Skip to content

fix: stale backend holding port blocks restart (reap own unhealthy listener; Windows netstat/taskkill fallback) #116

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
smoochy wants to merge 2 commits into
base: main
Choose a base branch
from
+66 −14
Open

fix: stale backend holding port blocks restart (reap own unhealthy listener; Windows netstat/taskkill fallback) #116

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7f39cff
fix: backend-service.sh reap stale own-backend + Windows lsof fallback
smoochy Jul 6, 2026
2a590d6
@
smoochy Jul 6, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
80 changes: 66 additions & 14 deletions plugin/scripts/backend-service.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -192,6 +192,27 @@ port_occupied() {
(echo >"/dev/tcp/127.0.0.1/$PORT") 2>/dev/null
}

# List PIDs listening on $1 without lsof, for Windows git-bash where lsof is
# absent. Parses `netstat -ano` (the Windows netstat, on PATH under git-bash):
# columns are Proto, Local, Foreign, State, PID; we want LISTENING rows whose
# local address ends in :$port. Empty if netstat is unavailable or nothing
# listens.
port_pids_netstat() {
command -v netstat >/dev/null 2>&1 || return 0
netstat -ano 2>/dev/null | awk -v p=":$1\$" '
$1 ~ /^TCP/ && $2 ~ p && $4 == "LISTENING" { print $5 }
' | sort -u
Comment on lines +202 to +204

@coderabbitai coderabbitai Bot Jul 6, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🩺 Stability & Availability | 🟠 Major | ⚡ Quick win

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
set -euo pipefail

printf '%s\n' "== plugin/scripts/backend-service.sh (around lines 195-210) =="
sed -n '190,215p' plugin/scripts/backend-service.sh | cat -n

printf '\n%s\n' "== Where port_pids_netstat is used =="
rg -n "port_pids_netstat|taskkill|tasklist|reap_port_listeners|port_holder|pid_cmdline_win" plugin/scripts/backend-service.sh plugin/scripts/_lib.sh

printf '\n%s\n' "== Relevant helper docs =="
sed -n '1,220p' plugin/scripts/_lib.sh | cat -n | sed -n '1,220p'

Repository: ReflexioAI/claude-smart

Length of output: 12950

🏁 Script executed:

#!/bin/bash
set -euo pipefail

sed -n '220,380p' plugin/scripts/backend-service.sh | cat -n

printf '\n%s\n' "== Simulate netstat CRLF parsing with awk =="
python3 - <<'PY'
import subprocess, textwrap, os, json, sys, tempfile

sample = textwrap.dedent("""\
TCP    127.0.0.1:8080    0.0.0.0:0    LISTENING    1234\r
TCP    127.0.0.1:8080    0.0.0.0:0    LISTENING    5678\r
""")

script = r'''awk -v p=":8080$" '
  $1 ~ /^TCP/ && $2 ~ p && $4 == "LISTENING" { print $5 }
' | sort -u'''
proc = subprocess.run(
    ["bash", "-lc", script],
    input=sample.encode(),
    stdout=subprocess.PIPE,
    stderr=subprocess.PIPE,
)
print("stdout repr:", repr(proc.stdout.decode()))
print("stderr repr:", repr(proc.stderr.decode()))
PY

Repository: ReflexioAI/claude-smart

Length of output: 7778

Strip \r from Windows netstat PIDs.
netstat -ano on Git Bash leaves CRLF on $5, so the PID becomes 1234\r and won’t match tasklist, ps, or taskkill. That breaks stale-listener cleanup on Windows.

