fix(runtime): keep supervision and absence retries in container

Co-authored-by: GPT-6 <gpt-6@openai.com>
This commit is contained in:
ZeusinaandGPT-6 Bot committed 2026-10-11 09:01:52 +03:00
1 parent 226d93e2aa
commit 3f9fd2f36f
21 files changed
+1056 -137

No files matched your search

+100 -21
View File
@@ -30,54 +30,97 @@ server_pid=
local_pid=
monitor_pid=
zapret_started=0
cooldown_pid=
# Signal only descendants of an owned child, never the container/host namespace.
# The monitor gets TERM itself first so it can join scans and restore bytes;
# descendant signalling is reserved for SS shutdown and forced watchdog cleanup.
. /opt/managed-target-supervision.sh
pause() {
sleep "$1" </dev/null >/dev/null 2>&1 & cooldown_pid=$!
wait "$cooldown_pid"
cooldown_pid=
}
# Wait with a watchdog: never let an unresponsive child exceed Docker's grace.
bounded_wait() {
child=$1
seconds=$2
(
sleep "$seconds" & sleeper=$!
trap 'kill "$sleeper" 2>/dev/null || true; exit 0' TERM INT
wait "$sleeper"
kill -KILL "$child" 2>/dev/null || true
) </dev/null >/dev/null 2>&1 &
rm -f "$config_dir/join-timeout" "$config_dir/join-ready"
# Fresh interpreter: a forked shell can inherit a deferred EXIT/TERM trap
# context and never run its cancellation handler during PID 1 shutdown.
/bin/sh /opt/managed-target-supervision.sh --watchdog "$child" "$seconds" "$config_dir" </dev/null >/dev/null 2>&1 &
watchdog=$!
# A fast-exiting child must not cancel the watchdog before its trap exists.
# Fail closed if exec/setup fails before acknowledgement; never wait forever.
ack_deadline=$(($(date +%s) + 5))
while [ ! -e "$config_dir/join-ready" ]; do
if ! kill -0 "$watchdog" 2>/dev/null || [ "$(date +%s)" -ge "$ack_deadline" ]; then
: > "$config_dir/join-timeout"
signal_tree "$watchdog" KILL
signal_tree "$child" KILL
break
fi
done
waited_code=0
wait "$child" 2>/dev/null || waited_code=$?
kill -TERM "$watchdog" 2>/dev/null || true
wait "$watchdog" 2>/dev/null || true
join_timed_out=0
[ ! -e "$config_dir/join-timeout" ] || join_timed_out=1
}
cleanup() {
final_code=$?
case "$final_code" in 0|64|75|78) ;; *) final_code=78 ;; esac
trap - EXIT INT TERM
cleanup_cycle() {
cleanup_code=0
# Let the monitor cancel/join scans and restore committed bytes BEFORE stop.
if [ -n "$monitor_pid" ]; then
kill -TERM "$monitor_pid" 2>/dev/null || true
bounded_wait "$monitor_pid" 60
bounded_wait "$monitor_pid" 45
# Cancellation must restore safely even when SS failure requested retry.
case "$waited_code" in 0|143) ;; *) final_code=78 ;; esac
case "$waited_code" in 0|143) ;; *) echo "error: monitor cleanup status $waited_code" >&2; cleanup_code=78 ;; esac
[ "$join_timed_out" -eq 0 ] || { echo 'error: monitor join timeout' >&2; cleanup_code=78; }
monitor_pid=
fi
# Dedicated PID/network namespace: no host/other-target stop.
if [ "$zapret_started" -eq 1 ]; then
timeout -k 5 30 /opt/zapret2/init.d/sysv/zapret2 stop || final_code=78
# Stop also sources config: clear trust before reconciliation, never
# source unknown bytes left by a crashed monitor, even for shutdown.
zapret_started=0
if timeout -k 5 45 /usr/local/bin/zapret-autotune managed reconcile; then
timeout -k 5 30 /opt/zapret2/init.d/sysv/zapret2 stop || cleanup_code=78
else
cleanup_code=78
fi
fi
for pid in "$local_pid" "$server_pid"; do
[ -z "$pid" ] || kill "$pid" 2>/dev/null || true
[ -z "$pid" ] || signal_tree "$pid" TERM
done
for pid in "$local_pid" "$server_pid"; do
