grafana, an inference.py script, and code for retraining on custom data.

What you can do

  • Integrate the classifier into a Prometheus monitoring pipeline to automatically switch Kafka broker rebalancing settings as traffic profiles shift.
  • Retrain the model on metrics from your own production cluster using the repository's train_model.py script.

Example

import joblib, pandas as pd
from huggingface_hub import hf_hub_download

repo = "nwaoghorpraise/arms-kafka-workload-classifier"
model = joblib.load(hf_hub_download(repo, "kafka_logistic_regression_model.pkl"))
scaler = joblib.load(hf_hub_download(repo, "kafka_feature_scaler.pkl"))

snapshot = pd.DataFrame([[96.19, 5.50e7, 235.35, 5.176e9]], columns=["request_rate_variance", "node_disk_write_bytes_total", "message_rate_variance", "node_memory_MemAvailable_bytes"])
print(model.predict_proba(scaler.transform(snapshot))[0])
Read moreDetails
Holdout accuracy (600 samples) is 0.9617, and 5-fold cross-validation accuracy is 0.9677.
Within ARMS, it cuts mean recovery time for real-time workloads by 6.14 s (p < 0.001).
Trained on 3,000 synthetic snapshots from a single 3-broker Docker cluster; fine-tuning on your own metrics is recommended for production.
Requires scikit-learn 1.6.x; files are saved in pickle format.
Examples
Discuss Telegram 1

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # nwaoghorpraise/arms-kafka-workload-classifier
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download nwaoghorpraise/arms-kafka-workload-classifier --local-dir ~/models/arms-kafka-workload-classifier
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of arms-kafka-workload-classifier (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

nwaoghorpraise/arms-kafka-workload-classifier
Author
nwaoghorpraise
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments