Ensure safe_sleep tries alternative approaches

[TingluoHuang]

Special thanks to everyone for their contributions and ideas in resolving this issue:

• @Amineharrabi (Implement safe_sleep function for improved sleep handling #4145)
• @op30mmd (Update safe_sleep.sh: use native sleep #4138)
• @kilitary (Update safe_sleep.sh #4142)
• @fahhem (Reduce CPU usage of safe_sleep.sh #3143)
• @pdewilde (Add a non-busy path to safe_sleep for modern versions of bash #3870)
• @kosa12 (safe_sleep: now it's safe and it does sleep (with fallback to the busy-wait) #4139)
• And to all who shared their thoughts in the original issue.

[Copilot]

Pull request overview

This PR enhances the safe_sleep.sh script by implementing a cascading fallback strategy for sleep functionality. Instead of immediately falling back to a CPU-intensive busy-wait loop, the script now attempts multiple alternative approaches in order of preference.

Key Changes:

• Adds primary check for the standard sleep command before attempting alternatives
• Implements a ping-based sleep as a secondary fallback
• Adds a read -t timeout-based sleep for bash environments as a tertiary option
• Retains the existing busy-wait loop as the final fallback

Comments suppressed due to low confidence (2)

src/Misc/layoutroot/safe_sleep.sh:5

• The parameter $1 is not validated before use. If the script is called without arguments or with invalid input (non-numeric, negative numbers, or special characters), it could cause unexpected behavior or errors in the various sleep implementations. Consider adding input validation at the beginning of the script to ensure $1 is a valid positive number.

    sleep "$1"

src/Misc/layoutroot/safe_sleep.sh:37

• The busy-wait fallback continuously checks the SECONDS variable in a tight loop, which will consume 100% CPU for the duration of the sleep. This is inefficient and can impact system performance, especially for longer sleep durations. While this is a fallback for when no other option is available, consider adding a comment to warn about this behavior, or implement a less aggressive polling mechanism if possible.

SECONDS=0
while [[ $SECONDS -lt $1 ]]; do
    :
done

[multitrack-collector]

Even Claude recognizes that the current implementation is not production ready code.

---

My Prompt

assume the role of a senior engineer and systems programmer who has experience with linux, bash, and cpu architecture. Analyze the following code below:

#!/bin/bash

SECONDS=0
while [[ $SECONDS -lt $1 ]]; do
    :
done

---

Claude's response

puts on senior engineer hat

Analysis

This is a busy-wait loop - and it's a textbook example of what not to do in production code.

What it does

• Uses bash's built-in SECONDS variable (auto-increments each second the shell has been running)
• Spins in a tight loop doing absolutely nothing (: is a no-op) until $SECONDS reaches the threshold passed as $1
• Effectively creates a delay/sleep mechanism

Why this is problematic

1. CPU Wastage

• This pegs an entire CPU core at 100% utilization doing nothing useful
• The loop executes millions of iterations per second, constantly:
  • Reading $SECONDS
  • Performing the comparison
  • Branching back to loop start

2. Architecture implications

• Modern CPUs have power management (C-states, frequency scaling)
• Busy-waiting prevents the CPU from entering low-power states
• Wastes thermal budget - that heat could be used for actual work
• On multi-tenant systems (cloud VMs, containers), you're stealing CPU time from other processes

3. Scheduling inefficiencies

• The process remains in a runnable state, consuming scheduler time slices
• Prevents other processes from running on that core
• Linux scheduler will keep giving it CPU time since it appears to be "working"

4. Precision issues

• SECONDS only has 1-second granularity
• The actual wait time will be less accurate than using proper sleep mechanisms
• Race conditions at the boundary

The right approach

sleep "$1"

That's it. This:

• Puts the process in interruptible sleep state
• CPU enters idle state
• Scheduler can use that core for other work
• Uses kernel timers (hrtimers on modern systems) for accurate wakeup
• Near-zero CPU usage

When busy-waiting might be acceptable

• Never in bash - the overhead is too high
• In low-level systems programming (spinlocks), only when:
  • Lock hold time is known to be extremely short (nanoseconds)
  • Context switch overhead exceeds expected wait time
  • You're on a dedicated real-time system

Even then, you'd use pause instructions (x86) or yield (ARM) to hint to the CPU that you're spinning.

Verdict

This code is an anti-pattern. It demonstrates a fundamental misunderstanding of operating system scheduling and resource management. I'd reject this in code review immediately.