if [ -n "$pid" ]; then
bounded_wait "$pid" 5
# Its original failure triggered supervision; only forced shutdown
# is a cleanup failure, not the already-dead Shadowsocks status.
[ "$waited_code" -ne 137 ] || final_code=78
[ "$join_timed_out" -eq 0 ] || { echo 'error: Shadowsocks join timeout' >&2; cleanup_code=78; }
fi
done
server_pid=
local_pid=
return "$cleanup_code"
}
cleanup() {
final_code=$?
trap - EXIT
# A caught (not ignored) signal lets watchdog children install TERM traps.
trap ':' INT TERM
if [ -n "$cooldown_pid" ]; then
kill "$cooldown_pid" 2>/dev/null || true
wait "$cooldown_pid" 2>/dev/null || true
fi
cleanup_cycle || final_code=78
exit "$final_code"
}
trap cleanup EXIT
trap 'exit 0' INT TERM
# Never invoke /opt/entrypoint.sh: zapret must start AFTER reconciliation.
run_cycle() {
ss-server -c "$config_dir/server.json" &
server_pid=$!
ss-local -c "$config_dir/local.json" &
@@ -86,10 +129,10 @@ local_pid=$!
# Bounded readiness, no network probe. Managed commands verify socket ownership.
attempt=0
until awk '$2 == "00000000:0438" && $4 == "0A" { found=1 } END { exit !found }' /proc/net/tcp; do
{ kill -0 "$server_pid" && kill -0 "$local_pid"; } || exit 75
{ kill -0 "$server_pid" && kill -0 "$local_pid"; } || return 75
attempt=$((attempt + 1))
[ "$attempt" -lt 30 ] || { echo 'error: target SOCKS listener timeout' >&2; exit 75; }
sleep 1
[ "$attempt" -lt 30 ] || { echo 'error: target SOCKS listener timeout' >&2; return 75; }
pause 1
done
if [ ! -e "$ZAPRET_CONFIG" ]; then
timeout -k 5 45 /usr/local/bin/zapret-autotune managed init || exit 78
@@ -100,10 +143,14 @@ timeout -k 5 45 /usr/local/bin/zapret-autotune managed reconcile || exit 78
zapret_started=1
timeout -k 5 30 /opt/zapret2/init.d/sysv/zapret2 start || exit 78
# A replacement SS child can fail during reconciliation/activation. Do not
# fork a monitor we already need to cancel: an early TERM can race interpreter
# startup before its cancellation handler exists. Clean this cycle directly.
{ kill -0 "$server_pid" 2>/dev/null && kill -0 "$local_pid" 2>/dev/null; } || return 75
/usr/local/bin/zapret-autotune managed monitor &
monitor_pid=$!
while kill -0 "$server_pid" && kill -0 "$local_pid" && kill -0 "$monitor_pid"; do
sleep 1
pause 1
done
if ! kill -0 "$monitor_pid" 2>/dev/null; then
code=0
@@ -115,7 +162,39 @@ if ! kill -0 "$monitor_pid" 2>/dev/null; then
64|75|78) ;;
*) code=78 ;;
esac
exit "$code"
return "$code"
fi
echo 'error: target child process exited' >&2
exit 75
return 75
}
# Three starts per rolling six hours, including initial startup. PID 1 owns
# the entire runtime cycle; cooldown is cancellable without any external manager.
starts=
backoff=15
while :; do
now=$(date +%s)
recent=
count=0
for started in $starts; do
if [ "$((now - started))" -lt 21600 ]; then
recent="$recent $started"
count=$((count + 1))
fi
done
[ "$count" -lt 3 ] || { echo 'error: target restart budget exhausted' >&2; exit 78; }
starts="$recent $now"
code=0
run_cycle || code=$?
cleanup_cycle || code=78
[ "$code" -eq 75 ] || exit "$code"
# A long-running monitor may outlive the window; recheck at failure time.
now=$(date +%s)
retry_count=0
for started in $starts; do
[ "$((now - started))" -ge 21600 ] || retry_count=$((retry_count + 1))
done
[ "$retry_count" -lt 3 ] || { echo 'error: target restart budget exhausted' >&2; exit 78; }
pause "$backoff"
[ "$backoff" -ge 60 ] || backoff=$((backoff * 2))
done