Somehow duct negatively affects running the bash script (grep or pipe)

[yarikoptic]

trying to process s3 logs for dandi and thought to use duct, but it causes the goaccess to error out as if there were no input from grep. It is very odd

Here is without duct

dandi@drogon:/mnt/backup/dandi/dandiarchive-logs-goaccess-logs$ bash -x code/process_log ../dandiarchive-logs/2025/01/01.log
+ set -eu
+ logs_topdir=/mnt/backup/dandi/dandiarchive-logs
++ dirname code/process_log
+ topdir=code/..
+ force=
+ for in_log in "$@"
++ sed -E 's#.*/([0-9]{4}/[0-9]{2}(/[0-9]{2})?).*#\1#'
++ echo ../dandiarchive-logs/2025/01/01.log
+ log_date=2025/01/01
+ o_log=code/../2025/01/01.html
+ '[' -e code/../2025/01/01.html ']'
++ dirname code/../2025/01/01.html
+ mkdir -p code/../2025/01
+ find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f '!' -path '*/.git/*'
+ goaccess -o code/../2025/01/01.html --log-format AWSS3 --no-parsing-spinner --no-progress
+ xargs grep -v 'S3\.EXPIRE\.OBJECT'

and then if I remove that generated file and rerun -- kaboom

dandi@drogon:/mnt/backup/dandi/dandiarchive-logs-goaccess-logs$ rm 2025/01/01.html
dandi@drogon:/mnt/backup/dandi/dandiarchive-logs-goaccess-logs$ duct bash -x code/process_log ../dandiarchive-logs/2025/01/01.log
2025-03-28T07:41:53-0400 [INFO    ] con-duct: duct is executing 'bash -x code/process_log ../dandiarchive-logs/2025/01/01.log'...
2025-03-28T07:41:53-0400 [INFO    ] con-duct: Log files will be written to .duct/logs/2025.03.28T07.41.53-680321_
+ set -eu
+ logs_topdir=/mnt/backup/dandi/dandiarchive-logs
++ dirname code/process_log
+ topdir=code/..
+ force=
+ for in_log in "$@"
++ sed -E 's#.*/([0-9]{4}/[0-9]{2}(/[0-9]{2})?).*#\1#'
++ echo ../dandiarchive-logs/2025/01/01.log
+ log_date=2025/01/01
+ o_log=code/../2025/01/01.html
+ '[' -e code/../2025/01/01.html ']'
++ dirname code/../2025/01/01.html
+ mkdir -p code/../2025/01
+ goaccess -o code/../2025/01/01.html --log-format AWSS3 --no-parsing-spinner --no-progress

GoAccess - version 1.7 - Jan  4 2023 18:36:47
Config file: /etc/goaccess/goaccess.conf

Fatal error has occurred
Error occurred at: src/goaccess.c - initializer - 1471
No input data was provided nor there's data to restore.

+ xargs grep -v 'S3\.EXPIRE\.OBJECT'
+ find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f '!' -path '*/.git/*'
xargs: grep: terminated by signal 13
2025-03-28T07:41:53-0400 [INFO    ] con-duct: Summary:
Exit Code: 1
Command: bash -x code/process_log ../dandiarchive-logs/2025/01/01.log
Log files location: .duct/logs/2025.03.28T07.41.53-680321_
Wall Clock Time: 0.041 sec
Memory Peak Usage (RSS): 3.3 MB
Memory Average Usage (RSS): 3.3 MB
Virtual Memory Peak Usage (VSZ): 7.1 MB
Virtual Memory Average Usage (VSZ): 7.1 MB
Memory Peak Percentage: 0.0%
Memory Average Percentage: 0.0%
CPU Peak Usage: 0.0%
Average CPU Usage: 0.0%

happened originally with 0.8.0 and this one with bleeding edge 0.12.0.post8+g1650794 .