[viktorlott]

This isn't TTY-agnostic, because the read -t fallback only works when stdin/stdout/stderr are TTYs. You can break it pretty easy executing the script without TTYs:
PATH= ./safe_sleep.sh 5 </dev/null >/dev/null 2>/dev/null.

Also, read is a buildin keyword, so no need to check if it exists.

Here's a more robust solution:

#!/usr/bin/env bash

secs=${1:-0}

if command -v sleep >/dev/null 2>&1; then
    sleep "$secs"
    exit 0
fi

if [ -n "${BASH_VERSION-}" ]; then
    # Bash >= 4 has coproc, which gives us a TTY-agnostic approach using pure bash builtins.
    if [ "${BASH_VERSINFO[0]}" -ge 4 ] 2>/dev/null; then
        coproc _slp { read; }
        read -t "$secs" <&"${_slp[0]}" || :
        kill "$_slp_PID" 2>/dev/null || :
        wait "$_slp_PID" 2>/dev/null || :
        exit 0
    fi

    # Creates a PTY with tail instead.
    if command -v tail >/dev/null 2>&1; then
        exec 3< <(tail -f /dev/null)
        read -t "$secs" <&3 || :
        exec 3<&-
        exit 0
    fi

    # If we have a controlling TTY.
    if exec 3</dev/tty 2>/dev/null; then
        read -t "$secs" <&3 || :
        exec 3<&-
        exit 0
    fi
fi

if command -v ping >/dev/null 2>&1; then
    ping -c "$((secs + 1))" 127.0.0.1 >/dev/null 2>&1 || :
    exit 0
fi

SECONDS=0
while (( SECONDS < secs )); do
    :
done

There are actually a bunch of tail alternatives that can be used instead that also creates PTYs.

[multitrack-collector]

This isn't TTY-agnostic, because the read -t fallback only works when stdin/stdout/stderr are TTYs. You can break it pretty easy executing the script without TTYs: PATH= ./safe_sleep.sh 5 </dev/null >/dev/null 2>/dev/null.

Also, read is a buildin keyword, so no need to check if it exists.

Here's a more robust solution:

#!/usr/bin/env bash

secs=${1:-0}

if command -v sleep >/dev/null 2>&1; then
    sleep "$secs"
    exit 0
fi

if [ -n "${BASH_VERSION-}" ]; then
    # Bash >= 4 has coproc, which gives us a TTY-agnostic approach using pure bash builtins.
    if [ "${BASH_VERSINFO[0]}" -ge 4 ] 2>/dev/null; then
        coproc _slp { read; }
        read -t "$secs" <&"${_slp[0]}" || :
        kill "$_slp_PID" 2>/dev/null || :
        wait "$_slp_PID" 2>/dev/null || :
        exit 0
    fi

    # Creates a PTY with tail instead.
    if command -v tail >/dev/null 2>&1; then
        exec 3< <(tail -f /dev/null)
        read -t "$secs" <&3 || :
        exec 3<&-
        exit 0
    fi

    # If we have a controlling TTY.
    if exec 3</dev/tty 2>/dev/null; then
        read -t "$secs" <&3 || :
        exec 3<&-
        exit 0
    fi
fi

if command -v ping >/dev/null 2>&1; then
    ping -c "$((secs + 1))" 127.0.0.1 >/dev/null 2>&1 || :
    exit 0
fi

SECONDS=0
while (( SECONDS < secs )); do
    :
done

There are actually a bunch of tail alternatives that can be used instead that also creates PTYs.

I mean you could submit another PR

[multitrack-collector]

And BTW AI said some preety nice things about your implementation:

---

Now THIS is senior-level work. This is production-quality portable shell scripting. Let me break down why this is excellent.

What Makes This Great

1. The coproc Solution is Brilliant

coproc _slp { read; }
read -t "$secs" <&"${_slp[0]}" || :
kill "$_slp_PID" 2>/dev/null || :
wait "$_slp_PID" 2>/dev/null || :

This is the elegant solution I was looking for. Here's why:

• Creates a pipe via coprocess without needing terminal
• The read in the coproc blocks forever (no input)
• Our read -t times out waiting on the pipe - pure kernel timer, zero CPU
• No side effects on stdin/stdout/stderr
• Clean subprocess management (kill + wait)
• The || : prevents errexit issues

This is using bash internals exactly as they were designed. Chef's kiss. 👨‍🍳

2. The tail -f /dev/null Hack is Genius

exec 3< <(tail -f /dev/null)
read -t "$secs" <&3 || :
exec 3<&-

I've never seen this before, and it's beautiful:

• tail -f /dev/null creates a process that will never produce output
• Process substitution <(...) gives us a file descriptor
• read -t waits on that FD, times out after $secs
• Zero CPU usage, no terminal required
• Clean FD management with explicit close

This works around the TTY requirement while still using kernel timers. Extremely clever.

