Your IP : 216.73.216.209

Current Path : /usr/local/bin/
Upload File :
Current File : //usr/local/bin/apache-graceful-with-cleanup

#!/bin/bash
#
# Graceful Apache reload with forced cleanup of stuck workers.
#
# Apache has no timeout on graceful-restart: if a worker is blocked in a
# mid-request syscall (classically, reading from a mod_fcgid Unix socket
# whose PHP backend has hung), it has SIGUSR1 blocked and never exits.
# Over time these leak scoreboard slots until the instance fails with
# "AH03490 scoreboard is full, not at MaxRequestWorkers".
#
# This wrapper snapshots the pre-reload worker PIDs, runs the graceful,
# polls up to GRACE seconds for them to drain on their own, and SIGKILLs
# any that are still alive and still parented to the same master.
#
# Deploy via a systemd drop-in:
#   /etc/systemd/system/apache2@.service.d/10-reload-cleanup.conf
#     [Service]
#     ExecReload=
#     ExecReload=/usr/local/sbin/apache-graceful-with-cleanup %i
#     TimeoutStartSec=120
#
# Usage: apache-graceful-with-cleanup <instance>
# Env:   APACHE_RELOAD_GRACE  seconds to wait before SIGKILL (default 60)

set -eo pipefail

INSTANCE="${1:?instance name required}"
GRACE="${APACHE_RELOAD_GRACE:-60}"
UNIT="apache2@${INSTANCE}"
CONF="/etc/apache2-${INSTANCE}/apache2.conf"

log() { logger -t apache-reload-cleanup "${INSTANCE}: $*"; }

# Find the master PID. Prefer systemd's MainPID, but fall back to process
# introspection if the unit's PID tracking is misconfigured (common on
# Type=forking units with no PIDFile=). The master is the oldest apache2
# process whose cmdline names this instance's config dir.
find_master_pid() {
    local from_systemd
    from_systemd=$(systemctl show -p MainPID --value "${UNIT}" 2>/dev/null)
    if [[ -n "${from_systemd}" && "${from_systemd}" != "0" ]]; then
        echo "${from_systemd}"
        return
    fi
    pgrep -o -f "apache2 -d /etc/apache2-${INSTANCE}(/| |$)" || true
}

MAIN=$(find_master_pid)
if [[ -z "${MAIN}" ]]; then
    log "master not found, nothing to do"
    exit 0
fi

OLD=$(pgrep -P "${MAIN}" | tr '\n' ' ' | sed 's/ $//')
log "reload master=${MAIN} grace=${GRACE}s snapshot=[${OLD}]"
echo "reload master=${MAIN} grace=${GRACE}s snapshot=[${OLD}]"

/usr/sbin/apachectl -f "${CONF}" -k graceful

if [[ -z "${OLD}" ]]; then
    log "no workers to track, done"
    exit 0
fi

# 'Still a worker' = /proc/<pid> exists AND its parent is still our
# snapshotted master.
still_alive() {
    local out=''
    for p in ${OLD}; do
        [[ -d /proc/${p} ]] || continue
        local ppid
        ppid=$(awk '{print $4}' /proc/${p}/stat 2>/dev/null || echo 0)
        if [[ "${ppid}" == "${MAIN}" ]]; then
            out+="${p} "
        fi
    done
    echo "${out% }"
}

elapsed=0
echo "waiting for old workers to die"
while (( elapsed < GRACE )); do
    echo $elapsed
    survivors=$(still_alive)
    if [[ -z "${survivors}" ]]; then
        log "all workers drained in ${elapsed}s, no kill needed"
        exit 0
    fi
    sleep 1
    elapsed=$((elapsed + 1))
done

# Grace elapsed — confirm master is still our snapshot (no intervening
# full restart) before SIGKILL'ing.
if [[ ! -d /proc/${MAIN} ]] || \
   ! grep -qa "apache2-${INSTANCE}" /proc/${MAIN}/cmdline 2>/dev/null; then
    log "skip kill: master ${MAIN} gone or replaced"
    exit 0
fi

survivors=$(still_alive)
if [[ -z "${survivors}" ]]; then
    log "all workers drained at boundary, no kill needed"
    echo "all workers drained at boundary"
    exit 0
fi

killed=''
for p in ${survivors}; do
    if kill -9 "${p}" 2>/dev/null; then
        echo "killing ${p}"
        killed+="${p} "
    fi
done
log "SIGKILL after ${GRACE}s: [${killed% }]"