Proposed fix 
-    $1 ~ /^TCP/ && $2 ~ p && $4 == "LISTENING" { print $5 }
+    $1 ~ /^TCP/ && $2 ~ p && $4 == "LISTENING" { gsub(/\r$/, "", $5); print $5 }
📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change
netstat -ano 2>/dev/null | awk -v p=":$1\$" '
$1 ~ /^TCP/ && $2 ~ p && $4 == "LISTENING" { print $5 }
' | sort -u
netstat -ano 2>/dev/null | awk -v p=":$1\$" '
$1 ~ /^TCP/ && $2 ~ p && $4 == "LISTENING" { gsub(/\r$/, "", $5); print $5 }
' | sort -u
🤖 Prompt for AI Agents 
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@plugin/scripts/backend-service.sh` around lines 202 - 204, The PID extraction
in the netstat pipeline still carries Windows CRLF, which leaves a trailing
carriage return on the value returned from the listener lookup. Update the
parsing in the backend-service.sh netstat/awk flow so the emitted PID is
normalized by stripping any trailing \r before it is used by the stale-listener
cleanup logic, ensuring the values match downstream calls like tasklist, ps, and
taskkill.

}

# Command line for a Windows PID via tasklist, for cmdline pattern matching
# when ps/lsof are unavailable. Prints the image name (best available proxy
# for the command on Windows) or empty.
pid_cmdline_win() {
command -v tasklist >/dev/null 2>&1 || return 0
tasklist /FI "PID eq $1" /FO CSV /NH 2>/dev/null \
| awk -F'","' 'NR==1 { gsub(/"/,"",$1); print $1 }'
}
Comment thread
coderabbitai[bot] marked this conversation as resolved.

# Reap listeners still holding the given port after the PID file kill
# when their command line matches one of the supplied patterns. Filters
# by cmdline so we don't knock over an
Expand All @@ -201,12 +222,23 @@ reap_port_listeners() {
port="${1:-$PORT}"
shift || true
[ "$#" -eq 0 ] && return 0
command -v lsof >/dev/null 2>&1 || return 0
candidates=$(lsof -ti:"$port" 2>/dev/null) || candidates=""
if command -v lsof >/dev/null 2>&1; then
candidates=$(lsof -ti:"$port" 2>/dev/null) || candidates=""
else
candidates=$(port_pids_netstat "$port")
fi
[ -z "$candidates" ] && return 0
ours=""
for pid in $candidates; do
cmdline=$(ps -p "$pid" -o command= 2>/dev/null || true)
cmdline=""
if command -v ps >/dev/null 2>&1; then
cmdline=$(ps -p "$pid" -o command= 2>/dev/null || true)
fi
# Git Bash ships `ps` but it may not support `-o command=` or resolve a
# native PID from netstat, leaving cmdline empty — fall back to tasklist.
if [ -z "$cmdline" ]; then
cmdline=$(pid_cmdline_win "$pid")
fi
Comment thread
coderabbitai[bot] marked this conversation as resolved.
for pattern in "$@"; do
if [[ "$cmdline" == $pattern ]]; then
ours="$ours $pid"
Expand All @@ -215,6 +247,11 @@ reap_port_listeners() {
done
done
[ -z "$ours" ] && return 0
if command -v taskkill >/dev/null 2>&1 && ! command -v lsof >/dev/null 2>&1; then
# Windows: no POSIX signals to these PIDs; force-kill directly.
for pid in $ours; do taskkill /F /PID "$pid" >/dev/null 2>&1 || true; done
Comment on lines +250 to +252

@coderabbitai coderabbitai Bot Jul 6, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🩺 Stability & Availability | 🟠 Major | ⚡ Quick win

Use the shared Git Bash-safe kill helper.

This direct taskkill /F /PID path bypasses the existing claude_smart_kill_tree handling for Git Bash slash escaping and child-process termination. Reuse the helper instead.

