Rogeeandrogee
798ea43c2f
* HH-442: isolate runtime processes and harden shutdown * HH-442: harden worker shutdown races * HH-442: gate dependency shutdown on active handlers --------- Co-authored-by: Rogee <rogee@ipao.vip>
104 lines
3.3 KiB
Bash
Executable File
104 lines
3.3 KiB
Bash
Executable File
#!/bin/bash
|
|
# Exit codes: 0 healthy, 1 degraded, 2 unhealthy/invalid invocation.
|
|
|
|
set -uo pipefail
|
|
|
|
GOCHAT_HOST="${GOCHAT_HOST:-localhost}"
|
|
GOCHAT_PORT="${GOCHAT_PORT:-3000}"
|
|
GOCHAT_METRICS_PORT="${GOCHAT_METRICS_PORT:-${GOCHAT_PORT}}"
|
|
TIMEOUT="${TIMEOUT:-5}"
|
|
FULL_CHECK=false
|
|
JSON_OUTPUT=false
|
|
RESULTS=()
|
|
CRITICAL_FAIL=0
|
|
DEGRADED_FAIL=0
|
|
|
|
for arg in "$@"; do
|
|
[[ "${arg}" == "--json" ]] && JSON_OUTPUT=true
|
|
done
|
|
|
|
usage_error() {
|
|
if [[ "${JSON_OUTPUT}" == "true" ]]; then
|
|
printf '{"error":"%s","checks":[],"summary":{"critical_failures":1,"degraded_failures":0}}\n' "$1"
|
|
else
|
|
printf 'health_check: %s\n' "$1" >&2
|
|
fi
|
|
exit 2
|
|
}
|
|
|
|
while [[ $# -gt 0 ]]; do
|
|
case "$1" in
|
|
--full) FULL_CHECK=true; shift ;;
|
|
--json) JSON_OUTPUT=true; shift ;;
|
|
--timeout)
|
|
[[ $# -ge 2 && "$2" =~ ^[1-9][0-9]*$ ]] || usage_error "--timeout requires a positive integer"
|
|
TIMEOUT="$2"
|
|
shift 2
|
|
;;
|
|
--help)
|
|
if [[ "${JSON_OUTPUT}" == "true" ]]; then
|
|
printf '{"usage":"health_check.sh [--full] [--json] [--timeout SECONDS]"}\n'
|
|
else
|
|
printf 'Usage: %s [--full] [--json] [--timeout SECONDS]\n' "$0"
|
|
fi
|
|
exit 0
|
|
;;
|
|
*) usage_error "unknown option" ;;
|
|
esac
|
|
done
|
|
|
|
BASE_URL="http://${GOCHAT_HOST}:${GOCHAT_PORT}"
|
|
METRICS_URL="http://${GOCHAT_HOST}:${GOCHAT_METRICS_PORT}"
|
|
|
|
check_result() {
|
|
local name="$1" status="$2" detail="$3"
|
|
RESULTS+=("${name}|${status}|${detail}")
|
|
[[ "${status}" == "CRITICAL_FAIL" ]] && CRITICAL_FAIL=$((CRITICAL_FAIL + 1))
|
|
[[ "${status}" == "DEGRADED" ]] && DEGRADED_FAIL=$((DEGRADED_FAIL + 1))
|
|
}
|
|
|
|
check_url() {
|
|
local name="$1" severity="$2" url="$3" ok_detail="$4" fail_detail="$5"
|
|
if curl -fsS --max-time "${TIMEOUT}" -o /dev/null "${url}"; then
|
|
check_result "${name}" "OK" "${ok_detail}"
|
|
else
|
|
check_result "${name}" "${severity}" "${fail_detail}"
|
|
fi
|
|
}
|
|
|
|
check_url "liveness" "CRITICAL_FAIL" "${BASE_URL}/live" "process is alive" "liveness endpoint unreachable"
|
|
check_url "readiness" "CRITICAL_FAIL" "${BASE_URL}/ready" "dependencies are ready" "readiness endpoint failed"
|
|
|
|
if [[ "${FULL_CHECK}" == "true" ]]; then
|
|
check_url "database" "CRITICAL_FAIL" "${BASE_URL}/health?check=database" "database is healthy" "database check failed"
|
|
check_url "redis" "CRITICAL_FAIL" "${BASE_URL}/health?check=redis" "redis is healthy" "redis check failed"
|
|
else
|
|
check_result "database" "SKIPPED" "not in full mode"
|
|
check_result "redis" "SKIPPED" "not in full mode"
|
|
fi
|
|
|
|
check_url "metrics" "DEGRADED" "${METRICS_URL}/metrics" "metrics endpoint is responding" "metrics endpoint unreachable"
|
|
|
|
if [[ "${JSON_OUTPUT}" == "true" ]]; then
|
|
printf '{"checks":['
|
|
for i in "${!RESULTS[@]}"; do
|
|
IFS='|' read -r name status detail <<< "${RESULTS[$i]}"
|
|
[[ $i -gt 0 ]] && printf ','
|
|
printf '{"name":"%s","status":"%s","detail":"%s"}' "${name}" "${status}" "${detail}"
|
|
done
|
|
printf '],"summary":{"critical_failures":%d,"degraded_failures":%d}}\n' "${CRITICAL_FAIL}" "${DEGRADED_FAIL}"
|
|
else
|
|
printf '=== GoChat Health Check Report ===\n'
|
|
for result in "${RESULTS[@]}"; do
|
|
IFS='|' read -r name status detail <<< "${result}"
|
|
printf ' %s: %s — %s\n' "${name}" "${status}" "${detail}"
|
|
done
|
|
printf 'Critical failures: %d | Degraded: %d\n' "${CRITICAL_FAIL}" "${DEGRADED_FAIL}"
|
|
fi
|
|
|
|
if [[ ${CRITICAL_FAIL} -gt 0 ]]; then
|
|
exit 2
|
|
elif [[ ${DEGRADED_FAIL} -gt 0 ]]; then
|
|
exit 1
|
|
fi
|