Run any command — but never twice too soon.

once is a lightweight Bash wrapper that ensures a command executes at most once per time period (hour/day/week/month) or within a custom cooldown window (like “6h” or “2d”). It’s perfect for scripts, CI jobs, or ad-hoc shell tasks you don’t want to repeat accidentally.

Instead of scheduling jobs in cron, you just prefix your command and run it however many times you want:

# Run it every time you login or something.
$ once --period day -- ./backup.sh
$ once --window 6h -- make deploy

# only once though
$ once --window 1h -- echo 'yoohoo'
yoohoo
$ once --window 1h -- echo 'yoohoo'
Skipped: ran 1s ago; window 1h.

once automatically:

• Creates a unique hash for the command (including args and working dir).
• Tracks when it was last run.
• Skips repeated runs within the defined time window.
• Prevents concurrent duplicates via lockfiles.
• Works anywhere — interactive shells, scripts, CI — no cron required.

In short: run anything, but never twice too soon.

# Safe install (download → review → install)
curl -fsSL https://raw.githubusercontent.com/svandragt/once/main/once.sh -o /tmp/once.sh
less /tmp/once.sh # inspect before running and making it executable

sudo install -m 0755 /tmp/once.sh /usr/local/bin/once
chmod +x /usr/local/bin/once

# EXAMPLES
# Run something once per day
once --period day -- ./backup.sh

# Or with a rolling cooldown
once --window 6h -- ./sync.sh

# Explain what would happen
once --explain --dry-run -- ./job.sh

• Each command’s identity is based on its executable path, arguments, working directory, and an optional --key-extra string.

• The identity is hashed (SHA-256) and recorded under:

  ~/.local/state/once/
  

• When invoked again, once checks whether that command has already run during:

  • The same calendar period (hour, day, week, month), or
  • Within a rolling window (e.g., 6h or 2d).

• If it has, it skips execution and exits with code 3.

• It uses atomic lock directories to prevent concurrent runs.

once [--period {hour|day|week|month} | --window {Nh|Nd}] [options] -- <command> [args...]

# Once per hour
once --period hour -- ./poll-api.sh

# Once per week (ISO weeks)
once --period week -- ./report.sh

# Once every 36 hours
once --window 36h -- ./rebuild-index.sh

# Add environment context to key
once --period day --key-extra prod -- ./backup.sh

# Forced run ignoring window
once --force -- ./backup.sh

• Hashes: SHA-256 of a canonical identity string (exe, args, cwd, key-extra).

• State:

  • Period mode: ~/.local/state/once/periods/YYYY-MM-DD/<hash>.stamp
  • Window mode: ~/.local/state/once/windows/<hash>.stamp

• Locks: Atomic mkdir in ~/.local/state/once/locks/<hash>.lock

• Concurrency: Locks prevent two simultaneous invocations of the same job.

• Security: State dir is 0700 (owner-only). No arguments are logged by default.

Stamps accumulate over time. You can safely remove old entries:

find ~/.local/state/once/periods -type d -mtime +60 -exec rm -rf {} +
find ~/.local/state/once/windows -type f -mtime +90 -delete

A future version will include a --gc command for automatic cleanup.

✅ You should use it when:

• You trigger tasks manually or via scripts, but want idempotent frequency control.
• You want to avoid cron, or you’re running in containers/CI where time-based scheduling isn’t reliable.
• You have multiple scripts sharing cooldown logic.

❌ You don’t need it when:

• Your task is already handled by cron/systemd timers/anacron.
• You need distributed coordination across multiple hosts (use Redis locks, etc. instead).

GPL © 2025 Sander van Dragt