here is with debug output

dandi@drogon:/mnt/backup/dandi/dandiarchive-logs-goaccess-logs$ duct -l DEBUG bash -x code/process_log ../dandiarchive-logs/2025/01/01.log
2025-03-28T07:45:15-0400 [INFO    ] con-duct: duct is executing 'bash -x code/process_log ../dandiarchive-logs/2025/01/01.log'...
2025-03-28T07:45:15-0400 [INFO    ] con-duct: Log files will be written to .duct/logs/2025.03.28T07.45.15-680848_
2025-03-28T07:45:15-0400 [DEBUG   ] con-duct: Starting monitoring of the process <Popen: returncode: None args: ['bash', '-x', 'code/process_log', '../dandia...> on sample interval 1.000000 for report interval 60.000000
+ set -eu
+ logs_topdir=/mnt/backup/dandi/dandiarchive-logs
++ dirname code/process_log
+ topdir=code/..
+ force=
+ for in_log in "$@"
++ sed -E 's#.*/([0-9]{4}/[0-9]{2}(/[0-9]{2})?).*#\1#'
++ echo ../dandiarchive-logs/2025/01/01.log
+ log_date=2025/01/01
+ o_log=code/../2025/01/01.html
+ '[' -e code/../2025/01/01.html ']'
++ dirname code/../2025/01/01.html
+ mkdir -p code/../2025/01
+ find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f '!' -path '*/.git/*'
+ goaccess -o code/../2025/01/01.html --log-format AWSS3 --no-parsing-spinner --no-progress
+ xargs grep -v 'S3\.EXPIRE\.OBJECT'

GoAccess - version 1.7 - Jan  4 2023 18:36:47
Config file: /etc/goaccess/goaccess.conf

Fatal error has occurred
Error occurred at: src/goaccess.c - initializer - 1471
No input data was provided nor there's data to restore.

2025-03-28T07:45:15-0400 [DEBUG   ] con-duct: Process ended, setting stop_event to stop monitoring thread
2025-03-28T07:45:15-0400 [DEBUG   ] con-duct: Waiting for monitoring thread to finish
xargs: grep: terminated by signal 13
2025-03-28T07:45:15-0400 [DEBUG   ] con-duct: Breaking out because stop event was set
2025-03-28T07:45:15-0400 [DEBUG   ] con-duct: Monitoring thread finished
2025-03-28T07:45:15-0400 [DEBUG   ] con-duct: Waiting for environment collection thread to finish
2025-03-28T07:45:15-0400 [DEBUG   ] con-duct: Environment collection finished
2025-03-28T07:45:15-0400 [DEBUG   ] con-duct: Waiting for system information collection thread to finish
2025-03-28T07:45:15-0400 [DEBUG   ] con-duct: System information collection finished
2025-03-28T07:45:15-0400 [INFO    ] con-duct: Removing log files since command failed in less than 3.0 seconds.

seems to be specifically about `| xargs grep -v` component since if removed -- works fine

dandi@drogon:/mnt/backup/dandi/dandiarchive-logs-goaccess-logs$ bash -c "find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs grep -v 'S3\.EXPIRE\.OBJECT' | goaccess -o code/../2025/01/01.html --log-format AWSS3 --no-parsing-spinner --no-progress"
dandi@drogon:/mnt/backup/dandi/dandiarchive-logs-goaccess-logs$ ls -l 2025/01/01.html                                                                                                                                             -rw-r--r-- 1 dandi dandi 760070 Mar 28 07:48 2025/01/01.html
dandi@drogon:/mnt/backup/dandi/dandiarchive-logs-goaccess-logs$ duct bash -c "find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs grep -v 'S3\.EXPIRE\.OBJECT' | goaccess -o code/../2025/01/01.html --log-format AWSS3 --no-parsing-spinner --no-progress"
2025-03-28T07:48:21-0400 [INFO    ] con-duct: duct is executing "bash -c find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs grep -v 'S3\\.EXPIRE\\.OBJECT' | goaccess -o code/../2025/01/01.html --log-format AWSS3 --no-parsing-spinner --no-progress"...
2025-03-28T07:48:21-0400 [INFO    ] con-duct: Log files will be written to .duct/logs/2025.03.28T07.48.21-685844_

GoAccess - version 1.7 - Jan  4 2023 18:36:47
Config file: /etc/goaccess/goaccess.conf

Fatal error has occurred
Error occurred at: src/goaccess.c - initializer - 1471
No input data was provided nor there's data to restore.

