dockpulse 0.5.0

Container Resource Profiler & Right-Sizer for Docker

DockPulse

Your Docker containers are wasting resources. DockPulse tells you exactly how much.

---

The Problem

Most Docker containers run with either no resource limits (risking OOM kills and noisy neighbors) or wildly over-provisioned limits set by guesswork. Kubernetes has Vertical Pod Autoscaler for right-sizing, but standalone Docker has nothing.

DockPulse fills this gap. It profiles your containers over time, computes percentile-based resource usage, identifies waste, and automatically rewrites your docker-compose.yml with data-driven resource limits.

Features

• Container Profiling -- Collect CPU, memory, network, and block I/O statistics from running containers over configurable time windows (minutes to days)
• Percentile Analysis -- Compute p50 / p95 / p99 resource usage with anomaly detection for memory pressure, CPU spikes, and chronic over-provisioning
• Right-Sizing Engine -- Generate recommended deploy.resources.limits and reservations based on observed p95 usage plus configurable headroom
• Compose Rewriter -- Automatically patch docker-compose.yml files with optimized limits while preserving comments and formatting (via ruamel.yaml)
• Waste Reports -- Quantify total memory and CPU waste across your entire stack with actionable savings numbers
• Live Dashboard -- Real-time Rich terminal UI with CPU sparklines, memory bars, and color-coded health indicators
• SQLite Persistence -- All profiling data is stored locally with zero external dependencies
• Multiple Output Formats -- Terminal (Rich), JSON, and styled HTML reports
• Cloud Cost Estimation -- Map resource waste to real dollar amounts for AWS Fargate, GCP Cloud Run, and Azure ACI with per-container savings breakdown
• Grafana Dashboard -- Pre-built Grafana dashboard with 5 rows of panels covering CPU, memory, network, disk I/O, and PIDs across all monitored containers
• Startup Time Profiling -- Measure container startup times (create → running → healthy) with multi-run averaging and compose-file support

Quick Start

Installation

pip install dockpulse

Or install from source:

git clone https://github.com/hariharanragothaman/dockpulse.git
cd dockpulse
pip install -e ".[dev]"

Or run via Docker:

docker run --rm -v /var/run/docker.sock:/var/run/docker.sock \
  hariharanragothaman/dockpulse profile --duration 1h

Basic Usage

# Profile all running containers for 30 minutes
dockpulse profile --duration 30m

# Profile specific containers for 2 hours at 5-second intervals
dockpulse profile --duration 2h --containers web,db,redis --interval 5

# Analyze collected data
dockpulse analyze

# Right-size a compose file with 25% headroom
dockpulse right-size docker-compose.yml --headroom 25 -o docker-compose.optimized.yml

# View live dashboard
dockpulse dashboard

# Generate a waste report
dockpulse waste

CLI Reference

dockpulse profile

Profile running containers and record resource usage to a local SQLite database.

Option	Default	Description
--duration, -d	1h	Profiling duration (e.g. 30m, 1h, 2h30m, 1d)
--containers, -c	all	Comma-separated container IDs or names
--interval, -i	1.0	Seconds between stat samples

$ dockpulse profile --duration 30m
Profiling 3 containers for 30m (interval=1.0s)
  web      | collected 1800 samples
  db       | collected 1800 samples
  redis    | collected 1800 samples
Done. 5400 samples saved to ~/.dockpulse/profiles.db

dockpulse analyze

Analyze the most recent profile and display results.

Option	Default	Description
--format, -f	rich	Output format: rich, json, or html
--output, -o	--	Output file path (required for json/html)

$ dockpulse analyze
Container: web
  CPU   p50=12.3%  p95=34.1%  p99=52.8%   peak=67.2%
  MEM   p50=180MB  p95=245MB  p99=312MB   limit=1024MB
  Anomalies: Over-provisioned (p95 memory is 24% of limit)

Container: db
  CPU   p50=4.1%   p95=18.6%  p99=29.4%   peak=41.0%
  MEM   p50=420MB  p95=510MB  p99=580MB   limit=2048MB
  Anomalies: Over-provisioned (p95 memory is 25% of limit)

Container: redis
  CPU   p50=0.8%   p95=2.1%   p99=3.4%    peak=5.1%
  MEM   p50=28MB   p95=35MB   p99=42MB    limit=512MB
  Anomalies: Over-provisioned (p95 memory is 7% of limit)

dockpulse right-size

Right-size a Docker Compose file based on profiled resource usage.

