Files
platform-core/scripts/bootstrap/apply-host-nginx-gitea.sh
T

352 lines
12 KiB
Bash
Executable File

#!/usr/bin/env bash
set -Eeuo pipefail
# Do not expose headers, cookies, or future sensitive values through caller xtrace.
set +x
umask 077
readonly EXPECTED_CANDIDATE_SHA256="de7ebd4f69cd7d2204ee633e074bf6a4370a6f5e3f3fac9067b099d5d75269b5"
readonly EXPECTED_PRE_CUTOVER_SHA256="5b5941519ab677f751568827aa4f315193dfbd9eeac5e8a4ad85d724aaefe16f"
readonly EXPECTED_PRE_CUTOVER_HEALTH_SHA256="6b683cb16987ff1f5ded22e9847ac0a45995947927c459b21f427523a41c7484"
readonly REPOSITORY_ROOT="/home/donghyeon/workspace/platform"
readonly CANDIDATE="${REPOSITORY_ROOT}/infrastructure/networking/host-nginx/learn-services.conf"
readonly ACTIVE="/etc/nginx/sites-available/learn-services"
readonly ENABLED="/etc/nginx/sites-enabled/learn-services"
readonly CURL_BIN="/usr/bin/curl"
readonly JQ_BIN="/usr/bin/jq"
readonly NGINX_BIN="/usr/sbin/nginx"
readonly SYSTEMCTL_BIN="/usr/bin/systemctl"
readonly INSTALL_BIN="/usr/bin/install"
readonly SHA256SUM_BIN="/usr/bin/sha256sum"
readonly STAT_BIN="/usr/bin/stat"
readonly READLINK_BIN="/usr/bin/readlink"
readonly MKTEMP_BIN="/usr/bin/mktemp"
readonly RM_BIN="/usr/bin/rm"
readonly DATE_BIN="/usr/bin/date"
readonly AWK_BIN="/usr/bin/awk"
readonly SLEEP_BIN="/usr/bin/sleep"
fail() {
printf 'ERROR: %s\n' "$*" >&2
return 1
}
usage() {
printf '%s\n' \
'Usage: sudo bash scripts/bootstrap/apply-host-nginx-gitea.sh --execute' \
'' \
'Backs up the active learn-services site, installs the reviewed Gitea proxy,' \
'tests and reloads Nginx, then runs local acceptance checks. Any failure after' \
'the active file changes triggers an automatic restore and reload.'
}
[[ "${1:-}" == "--execute" && "$#" -eq 1 ]] || {
usage
exit 2
}
[[ "$EUID" -eq 0 ]] || fail "run this script through sudo"
[[ -t 0 ]] || fail "an interactive terminal is required"
for required_binary in \
"$CURL_BIN" "$JQ_BIN" "$NGINX_BIN" "$SYSTEMCTL_BIN" "$INSTALL_BIN" \
"$SHA256SUM_BIN" "$STAT_BIN" "$READLINK_BIN" "$MKTEMP_BIN" \
"$RM_BIN" "$DATE_BIN" "$AWK_BIN" "$SLEEP_BIN"; do
[[ -x "$required_binary" ]] || fail "required executable is missing: ${required_binary}"
done
[[ -f "$CANDIDATE" && ! -L "$CANDIDATE" ]] || fail "unsafe candidate: ${CANDIDATE}"
[[ -f "$ACTIVE" && ! -L "$ACTIVE" ]] || fail "unsafe active file: ${ACTIVE}"
[[ -L "$ENABLED" ]] || fail "enabled path is not a symlink: ${ENABLED}"
[[ "$("$READLINK_BIN" -f "$ENABLED")" == "$ACTIVE" ]] || fail "enabled symlink target changed"
[[ "$("$STAT_BIN" --format='%U:%G %a' "$ACTIVE")" == "root:root 644" ]] || \
fail "active file owner or mode changed"
"$SYSTEMCTL_BIN" is-active --quiet nginx || fail "nginx is not active"
readonly TEMP_DIR="$("$MKTEMP_BIN" -d /tmp/nginx-gitea-cutover.XXXXXX)"
readonly CANDIDATE_SNAPSHOT="${TEMP_DIR}/learn-services.candidate"
readonly DIRECT_HEALTH="${TEMP_DIR}/direct-health.json"
readonly NGINX_HEALTH="${TEMP_DIR}/nginx-health.json"
readonly LOGIN_HEADERS="${TEMP_DIR}/login-headers"
rollback_armed=0
backup=""
cleanup() {
case "$TEMP_DIR" in
/tmp/nginx-gitea-cutover.*)
"$RM_BIN" -rf -- "$TEMP_DIR"
;;
*)
printf 'WARNING: refusing to remove unexpected temp path: %s\n' "$TEMP_DIR" >&2
;;
esac
}
rollback() {
local restore_install_rc
local restore_test_rc
local restore_reload_rc
set +e
printf '\nROLLBACK: restoring %s\n' "$backup" >&2
if "$INSTALL_BIN" -o root -g root -m 0644 "$backup" "$ACTIVE" &&
[[ "$("$SHA256SUM_BIN" "$ACTIVE" | "$AWK_BIN" '{print $1}')" == "$("$SHA256SUM_BIN" "$backup" | "$AWK_BIN" '{print $1}')" ]]; then
restore_install_rc=0
else
restore_install_rc=1
fi
"$NGINX_BIN" -t
restore_test_rc=$?