xargs: grep: terminated by signal 13
2025-03-28T07:48:21-0400 [INFO    ] con-duct: Removing log files since command failed in less than 3.0 seconds.
dandi@drogon:/mnt/backup/dandi/dandiarchive-logs-goaccess-logs$ ls -l 2025/01/01.html                                                                                                                                             -rw-r--r-- 1 dandi dandi 760070 Mar 28 07:48 2025/01/01.html
dandi@drogon:/mnt/backup/dandi/dandiarchive-logs-goaccess-logs$ duct bash -c "find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs goaccess -o code/../2025/01/01.html --log-format AWSS3 --no-parsing-spinner --no-progress"
2025-03-28T07:48:45-0400 [INFO    ] con-duct: duct is executing "bash -c find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs goaccess -o code/../2025/01/01.html --log-format AWSS3 --no-parsing-spinner --no-progress"...
2025-03-28T07:48:45-0400 [INFO    ] con-duct: Log files will be written to .duct/logs/2025.03.28T07.48.45-685857_
2025-03-28T07:48:52-0400 [INFO    ] con-duct: Summary:
Exit Code: 0
Command: bash -c find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs goaccess -o code/../2025/01/01.html --log-format AWSS3 --no-parsing-spinner --no-progress
Log files location: .duct/logs/2025.03.28T07.48.45-685857_
Wall Clock Time: 6.820 sec
Memory Peak Usage (RSS): 27.2 MB
Memory Average Usage (RSS): 21.3 MB
Virtual Memory Peak Usage (VSZ): 46.3 MB
Virtual Memory Average Usage (VSZ): 39.6 MB
Memory Peak Percentage: 0.00%
Memory Average Percentage: 0.00%
CPU Peak Usage: 40.90%
Average CPU Usage: 29.51%

dandi@drogon:/mnt/backup/dandi/dandiarchive-logs-goaccess-logs$ ls -l 2025/01/01.html                                                                                                                                             -rw-r--r-- 1 dandi dandi 760111 Mar 28 07:48 2025/01/01.html
dandi@drogon:/mnt/backup/dandi/dandiarchive-logs-goaccess-logs$ rm 2025/01/01.html
dandi@drogon:/mnt/backup/dandi/dandiarchive-logs-goaccess-logs$ ls -l 2025/01/01.html
ls: cannot access '2025/01/01.html': No such file or directory
dandi@drogon:/mnt/backup/dandi/dandiarchive-logs-goaccess-logs$ duct bash -c "find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs goaccess -o code/../2025/01/01.html --log-format AWSS3 --no-parsing-spinner --no-progress"
2025-03-28T07:49:14-0400 [INFO    ] con-duct: duct is executing "bash -c find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs goaccess -o code/../2025/01/01.html --log-format AWSS3 --no-parsing-spinner --no-progress"...
2025-03-28T07:49:14-0400 [INFO    ] con-duct: Log files will be written to .duct/logs/2025.03.28T07.49.14-685878_
2025-03-28T07:49:18-0400 [INFO    ] con-duct: Summary:
Exit Code: 0
Command: bash -c find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs goaccess -o code/../2025/01/01.html --log-format AWSS3 --no-parsing-spinner --no-progress
Log files location: .duct/logs/2025.03.28T07.49.14-685878_
Wall Clock Time: 3.263 sec
Memory Peak Usage (RSS): 26.4 MB
Memory Average Usage (RSS): 19.0 MB
Virtual Memory Peak Usage (VSZ): 45.5 MB
Virtual Memory Average Usage (VSZ): 37.5 MB
Memory Peak Percentage: 0.00%
Memory Average Percentage: 0.00%
CPU Peak Usage: 51.80%
Average CPU Usage: 37.38%

[yarikoptic]

I would first check

• what is env vars are under duct -- are they the same as if without duct
• simulate session id assignment in a simpler bash or python wrapper

[asmacdo]

Heres a reproducer and a workaround for you.