Argument / Option	Default	Description
COMPOSE_FILE	required	Path to the Docker Compose file
--headroom, -H	20	Headroom percentage above p95
--output, -o	auto	Output path for optimized file

$ dockpulse right-size docker-compose.yml --headroom 25
--- docker-compose.yml
+++ docker-compose.optimized.yml
@@ services.web.deploy.resources @@
+    limits:
+      memory: 306M
+      cpus: '0.43'
+    reservations:
+      memory: 180M

@@ services.db.deploy.resources @@
-    limits:
-      memory: 2048M
+    limits:
+      memory: 638M
+      cpus: '0.24'
+    reservations:
+      memory: 420M

Savings: 1.44 GB memory, 1.83 CPU cores freed
Written to docker-compose.optimized.yml

dockpulse dashboard

Launch a live terminal dashboard with real-time resource monitoring.

$ dockpulse dashboard
+----------------------------------------------------------------+
|  DockPulse - Container Resource Monitor          Ctrl+C to exit |
|                                                                 |
|  Container  CPU (sparkline)  Avg CPU  Memory         Status     |
|  web        ▂▃▅▃▂▁▂▃▆▄     12.3%    ██████░░ 45.2%  HEALTHY   |
|  db         ▁▁▂▁▁▁▁▂▃▂      4.1%    ████░░░░ 31.0%  HEALTHY   |
|  redis      ▁▁▁▁▁▁▁▁▁▁      0.8%    █░░░░░░░  8.2%  HEALTHY   |
|  worker     ▃▅▇▅▃▅▇█▇▅     78.4%    ███████░ 88.1%  WARNING   |
+----------------------------------------------------------------+

dockpulse waste

Show a waste report for the most recent profiling session.

$ dockpulse waste
DockPulse Waste Report
======================
Container   Allocated  Used(p95)  Wasted     Utilization
web         1024 MB    245 MB     779 MB     24%
db          2048 MB    510 MB     1538 MB    25%
redis       512 MB     35 MB      477 MB     7%
-------------------------------------------------------
Total       3584 MB    790 MB     2794 MB    22%

You are wasting 2.73 GB of memory and 2.1 CPU cores across 3 containers.
Right-size with: dockpulse right-size docker-compose.yml

dockpulse cost

Estimate cloud infrastructure costs and potential savings.

Option	Default	Description
--provider, -p	aws	Cloud provider: aws, gcp, azure
--hours	730	Monthly running hours
--headroom, -H	20	Headroom percentage for right-sizing
--format, -f	rich	Output format: rich, json

$ dockpulse cost --provider aws
Cloud Cost Estimate (aws_fargate)
Container   Current $/mo   Optimized $/mo   Savings $/mo
web         $32.14         $18.72           $13.42
db          $48.90         $27.55           $21.35
redis       $8.21          $2.10            $6.11
─────────────────────────────────────────────────────────
TOTAL       $89.25         $48.37           $40.88

Based on 730 hours/month at aws_fargate pricing.

dockpulse startup

Profile container startup times.

Argument / Option	Default	Description
IMAGE	--	Docker image to profile
--compose, -C	--	Path to docker-compose.yml
--runs, -n	3	Number of runs to average
--format, -f	rich	Output format: rich, json

$ dockpulse startup nginx:latest --runs 5
Startup Time Profile
Container   Image          Create→Running   Running→Healthy   Total      Image Size   Healthcheck
nginx       nginx:latest   142ms            —                 142ms      187.8 MB     ✗

Live Demo: Top 10 Docker Hub Images

See DockPulse in action profiling the most popular containers on Docker Hub — nginx, redis, postgres, python, node, memcached, mysql, mongo, httpd, and rabbitmq — with deliberately over-provisioned limits so you can see waste detection, right-sizing, and Grafana dashboards working end to end.

cd examples/demo
./run-demo.sh

The script pulls all 10 images, starts a traffic generator, profiles for 3 minutes, launches Prometheus + Grafana, and generates analysis, waste, cost, and HTML reports automatically. See examples/demo/README.md for options.