if (( restore_install_rc == 0 && restore_test_rc == 0 )); then
if "$SYSTEMCTL_BIN" reload nginx &&
"$SYSTEMCTL_BIN" is-active --quiet nginx &&
wait_for_rollback_state; then
restore_reload_rc=0
else
restore_reload_rc=1
fi
else
restore_reload_rc=1
fi
if (( restore_install_rc == 0 && restore_test_rc == 0 && restore_reload_rc == 0 )); then
rollback_armed=0
printf 'ROLLBACK complete. Active config and response were restored.\n' >&2
else
printf 'CRITICAL: automatic rollback failed; backup remains at %s\n' "$backup" >&2
fi
}
on_exit() {
local rc=$?
trap - EXIT INT TERM
if (( rc != 0 && rollback_armed == 1 )); then
rollback
fi
cleanup
exit "$rc"
}
trap on_exit EXIT
trap 'exit 130' INT
trap 'exit 143' TERM
"$INSTALL_BIN" -o root -g root -m 0600 "$CANDIDATE" "$CANDIDATE_SNAPSHOT"
readonly CANDIDATE_SHA256="$("$SHA256SUM_BIN" "$CANDIDATE_SNAPSHOT" | "$AWK_BIN" '{print $1}')"
readonly ACTIVE_SHA256="$("$SHA256SUM_BIN" "$ACTIVE" | "$AWK_BIN" '{print $1}')"
[[ "$CANDIDATE_SHA256" == "$EXPECTED_CANDIDATE_SHA256" ]] || \
fail "candidate digest changed: ${CANDIDATE_SHA256}"
if [[ "$ACTIVE_SHA256" != "$EXPECTED_PRE_CUTOVER_SHA256" && "$ACTIVE_SHA256" != "$CANDIDATE_SHA256" ]]; then
fail "active config has an unexpected digest: ${ACTIVE_SHA256}"
fi
check_health_json() {
local path="$1"
"$JQ_BIN" -e '
.status == "pass" and
([.checks["database:ping"][], .checks["cache:ping"][]] |
all(.status == "pass"))
' "$path" >/dev/null
}
pre_cutover_check() {
"$CURL_BIN" --noproxy '*' --fail-with-body --silent --show-error --max-time 10 \
--header 'Host: git.learn.hyeonworks.com' \
http://127.0.0.1:30080/api/healthz >"$DIRECT_HEALTH"
check_health_json "$DIRECT_HEALTH"
}
wait_for_nginx_health() {
local attempt
local probe_metadata=""
local http_code="curl-error"
local content_type="unavailable"
local size_download="0"
local body_sha256=""
local classification="transport-error"
local consecutive_passes=0
for ((attempt = 1; attempt <= 10; attempt++)); do
if probe_metadata="$(
"$CURL_BIN" --noproxy '*' --silent --connect-timeout 1 --max-time 2 \
--resolve git.learn.hyeonworks.com:443:127.0.0.1 \
--output "$NGINX_HEALTH" \
--write-out $'%{http_code}\t%{content_type}\t%{size_download}' \
https://git.learn.hyeonworks.com/api/healthz
)"; then
IFS=$'\t' read -r http_code content_type size_download <<<"$probe_metadata"
body_sha256="$(
"$SHA256SUM_BIN" "$NGINX_HEALTH" | "$AWK_BIN" '{print $1}'
)"
if [[ "$http_code" == "200" ]] &&
check_health_json "$NGINX_HEALTH" 2>/dev/null; then
classification="healthy-json"
consecutive_passes=$((consecutive_passes + 1))
if (( consecutive_passes >= 2 )); then
printf 'Nginx proxy health stabilized after %d probes.\n' "$attempt"
return 0
fi
else
consecutive_passes=0
if [[ "$body_sha256" == "$EXPECTED_PRE_CUTOVER_HEALTH_SHA256" ]]; then
classification="stale-old-generation"
elif [[ "$http_code" =~ ^(502|503|504)$ ]]; then
classification="transient-upstream"
elif "$JQ_BIN" -e '
type == "object" and
has("status") and
(.checks | type == "object")
' "$NGINX_HEALTH" >/dev/null 2>&1; then
classification="unhealthy-health-json"