I cant say that I fully understand whats going on, but with a lot of AI help, it seems to be a buffering issue. Perhaps the erroniously included filenames (without grep -h) are passed immediately to goaccess which is treated as 0 valid bytes and aborts before the rest of the data is piped. Without duct, (I dont understand why) the filenames and the data are passed at the same time, and the invalid data is filtered out?

tmp=/tmp/duct-261
rm -rf "$tmp" && mkdir -p "$tmp"/dandiarchive-logs/2025/01
cd "$tmp"/dandiarchive-logs/2025/01
set -o pipefail

# save a single valid AWSS3 line **in the file itself**
cat > 01.log <<'EOF'
79a4fe9d mybucket [01/Jan/2025:00:00:00 +0000] 192.0.2.1 - 3E57 REST.GET.OBJECT file.txt "GET /file.txt HTTP/1.1" 200 - 123 456 3 2 "-" "Mozilla/5.0" -
S3.EXPIRE.OBJECT this-line-should-be-filtered-out
EOF

mkdir -p "$tmp"/not-da
cd "$tmp"/not-da

# 1.  run the pipeline you showed in the GH issue – outside duct
bash -c "find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs grep -v 'S3\.EXPIRE\.OBJECT' | goaccess -o bash.html --log-format AWSS3 --no-parsing-spinner --no-progress"
echo "exit code outside duct = $?"

# 2.  same one-liner **inside duct** 
duct bash -c "find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs grep -v 'S3\.EXPIRE\.OBJECT'| goaccess -o duct.html --log-format AWSS3 --no-parsing-spinner --no-progress"
echo "exit code with duct using original reproducer = $?"

# 3.  fixed one-liner **inside duct** 
duct bash -c "find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs grep -v -h 'S3\.EXPIRE\.OBJECT'| goaccess -o duct-fixed.html --log-format AWSS3 --no-parsing-spinner --no-progress -"
echo "exit code with duct with adjusted reproducer = $?"
ls  # show the files that finished

The fixed one-liner has 2 changes (both required):

• -h to supress the filenames (which are not valid AWSS3)
• - which forces GoAccess to wait on stdin.

Heres the output

exit code outside duct = 0
2025-05-01T14:23:32-0500 [INFO    ] con-duct: duct is executing "bash -c find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs grep -v 'S3\\.EXPIRE\\.OBJECT'| goaccess -o duct.html --log-format AWSS3 --no-parsing-spinner --no-progress"...
2025-05-01T14:23:32-0500 [INFO    ] con-duct: Log files will be written to .duct/logs/2025.05.01T14.23.32-1156279_

GoAccess - version 1.9.4 - Apr  1 2025 00:00:00
Config file: /etc/goaccess/goaccess.conf

Fatal error has occurred
Error occurred at: src/goaccess.c - initializer - 1514
No input data was provided nor there's data to restore.

2025-05-01T14:23:32-0500 [INFO    ] con-duct: Removing log files since command failed in less than 3.0 seconds.
exit code with duct using original reproducer = 1
2025-05-01T14:23:32-0500 [INFO    ] con-duct: duct is executing "bash -c find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs grep -v -h 'S3\\.EXPIRE\\.OBJECT'| goaccess -o duct-fixed.html --log-format AWSS3 --no-parsing-spinner --no-progress -"...
2025-05-01T14:23:32-0500 [INFO    ] con-duct: Log files will be written to .duct/logs/2025.05.01T14.23.32-1156291_
2025-05-01T14:23:32-0500 [INFO    ] con-duct: Summary:
Exit Code: 0
Command: bash -c find ../dandiarchive-logs/2025/01/01.log -iname '[0-9][0-9]*' -type f -not -path '*/.git/*' | xargs grep -v -h 'S3\.EXPIRE\.OBJECT'| goaccess -o duct-fixed.html --log-format AWSS3 --no-parsing-spinner --no-progress -
Log files location: .duct/logs/2025.05.01T14.23.32-1156291_
Wall Clock Time: 0.005 sec
Memory Peak Usage (RSS): -
Memory Average Usage (RSS): -
Virtual Memory Peak Usage (VSZ): -
Virtual Memory Average Usage (VSZ): -
Memory Peak Percentage: -%
Memory Average Percentage: -%
CPU Peak Usage: -%
Average CPU Usage: -%

exit code with duct with adjusted reproducer = 0
bash.html  duct-fixed.html

So original bash works, original duct does not work, and duct-fixed works.

[asmacdo]

I've confirmed that the fundamental limitation with the current design is that duct is not guaranteed to preserve exact behavior with commands that expect a controlling TTY

This reproducer uses Popen directly, rather than duct.