Grafana Dashboards

DockPulse ships with three pre-built Grafana dashboards and Prometheus recording/alerting rules. To use them with the bundled Prometheus setup:

cd examples/prometheus
docker compose up -d
dockpulse export --port 9090   # start the Prometheus exporter

# Open Grafana at http://localhost:3000 (admin/admin)
# All three dashboards are auto-provisioned

To point Prometheus at a custom exporter address, set the environment variable before starting:

DOCKPULSE_EXPORTER_TARGET=my-host:9090 docker compose up -d

Container Resource Overview (dockpulse-overview)

• Overview row -- stat panels for total containers, avg CPU%, avg memory%, total network I/O
• CPU row -- time series of CPU usage per container with threshold lines
• Memory row -- time series of memory usage vs limits, plus memory utilization gauges
• Network & Disk I/O row -- time series of RX/TX bytes and block read/write
• Processes row -- time series of PID counts per container

Alerts & Thresholds (dockpulse-alerts)

• Active alerts list -- shows currently firing DockPulse alerts
• CPU/Memory warning counters -- stat panels showing how many containers exceed thresholds
• CPU & Memory with threshold overlays -- time series with 80%/95% warning/critical lines
• 5-minute rolling aggregates -- CPU p95 and memory averages from Prometheus recording rules

Right-Sizing & Waste Analysis (dockpulse-rightsizing)

• Fleet utilization gauge -- overall memory utilization across all containers
• Waste metrics -- estimated memory waste in bytes and as a percentage
• Per-container bar gauges -- memory and CPU utilization with waste-detection coloring
• Waste over time -- stacked time series showing the gap between limits and actual usage
• Usage vs Limits -- side-by-side comparison of actual usage against configured limits

All dashboards support the $container template variable for filtering and link to each other for easy navigation.

Prometheus Recording & Alerting Rules

The bundled recording_rules.yml provides pre-computed aggregates and alert definitions:

Recording rules -- 5-minute rolling CPU/memory averages and p95s, fleet-wide utilization ratios, per-container network and disk I/O totals.

Alerting rules:

Alert	Condition	Severity
DockPulseHighCPU	CPU > 80% for 5m	warning
DockPulseCriticalCPU	CPU > 95% for 2m	critical
DockPulseHighMemory	Memory > 85% for 5m	warning
DockPulseCriticalMemory	Memory > 95% for 2m	critical
DockPulseMemoryWaste	<20% memory used for 30m	info
DockPulseHighPIDs	PIDs > 500 for 5m	warning
DockPulseContainerDown	Exporter unreachable for 1m	critical

Architecture

Data Flow

flowchart LR
    Docker["Docker Daemon"]
    Collector["StatsCollector"]
    DB["SQLite DB"]
    Analyzer["Analyzer"]
    Profile["ProfileResult"]
    RightSizer["RightSizer"]
    Compose["ComposeRewriter"]
    Reporter["Reporter"]
    Dashboard["Dashboard"]
    Visualizer["Visualizer"]
    Prometheus["PrometheusExporter"]

    Docker -->|"/containers/stats API"| Collector
    Collector -->|"persist samples"| DB
    Collector -->|"live stream"| Dashboard
    Collector -->|"live stream"| Prometheus
    DB -->|"load samples"| Analyzer
    Analyzer -->|"percentiles + anomalies"| Profile
    Profile --> RightSizer
    Profile --> Reporter
    Profile --> Visualizer
    RightSizer -->|"recommendations"| Compose
    Compose -->|"optimized YAML"| ComposeFile["docker-compose.yml"]
    Reporter -->|"JSON / HTML / terminal"| Output["Reports"]
    Visualizer -->|"Plotly charts"| HTMLReport["Interactive HTML"]
    Dashboard -->|"Rich live UI"| Terminal["Terminal"]
    Prometheus -->|"/metrics"| PromScrape["Prometheus / Grafana"]

Module Dependency Graph

flowchart TD
    CLI["cli.py"]
    Collector["collector.py"]
    AnalyzerMod["analyzer.py"]
    Models["models.py"]
    Config["config.py"]
    DashboardMod["dashboard.py"]
    ReporterMod["reporter.py"]
    RightSizerMod["rightsizer.py"]
    ComposeRewriterMod["compose_rewriter.py"]
    VisualizerMod["visualizer.py"]
    PrometheusMod["prometheus.py"]
    CostMod["cost.py"]
    StartupMod["startup.py"]