Proposed fix 
   if command -v taskkill >/dev/null 2>&1 && ! command -v lsof >/dev/null 2>&1; then
-    # Windows: no POSIX signals to these PIDs; force-kill directly.
-    for pid in $ours; do taskkill /F /PID "$pid" >/dev/null 2>&1 || true; done
+    # Windows: netstat returns native PIDs; shared helper handles taskkill safely.
+    for pid in $ours; do claude_smart_kill_tree "$pid"; done
     return 0
   fi
📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change
if command -v taskkill >/dev/null 2>&1 && ! command -v lsof >/dev/null 2>&1; then
# Windows: no POSIX signals to these PIDs; force-kill directly.
for pid in $ours; do taskkill /F /PID "$pid" >/dev/null 2>&1 || true; done
if command -v taskkill >/dev/null 2>&1 && ! command -v lsof >/dev/null 2>&1; then
# Windows: netstat returns native PIDs; shared helper handles taskkill safely.
for pid in $ours; do claude_smart_kill_tree "$pid"; done
return 0
fi
🤖 Prompt for AI Agents 
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@plugin/scripts/backend-service.sh` around lines 246 - 248, The Windows-only
direct taskkill path in backend-service.sh should not bypass the existing
kill-tree logic. Update the branch in the shutdown/cleanup flow to reuse
claude_smart_kill_tree for each PID in ours so Git Bash slash escaping and
child-process termination are handled consistently, instead of calling taskkill
/F /PID directly in the loop.

return 0
fi
# shellcheck disable=SC2086
kill -TERM $ours 2>/dev/null || true
sleep 1
Expand All @@ -231,9 +268,14 @@ reap_port_listeners() {
# "<command> (pid <pid>)" or empty if the port is free or lsof is
# unavailable. Used to make port-conflict log lines diagnosable.
port_holder() {
command -v lsof >/dev/null 2>&1 || return 0
lsof -i:"$1" -sTCP:LISTEN -P -n 2>/dev/null \
| awk 'NR==2 {print $1" (pid "$2")"; exit}'
if command -v lsof >/dev/null 2>&1; then
lsof -i:"$1" -sTCP:LISTEN -P -n 2>/dev/null \
| awk 'NR==2 {print $1" (pid "$2")"; exit}'
return 0
fi
pid=$(port_pids_netstat "$1" | head -n1)
[ -z "$pid" ] && return 0
printf '%s (pid %s)' "$(pid_cmdline_win "$pid")" "$pid"
}

ensure_vendored_reflexio_active() {
Expand Down Expand Up @@ -304,15 +346,25 @@ case "$CMD" in
emit_ok; exit 0
fi
if port_occupied; then
# is_our_backend_running already ruled out a healthy Reflexio backend.
# Anything still holding the port would make uvicorn fail to bind, so
# surface the holder instead of silently starting into a collision.
# is_our_backend_running already ruled out a healthy Reflexio backend, so
# the holder is either our own wedged/unhealthy backend (reap + retry) or
# a foreign process (surface and skip, never stomp it).
holder="$(port_holder "$PORT" 2>/dev/null || true)"
msg="[claude-smart] backend: port $PORT held by another process${holder:+ ($holder)}; skipping start"
claude_smart_append_capped_log "$LOG_FILE" "$LOG_MAX_BYTES" "$msg"
echo "$msg" >&2
echo "Free port $PORT (or stop the process above) and run /claude-smart:restart again." >&2
emit_ok; exit 0
case "$holder" in
*reflexio*|*uvicorn*|*python*)
claude_smart_append_capped_log "$LOG_FILE" "$LOG_MAX_BYTES" \
"[claude-smart] backend: stale unhealthy holder on port $PORT (${holder:-unknown}); reaping and retrying"
reap_port_listeners "$PORT" '*reflexio*' '*uvicorn*' '*python*'
;;
esac
if port_occupied; then
holder="$(port_holder "$PORT" 2>/dev/null || true)"
msg="[claude-smart] backend: port $PORT held by another process${holder:+ ($holder)}; skipping start"
claude_smart_append_capped_log "$LOG_FILE" "$LOG_MAX_BYTES" "$msg"
echo "$msg" >&2
echo "Free port $PORT (or stop the process above) and run /claude-smart:restart again." >&2
emit_ok; exit 0
fi
fi
if ! command -v uv >/dev/null 2>&1; then
if [ "${CLAUDE_SMART_BOOTSTRAPPING:-}" != "1" ] && [ -x "$PLUGIN_ROOT/scripts/smart-install.sh" ]; then
Expand Down