Reproducer Script

#!/usr/bin/env python3
"""
POPEN-ONLY SESSION ISOLATION TEST
Shows how different Popen flags affect a piped GoAccess run.
"""

import subprocess, os

S3_LOG = (
    '79a4fe9d mybucket [01/Jan/2025:00:00:00 +0000] 192.0.2.1 - 3E57 REST.GET.OBJECT '
    'file.txt \\"GET /file.txt HTTP/1.1\\" 200 - 123 456 3 2 \\"-\\" \\"Mozilla/5.0\\" -'
)
cmd = f'echo "{S3_LOG}" | goaccess --log-format AWSS3 --no-parsing-spinner --no-progress -o test.html'

def run(label, **kwargs):
    proc = subprocess.Popen(['bash', '-c', cmd], stdout=subprocess.PIPE,
                            stderr=subprocess.PIPE, text=True, **kwargs)
    out, err = proc.communicate()
    print(f"{label:22} exit={proc.returncode} html={'YES' if subprocess.run(['test','-f','test.html']).returncode==0 else 'NO'}")
    if err: print(err.splitlines()[0])
    subprocess.run(['rm','-f','test.html'])

run("use current session")
run("start_new_session", start_new_session=True)
run("setpgrp", preexec_fn=os.setpgrp)
run("setsid",  preexec_fn=os.setsid)

print("=" *30)
print("Force goaccess to accept stdin")
print("=" *30)
print()
cmd = f'{cmd} -'
run("stdin, use current session")
run("stdin, start_new_session", start_new_session=True)
run("stdin, setpgrp", preexec_fn=os.setpgrp)
run("stdin, setsid",  preexec_fn=os.setsid)

Output:

use current session    exit=0 html=YES
start_new_session      exit=1 html=NO

setpgrp                exit=0 html=YES
setsid                 exit=1 html=NO

==============================
Force goaccess to accept stdin
==============================

stdin, use current session exit=0 html=YES
stdin, start_new_session exit=0 html=YES
stdin, setpgrp         exit=0 html=YES
stdin, setsid          exit=0 html=YES

As you can see in the reproducer, the problem exists whether we use start_new_session or setsid, because the new session is not a TTY.

Try with pty had no change

def run_with_pty(label, **kwargs):
    """Run with PTY allocation"""
    try:
        master, slave = pty.openpty()
        proc = subprocess.Popen(['bash', '-c', cmd],
                               stdin=slave,
                               stdout=subprocess.PIPE,
                               stderr=subprocess.PIPE,
                               text=True, **kwargs)
        os.close(slave)  # Close slave in parent
        os.close(master)  # Close master too

        out, err = proc.communicate()
        print(f"{label:22} exit={proc.returncode} html={'YES' if subprocess.run(['test','-f','test.html']).returncode==0 else 'NO'}")
        subprocess.run(['rm','-f','test.html'])
    except Exception as e:
        print(f"{label:22} ERROR: {e}")
        subprocess.run(['rm','-f','test.html'])

Standard tests
==============================
use current session    exit=0 html=YES
start_new_session      exit=1 html=NO

setpgrp                exit=0 html=YES
setsid                 exit=1 html=NO

pty + current session  exit=0 html=YES
pty + start_new_session exit=1 html=NO
pty + setpgrp          exit=0 html=YES
pty + setsid           exit=1 html=NO

Force goaccess to accept stdin
==============================
stdin, use current session exit=0 html=YES
stdin, start_new_session exit=0 html=YES
stdin, setpgrp         exit=0 html=YES
stdin, setsid          exit=0 html=YES

Options, (all need additional research):

1. keep current behavior, clearly document limitation of using programs that implicitly expect a controlling TTY
2. use pgrps rather than sessions, needs research to determine limitations of that approach
3. use cgroups or systemd, this would add complexity and also limit the systems we can run on

[asmacdo]

Take-away: like cron, any command that requires a controlling TTY will fail under duct’s start_new_session=True. GoAccess is a concrete example.

