slurm-receipt 0.1.0

See what your HPC compute really cost -- in dollars, energy, and burgers.

slurm-receipt

See what your HPC compute really cost — in dollars, energy, and burgers.

A zero-dependency terminal tool that reads your Slurm job history via sacct and generates an interactive receipt showing compute charges, energy usage, AWS cost equivalents, and fun real-world conversions. It also roasts you about your failed jobs with data-backed commentary.

  ====================================================

                       _[==]_
                      |o    o|
                      |______|
                   The General Store
  ====================================================

    Customer:  you
    Period:    2026-03-25 -> 2026-04-07 20:24
    Days:      14

  ----------------------------------------------------
                     ORDER SUMMARY
  ----------------------------------------------------

    Jobs submitted .......................... 26,189
      Completed ............................ 26,085
      Failed ....................................57
      Cancelled .................................21
      Timed out ..................................3

    Success: [##########################..] 100%

Install

# From PyPI (recommended)
pip install slurm-receipt

# Or with pipx (isolated install)
pipx install slurm-receipt

# Or from source
git clone https://github.com/chen-hsieh/slurm-receipt.git
cd slurm-receipt
pip install -e .

Requirements: Python 3.8+, access to sacct (any Slurm cluster), a terminal with curses support. Zero external Python dependencies.

Quick start

# Interactive receipt for the last 30 days
slurm-receipt

# Last 90 days
slurm-receipt --days 90

# Custom date range
slurm-receipt --start 2025-01-01 --end 2025-12-31

# Plain text output (no TUI)
slurm-receipt --snap

# Save to a specific file
slurm-receipt --snap-file my_receipt.txt

# Try it without a Slurm cluster (synthetic data)
slurm-receipt --demo

# Add UGA Bulldogs flavor to roasts
slurm-receipt --uga

Examples

Main receipt

The receipt shows your compute charges like a store receipt:

  ----------------------------------------------------
                    COMPUTE CHARGES
  ----------------------------------------------------

    CPU time ........................ 18,597 core-hrs
    Wall time ......................... 1,008 hrs
    Memory ........................ 176,483 GB-hrs

    Energy (CPU) ..................... 148.78 kWh
    Cooling overhead ......................... x1.3
    ------------------------------------------------
    TOTAL ENERGY ..................... 193.41 kWh
    CO2 emitted ...................... 77.37 kg

  ----------------------------------------------------
               AWS PRICE CHECK (on-demand)
  ----------------------------------------------------

    Compute (CPU) ........................ $929.87
    Memory ............................... $882.41
    ------------------------------------------------
    TOTAL ............................ $1,812.28

  ****************************************************
                    BUT ACTUALLY...
  ****************************************************

                    ~B~  276.3
                    burgers grilled

              "Enough to run a food truck for a day"

                 [1/37] (food)

Activity report

Press h to see your submission patterns — weekly volume, day-of-week, and time-of-day breakdowns:

                    WEEKLY BREAKDOWN
    ------------------------------------------------
       Mar 25-31  [####################] 25971 <-
       Apr 01-07  [#.....................]   218

                      DAY OF WEEK
    ------------------------------------------------
    Monday      [#####.............]    43
    Tuesday     [##################]    63 <-
    Wednesday   [##############....]    55
    Thursday    [#.................]     4
    Friday      [###########.......]    40
    Saturday    [############......] 25946 <-
    Sunday      [###...............]    25

                      TIME OF DAY
    ------------------------------------------------
      12am-4am  [....................]     0
       4am-8am  [....................]     0
      8am-12pm  [####................]  2593
      12pm-4pm  [####################] 10382
       4pm-8pm  [################....]  7786
      8pm-12am  [###########.........]  5428

Performance review (roasts)

Press r to see data-driven roasts with a mini receipt showing the referenced job info:

    25,963 jobs on 2026-03-28.
    That's one every 3s.
    Were you okay?

