Files
gochat/scripts/health_check.sh
T
2026-06-04 15:44:48 +08:00

175 lines
5.1 KiB
Bash

#!/bin/bash
# GoChat Health Check Script
# Reference: Chatwoot health_check endpoint pattern
# Checks: HTTP server, database, Redis, worker queues, WebSocket hub
#
# Usage:
# ./scripts/health_check.sh [--full] [--json] [--timeout SECONDS]
#
# Exit codes:
# 0 = healthy
# 1 = degraded (some checks failed but core is OK)
# 2 = unhealthy (critical checks failed)
set -euo pipefail
# Configuration
GOCHAT_HOST="${GOCHAT_HOST:-localhost}"
GOCHAT_PORT="${GOCHAT_PORT:-3000}"
GOCHAT_METRICS_PORT="${GOCHAT_METRICS_PORT:-9090}"
TIMEOUT="${TIMEOUT:-5}"
FULL_CHECK=false
JSON_OUTPUT=false
# Parse arguments
while [[ $# -gt 0 ]]; do
case "$1" in
--full) FULL_CHECK=true; shift ;;
--json) JSON_OUTPUT=true; shift ;;
--timeout) TIMEOUT="$2"; shift 2 ;;
--help) echo "Usage: $0 [--full] [--json] [--timeout SECONDS]"; exit 0 ;;
*) echo "Unknown option: $1"; exit 1 ;;
esac
done
BASE_URL="http://${GOCHAT_HOST}:${GOCHAT_PORT}"
METRICS_URL="http://${GOCHAT_HOST}:${GOCHAT_METRICS_PORT}"
# Result tracking
RESULTS=()
CRITICAL_FAIL=0
DEGRADED_FAIL=0
check_result() {
local name="$1" status="$2" detail="$3"
RESULTS+=("${name}|${status}|${detail}")
if [[ "$status" == "CRITICAL_FAIL" ]]; then
CRITICAL_FAIL=$((CRITICAL_FAIL + 1))
elif [[ "$status" == "DEGRADED" ]]; then
DEGRADED_FAIL=$((DEGRADED_FAIL + 1))
fi
}
# ---- Check 1: HTTP liveness endpoint ----
check_liveness() {
local response
response=$(curl -sf --max-time "${TIMEOUT}" "${BASE_URL}/live" 2>&1) && {
check_result "liveness" "OK" "${response}"
} || {
check_result "liveness" "CRITICAL_FAIL" "HTTP /live endpoint unreachable"
}
}
# ---- Check 2: HTTP readiness endpoint ----
check_readiness() {
local response
response=$(curl -sf --max-time "${TIMEOUT}" "${BASE_URL}/ready" 2>&1) && {
check_result "readiness" "OK" "${response}"
} || {
check_result "readiness" "CRITICAL_FAIL" "HTTP /ready endpoint unreachable"
}
}
# ---- Check 3: Database connectivity ----
check_database() {
if [[ "${FULL_CHECK}" == "true" ]]; then
local response
response=$(curl -sf --max-time "${TIMEOUT}" "${BASE_URL}/health?check=database" 2>&1) && {
check_result "database" "OK" "${response}"
} || {
check_result "database" "CRITICAL_FAIL" "Database connectivity failed"
}
else
check_result "database" "SKIPPED" "Not in full mode"
fi
}
# ---- Check 4: Redis connectivity ----
check_redis() {
if [[ "${FULL_CHECK}" == "true" ]]; then
local response
response=$(curl -sf --max-time "${TIMEOUT}" "${BASE_URL}/health?check=redis" 2>&1) && {
check_result "redis" "OK" "${response}"
} || {
check_result "redis" "DEGRADED" "Redis connectivity failed (non-critical)"
}
else
check_result "redis" "SKIPPED" "Not in full mode"
fi
}
# ---- Check 5: Metrics endpoint ----
check_metrics() {
local response
response=$(curl -sf --max-time "${TIMEOUT}" "${METRICS_URL}/metrics" 2>&1) && {
check_result "metrics" "OK" "Prometheus metrics endpoint responding"
} || {
check_result "metrics" "DEGRADED" "Metrics endpoint unreachable (non-critical)"
}
}
# ---- Check 6: WebSocket hub (full mode) ----
check_websocket() {
if [[ "${FULL_CHECK}" == "true" ]]; then
# WebSocket health check via HTTP status endpoint
local response
response=$(curl -sf --max-time "${TIMEOUT}" "${BASE_URL}/health?check=websocket" 2>&1) && {
check_result "websocket" "OK" "${response}"
} || {
check_result "websocket" "DEGRADED" "WebSocket hub unreachable"
}
else
check_result "websocket" "SKIPPED" "Not in full mode"
fi
}
# ---- Run all checks ----
check_liveness
check_readiness
check_database
check_redis
check_metrics
check_websocket
# ---- Output results ----
if [[ "${JSON_OUTPUT}" == "true" ]]; then
# JSON output for programmatic consumption
echo '{'
echo ' "checks": ['
for i in "${!RESULTS[@]}"; do
IFS='|' read -r name status detail <<< "${RESULTS[$i]}"
comma=""
[[ $i -gt 0 ]] && comma=","
echo " ${comma}{\"name\": \"${name}\", \"status\": \"${status}\", \"detail\": \"${detail}\"}"
done
echo ' ],'
echo ' "summary": {'
echo ' "critical_failures": ${CRITICAL_FAIL},'
echo ' "degraded_failures": ${DEGRADED_FAIL}'
echo ' }'
echo '}'
else
# Human-readable output
echo "=== GoChat Health Check Report ==="
for result in "${RESULTS[@]}"; do
IFS='|' read -r name status detail <<< "${result}"
case "$status" in
OK) icon="✅" ;;
CRITICAL_FAIL) icon="❌" ;;
DEGRADED) icon="⚠️" ;;
SKIPPED) icon="⏭️" ;;
esac
echo " ${icon} ${name}: ${status}${detail}"
done
echo ""
echo "Critical failures: ${CRITICAL_FAIL} | Degraded: ${DEGRADED_FAIL}"
fi
# ---- Determine exit code ----
if [[ ${CRITICAL_FAIL} -gt 0 ]]; then
exit 2
elif [[ ${DEGRADED_FAIL} -gt 0 ]]; then
exit 1
else
exit 0
fi