Heres an even simpler reproducer

#!/usr/bin/sh

# S3 log line for testing
S3_LOG='79a4fe9d mybucket [01/Jan/2025:00:00:00 +0000] 192.0.2.1 - 3E57 REST.GET.OBJECT file.txt "GET /file.txt HTTP/1.1" 200 - 123 456 3 2 "-" "Mozilla/5.0" -'

echo "=== Test 1: With TTY  === "
echo "$S3_LOG" | goaccess --log-format AWSS3 --no-parsing-spinner --no-progress -o test_tty.html
echo "Exit code: $?"
echo "HTML created: $([ -f test_tty.html ] && echo YES || echo NO)"
rm -f test_tty.html

echo "=== Test 2: Without TTY (setsid) ==="
setsid bash -c "echo '$S3_LOG' | goaccess --log-format AWSS3 --no-parsing-spinner --no-progress -o test_notty.html" < /dev/null > /dev/null 2>&1
echo "Exit code: $?"
echo "HTML created: $([ -f test_notty.html ] && echo YES || echo NO)"
rm -f test_notty.html

echo "=== Test 3: Without TTY but with - flag ==="
setsid bash -c "echo '$S3_LOG' | goaccess --log-format AWSS3 --no-parsing-spinner --no-progress -o test_stdin.html -" < /dev/null > /dev/null 2>&1
echo "Exit code: $?"
echo "HTML created: $([ -f test_stdin.html ] && echo YES || echo NO)"
rm -f test_stdin.html

Output:

=== Test 1: With TTY ===
Exit code: 0
HTML created: YES
=== Test 2: Without TTY (setsid) ===
Exit code: 1
HTML created: NO
=== Test 3: Without TTY but with - flag ===
Exit code: 0
HTML created: YES

When setsid is called, the process is detatched from the controlling TTY(POSIX rule), and this breaks goaccess, which believes it is in interactive mode (and expects TTY) unless the - is specified. See this comment

FWIW, the **same issue** exists if execute that command as a cronjob.

33 11 * * * /home/austin/devel/duct/asmacdo/claude-bf-261/cron_test.sh

#!/bin/bash

# S3 log line for testing
S3_LOG='79a4fe9d mybucket [01/Jan/2025:00:00:00 +0000] 192.0.2.1 - 3E57 REST.GET.OBJECT file.txt "GET /file.txt HTTP/1.1" 200 - 123 456 3 2 "-" "Mozilla/5.0" -'

cd /home/austin/devel/duct/asmacdo/claude-bf-261

# Test without hyphen
echo "$S3_LOG" | goaccess --log-format AWSS3 --no-parsing-spinner --no-progress -o /home/austin/devel/duct/asmacdo/claude-bf-261/cron_no_hyphen.html
echo "No hyphen exit: $?" > /home/austin/devel/duct/asmacdo/claude-bf-261/cron_results.txt
echo "No hyphen HTML: $([ -f /home/austin/devel/duct/asmacdo/claude-bf-261/cron_no_hyphen.html ] && echo YES || echo NO)" >> /home/austin/devel/duct/asmacdo/claude-bf-261/cron_results.txt

# Test with hyphen  
echo "$S3_LOG" | goaccess --log-format AWSS3 --no-parsing-spinner --no-progress -o /home/austin/devel/duct/asmacdo/claude-bf-261/cron_with_hyphen.html -
echo "With hyphen exit: $?" >> /home/austin/devel/duct/asmacdo/claude-bf-261/cron_results.txt
echo "With hyphen HTML: $([ -f /home/austin/devel/duct/asmacdo/claude-bf-261/cron_with_hyphen.html ] && echo YES || echo NO)" >> /home/austin/devel/duct/asmacdo/claude-bf-261/cron_results.txt

Output

cat cron_results.txt

No hyphen exit: 1
No hyphen HTML: NO
With hyphen exit: 0
With hyphen HTML: YES

I think I've demonstrated that this is a goaccess quirk, and not a duct transparency issue. However, it might still be worth trying out using pgroups instead of sessions (though I imagine a different set of quirks might eventually turn up?)

In the meantime I think we should do 2 things.

1. Document that duct commands are run without access to TTY, so inner commands should not require it.
2. Add a troubleshooting section that shows how to use setsid like the reproducer above, so users can develop their commands outside of duct and expect the same results.