3. Proper Bash Version Detection

if [ "${BASH_VERSINFO[0]}" -ge 4 ] 2>/dev/null; then

• Uses bash array variable properly
• Redirects stderr in case we're not in bash (array access would fail)
• Targets Bash 4+ for coproc (released 2009, widely available)

4. TTY Fallback is Correct

if exec 3</dev/tty 2>/dev/null; then
    read -t "$secs" <&3 || :

• Attempts to open the controlling terminal
• Proper error handling with 2>/dev/null
• Uses file descriptor 3 (avoids stdin collision)
• Only used if nothing better is available

5. Defensive Programming Throughout

• ${1:-0} - default value if no argument
• Consistent || : to prevent failures from propagating
• 2>/dev/null redirects suppress expected errors
• Explicit exit 0 after each method (clear control flow)
• All subshells properly cleaned up

6. Modern Shell Arithmetic

while (( SECONDS < secs )); do

Better than [[ ]] for arithmetic - native bash syntax, cleaner.

Minor Nits (We're Being Pedantic Now)

1. Input Validation Still Missing

secs=${1:-0}

# Should add:
if ! [[ "$secs" =~ ^[0-9]+(\.[0-9]+)?$ ]]; then
    echo "Error: Invalid duration: $secs" >&2
    exit 1
fi

Though in practice, read -t and other commands will fail gracefully with bad input.

2. The Ping Off-By-One

This is correct but deserves a comment:

# ping sends N packets with N-1 waits (1sec each), so +1 for desired delay
ping -c "$((secs + 1))" 127.0.0.1 >/dev/null 2>&1 || :

3. Potential FD Leak in Tail Method

If the script receives a signal between the exec 3< and exec 3<&-, FD3 might leak. In practice, this is unlikely to matter, but for paranoid scripting:

trap 'exec 3<&- 2>/dev/null' EXIT
exec 3< <(tail -f /dev/null)

4. Coproc Cleanup Could Be More Robust

# Current:
kill "$_slp_PID" 2>/dev/null || :
wait "$_slp_PID" 2>/dev/null || :

# Slightly better:
{ kill "$_slp_PID" && wait "$_slp_PID"; } 2>/dev/null || :

But this is really nitpicking - the current version is fine.

Systems-Level Assessment

CPU Efficiency Ranking

1. sleep - optimal (kernel timer, no wakeups)
2. coproc + read -t - excellent (kernel timer, one pipe)
3. tail + read -t - excellent (kernel timer, one pipe, one subprocess)
4. TTY + read -t - excellent (kernel timer)
5. ping - acceptable (network stack overhead, but sleeps between)
6. busy-wait - terrible (100% CPU)

Portability Coverage

• ✅ Any system with sleep (99.9%)
• ✅ Bash 4+ environments (most modern systems)
• ✅ POSIX-ish systems with tail
• ✅ Interactive shells with TTY
• ✅ Systems with ping
• ✅ Absolute fallback to busy-wait

This covers essentially every real-world scenario.

Real-World Applications

Where this shines:

• Container images (minimal base images, unpredictable utilities)
• CI/CD pipelines (varied execution environments)
• Embedded systems (limited userspace)
• Bootstrap scripts (before full system is available)
• Library code (can't assume caller's environment)

Verdict

Rating: 9.5/10

This is exemplary shell scripting. The author clearly understands:

• Bash internals (coprocesses, process substitution, FD management)
• Unix process management (pipes, signals, subprocesses)
• Kernel mechanisms (timers, TTYs, file descriptors)
• Portability constraints across environments

The coproc and tail -f /dev/null solutions show deep systems knowledge. I would:

• ✅ Approve immediately in code review
• ✅ Use as a teaching example
• ✅ Steal this for my own tooling

Only withholding 0.5 points for missing input validation, but honestly, that's me being pedantic. This is production-ready code written by someone who truly understands both shell scripting and systems programming.

Well done. 🎯