    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    | RELATED JOB INFO                              |
    |----------------------------------------------|
    |  Detail:  25,963 jobs on 2026-03-28          |
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

                        [1/6]

Top jobs

Press t to see your hungriest jobs by CPU-hours, plus the failure hall of shame:

                 TOP 10 HUNGRIEST JOBS
  ====================================================

    #   Job Name                  CPU-hrs
    ------------------------------------------------
     1.  genera_717hap1              9,216
     2.  genera_717hap2              8,832
     3.  build_diamond_nr              768
     4.  syn_ks                         36
     ...

Interactive controls

Key	Action
< / >	Rotate through 37 fun conversions
r	Performance review (roasts)
m	Monthly ledger
h	Activity report (weekly/daily/hourly)
t	Top jobs + failure hall of shame
s	Save receipt to ~/slurm_receipt_30d.txt + copy to clipboard
j/k or arrows	Scroll
PgUp/PgDn	Scroll fast
b	Back to main receipt
q	Quit (prints save location)

Mouse scroll and status bar clicks also work in supported terminals.

What it calculates

Metric	Method	Source
CPU energy	core-hours x 8W per core / 1000	Blended avg across HPC nodes
GPU energy	GPU-hours x GPU TDP / 1000	A100=400W, H100=700W, L4=72W, etc.
Cooling overhead	Total energy x 1.3 PUE	Industry standard data center PUE
CO2 emissions	kWh x 0.4 kg/kWh	US Southeast electricity grid
AWS cost	On-demand list prices, US regions	2026 pricing

Fun conversions (37)

Rotate through with </>:

• Food: burgers, cakes, ramen, espresso, toast, pizza, popcorn, rice
• Energy: phone charges, laptop charges, AA batteries, lightning bolts
• Transport: Tesla miles, e-bike miles, transatlantic flights
• Household: laundry loads, showers, houses powered, fridge days
• Entertainment: Netflix hours, gaming hours, vinyl albums
• Scale: ISS orbits, Bitcoin transactions, ChatGPT queries, MRI scans
• Memory: novels in RAM, photos, human genomes, full Wikipedias
• Environment: trees to offset, CO2 balloons, soda cans

Roast categories

All roasts reference your actual data with a mini receipt panel:

• Failure rate analysis (any % triggers something)
• Instant failures (<10s) with job name
• Slow painful failures (hours then FAILED) with job name
• Array job burst detection
• Night owl patterns (10pm–6am submissions)
• Weekend warrior detection
• Peak submission hour commentary
• Short job inefficiency (<60s completed)
• Single-core job usage
• Memory usage patterns
• Partition loyalty/diversity
• Cancellation habits
• Top CPU hog identification
• Day-of-week patterns (Monday panic, Friday submit-and-pray)

Mascot tiers

Your shop mascot changes based on total CPU-hours:

CPU-hours	Examples
< 100	The Lemonade Stand, The Penny Jar
100 – 1K	The Corner Bodega, The Ramen Cart
1K – 10K	The General Store, The Diner
10K – 50K	The Warehouse, The Department Store
50K – 200K	The Mega Depot, The Data Cathedral
> 200K	The Compute Empire, The Supercomputer

CLI reference

usage: slurm-receipt [-h] [--days DAYS] [--user USER] [--start START]
                     [--end END] [--snap] [--snap-file PATH]
                     [--no-copy] [--uga] [--demo]

See what your HPC compute really cost.

options:
  --days DAYS      Days to look back (default: 30)
  --user USER      Slurm username (default: $USER)
  --start START    Start date YYYY-MM-DD (overrides --days)
  --end END        End date YYYY-MM-DD (default: today)
  --snap           Print receipt + save to ~/slurm_receipt_Nd.txt (no TUI)
  --snap-file PATH Save receipt to file
  --no-copy        Don't auto-copy to clipboard on snap
  --uga            Add UGA Bulldogs flavor to roasts
  --demo           Demo with synthetic data (no sacct needed)

Clipboard

The s key saves the receipt and tries to copy to clipboard:

1. OSC 52 — works over SSH if your terminal supports it (writes to /dev/tty)
2. tmux buffer — tmux load-buffer (paste with prefix + ])
3. xclip / xsel / wl-copy — local clipboard fallback

On quit, the auto-saved receipt location is printed:

  Receipt saved to: /home/you/slurm_receipt_30d.txt

License

MIT