    CLI --> Collector
    CLI --> AnalyzerMod
    CLI --> DashboardMod
    CLI --> ReporterMod
    CLI --> RightSizerMod
    CLI --> ComposeRewriterMod
    CLI --> VisualizerMod
    CLI --> PrometheusMod
    CLI --> CostMod
    CLI --> StartupMod
    CLI --> Config
    CLI --> Models
    Collector --> Models
    AnalyzerMod --> Models
    DashboardMod --> Models
    ReporterMod --> Models
    RightSizerMod --> Models
    ComposeRewriterMod --> Models
    CostMod --> Models
    StartupMod --> Models
    PrometheusMod --> Collector

CLI Command Map

flowchart TD
    CLI["dockpulse"]
    Profile["profile"]
    Analyze["analyze"]
    RightSize["right-size"]
    Dash["dashboard"]
    Waste["waste"]
    Report["report"]
    Sessions["sessions"]
    Compare["compare"]
    Stack["stack"]
    Clean["clean"]
    Export["export"]
    Cost["cost"]
    Startup["startup"]

    CLI --> Profile
    CLI --> Analyze
    CLI --> RightSize
    CLI --> Dash
    CLI --> Waste
    CLI --> Report
    CLI --> Sessions
    CLI --> Compare
    CLI --> Stack
    CLI --> Clean
    CLI --> Export
    CLI --> Cost
    CLI --> Startup

    Profile ---|"collect stats over time"| SQLite["SQLite DB"]
    Analyze ---|"percentile analysis"| TermOut["Terminal / JSON / HTML"]
    RightSize ---|"optimize limits"| ComposeOut["Compose YAML"]
    Dash ---|"live monitoring"| LiveUI["Rich Live UI"]
    Waste ---|"quantify waste"| WasteOut["Waste Report"]
    Report ---|"interactive charts"| PlotlyOut["Plotly HTML"]
    Sessions ---|"list sessions"| SessionTable["Session Table"]
    Compare ---|"diff two sessions"| DeltaTable["Delta Table"]
    Stack ---|"multi-container analysis"| StackOut["Rankings + Bottleneck"]
    Clean ---|"delete data"| CleanDB["SQLite Cleanup"]
    Export ---|"Prometheus metrics"| MetricsOut["/metrics endpoint"]
    Cost ---|"cloud pricing"| CostOut["Cost Report"]
    Startup ---|"timing analysis"| StartupOut["Startup Profile"]

DockPulse talks directly to the Docker daemon via the Docker SDK for Python. Stats are collected using the /containers/{id}/stats API endpoint and persisted to a local SQLite database for offline analysis.

The right-sizing engine applies a configurable headroom percentage on top of observed p95 usage. The compose rewriter uses ruamel.yaml to update files in-place without destroying comments or formatting.

Comparison

Feature	DockPulse	docker stats	Kubernetes VPA
Time-series profiling	Yes	No (snapshot only)	Yes
Percentile analysis	p50/p95/p99	No	Yes
Anomaly detection	Yes	No	No
Compose file rewriting	Yes	No	N/A (k8s only)
Waste quantification	Yes	No	No
Live dashboard	Yes	Basic	No
Works without Kubernetes	Yes	Yes	No
Zero external dependencies	Yes	Yes	No (requires k8s)

Roadmap

• Prometheus metrics export
• Historical trend analysis and regression detection
• GitHub Action for CI resource regression checks
• Interactive Plotly HTML reports
• Session management and comparison
• Multi-container stack analysis
• Slack / Discord alert integration
• Cost estimation (map waste to cloud provider pricing)
• Grafana dashboard templates
• Grafana alerting rules and right-sizing dashboards
• Prometheus recording rules and configurable scrape targets
• Container startup time profiling
• PyPI package publishing with auto-versioning

Contributing

Contributions are welcome! See CONTRIBUTING.md for development setup, workflow, and guidelines.

# Development setup
git clone https://github.com/hariharanragothaman/dockpulse.git
cd dockpulse
pip install -e ".[dev]"

# Run tests
pytest

# Run linter
ruff check src/ tests/

# Run type checker
mypy src/

License

MIT License. See LICENSE for details.

---

Built with the Docker SDK for Python, Typer, and Rich.