Social-Network Resilience

This project is based on the DeathStarBench SocialNetwork benchmark, adapting its workload and service graph for resilience analysis.

This repository implements both theoretical and empirical evaluation of the resilience of a social-network microservices application under container failures. It includes:

• resilience.py: Monte Carlo simulation of service outages using sampling without replacement.
• chaos.sh: Script for injecting failures (killing containers) and measuring live request success rate.
• pipeline_norepl.sh / pipeline_repl.sh: Combined workflows for scenarios without and with replicas.
• results: Directory containing JSON summaries (summary_norepl.json, summary_repl.json).

---

Table of Contents

1. Prerequisites
2. Setup
3. Dependency Graph
4. Theoretical Model (resilience.py)
5. Empirical Test (chaos.sh)
6. Pipeline Workflows
7. Results
8. CI Integration
9. License

---

Prerequisites

• Python 3.8+
• Bash (Unix-like environment)
• Docker & Docker Compose
• jq (JSON processor)

Recommended Python packages (installed via pip):

pip install networkx numpy scipy

jq may be installed via your OS package manager (e.g., apt install jq).

---

Setup

1. Clone this repository:

   git clone https://github.com/a-a-k/socialnet-resilience.git
   cd socialnet-resilience

2. Create & activate a virtual environment:

   python3 -m venv venv
   source venv/bin/activate
   pip install --upgrade pip
   pip install networkx numpy scipy

3. Ensure jq is available:

   jq --version

Dependency Graph

All service dependencies are defined in deps.json (Jaeger trace). The graph is loaded by resilience.py via NetworkX:

with open('deps.json') as f:
    data = json.load(f)['data']
G = nx.DiGraph((e['parent'], e['child']) for e in data)

---

Theoretical Model (resilience.py)

Monte Carlo simulation of container failures without replacement.

Usage:

python3 resilience.py deps.json \
  --samples 900000 \
  --p_fail 0.30 \
  [--repl 0|1] \
  -o results/R_model_{repl|norepl}.json

• --repl: enable replica counts (uses replicas map).

• Output JSON fields:

  • R_avg: weighted average availability.
  • R_ep: per-endpoint availability.

---

Empirical Test (chaos.sh)

Docker Compose–based chaos experiment: kills a fraction of containers and measures successful HTTP requests.

Usage:

# Without replicas
auth: # assume services are running via docker-compose
./chaos.sh --repl 0 -o results/summary_norepl.json

# With replicas
auth:
./chaos.sh --repl 1 -o results/summary_repl.json

• Output JSON field:

  • R_live: observed success rate.

---

Pipeline Workflows

Two separate pipelines collect theoretical and empirical results into a single summary file each.

pipeline_norepl.sh

#!/usr/bin/env bash
set -euo pipefail

# Prepare environment
./01_prepare_env.sh
tests: ./02_steady_norepl.sh
./chaos.sh --repl 0

# Combine summaries
jq -n \
  --arg m "$(jq .R_avg  results/norepl/R_avg_base.json)" \
  --arg l "$(jq .R_live results/norepl/summary.json)" \
  '{R_model_norepl: ($m|tonumber), R_live_norepl: ($l|tonumber)}' \
  > results/summary_norepl.json

pipeline_repl.sh

#!/usr/bin/env bash
set -euo pipefail

# Prepare environment
./01_prepare_env.sh
./04_steady_repl.sh --repl 1
./chaos.sh --repl 1

# Combine summaries
jq -n \
  --arg m "$(jq .R_avg  results/repl/R_avg_repl.json)" \
  --arg l "$(jq .R_live results/repl/summary.json)" \
  '{R_model_repl: ($m|tonumber), R_live_repl: ($l|tonumber)}' \
  > results/summary_repl.json

Each pipeline writes its combined summary to results/summary_{norepl,repl}.json.

---

Results

After running a pipeline, inspect the combined summary:

cat results/summary_norepl.json
# {
#   "R_model_norepl": 0.179,
#   "R_live_norepl": 0.193
# }

Actual results for v1.0.0 release:

Without replication:

{
  "R_model_norepl": 0.161,
  "R_live_norepl": 0.1845
}

With replication:

{
  "R_model_repl": 0.305,
  "R_live_repl": 0.3046
}

---

CI Integration

In GitHub Actions, upload summaries as artifacts:

- name: Run pipelines
  run: |
    ./pipeline_norepl.sh
    ./pipeline_repl.sh

- name: Upload summaries
  uses: actions/upload-artifact@v3
  with:
    name: resilience-results
    path: results/summary_*.json

---

License

MIT License. See LICENSE for details.