else
printf 'ERROR: unexpected Nginx health response (status=%s, content-type=%s, bytes=%s, sha256=%s)\n' \
"$http_code" "${content_type:-none}" "$size_download" "$body_sha256" >&2
return 1
fi
fi
else
http_code="curl-error"
content_type="unavailable"
size_download="0"
body_sha256=""
classification="transport-error"
consecutive_passes=0
fi
if (( attempt < 10 )); then
"$SLEEP_BIN" 1
fi
done
fail "Nginx health did not converge during the bounded retry window (classification=${classification}, status=${http_code}, content-type=${content_type:-none}, bytes=${size_download})"
}
wait_for_rollback_state() {
local attempt
local probe_metadata=""
local http_code="curl-error"
local content_type="unavailable"
local size_download="0"
local body_sha256=""
for ((attempt = 1; attempt <= 10; attempt++)); do
if probe_metadata="$(
"$CURL_BIN" --noproxy '*' --silent --connect-timeout 1 --max-time 2 \
--resolve git.learn.hyeonworks.com:443:127.0.0.1 \
--output "$NGINX_HEALTH" \
--write-out $'%{http_code}\t%{content_type}\t%{size_download}' \
https://git.learn.hyeonworks.com/api/healthz
)"; then
IFS=$'\t' read -r http_code content_type size_download <<<"$probe_metadata"
body_sha256="$(
"$SHA256SUM_BIN" "$NGINX_HEALTH" | "$AWK_BIN" '{print $1}'
)"
if [[ "$http_code" == "200" &&
"$body_sha256" == "$EXPECTED_PRE_CUTOVER_HEALTH_SHA256" ]]; then
if (( attempt > 1 )); then
printf 'Rollback proxy state stabilized after %d probes.\n' "$attempt" >&2
fi
return 0
fi
else
http_code="curl-error"
content_type="unavailable"
size_download="0"
body_sha256=""
fi
if (( attempt < 10 )); then
"$SLEEP_BIN" 1
fi
done
printf 'CRITICAL: rollback response did not converge (status=%s, content-type=%s, bytes=%s, sha256=%s)\n' \
"$http_code" "${content_type:-none}" "$size_download" "${body_sha256:-none}" >&2
return 1
}
post_cutover_checks() {
local id_body
local redirect_result
wait_for_nginx_health
id_body="$(
"$CURL_BIN" --noproxy '*' --fail-with-body --silent --show-error --max-time 10 \
--resolve id.learn.hyeonworks.com:443:127.0.0.1 \
https://id.learn.hyeonworks.com/
)"
[[ "$id_body" == "Keycloak domain reached Nginx successfully" ]] || \
fail "Keycloak hold response changed"
redirect_result="$(
"$CURL_BIN" --noproxy '*' --silent --show-error --max-time 10 \
--resolve git.learn.hyeonworks.com:80:127.0.0.1 \
--output /dev/null --write-out $'%{http_code}\n%{redirect_url}' \
http://git.learn.hyeonworks.com/api/healthz
)"
[[ "$redirect_result" == $'301\nhttps://git.learn.hyeonworks.com/api/healthz' ]] || \
fail "HTTP redirect check failed: ${redirect_result}"
if "$CURL_BIN" --noproxy '*' --insecure --silent --output /dev/null --max-time 5 \
--resolve unconfigured.invalid:443:127.0.0.1 \
https://unconfigured.invalid/ 2>/dev/null; then
fail "unknown TLS hostname was not rejected"
fi
"$CURL_BIN" --noproxy '*' --fail-with-body --silent --show-error --max-time 10 \
--resolve git.learn.hyeonworks.com:443:127.0.0.1 \
--dump-header "$LOGIN_HEADERS" --output /dev/null \
https://git.learn.hyeonworks.com/user/login
"$AWK_BIN" '
BEGIN { found = 0; insecure = 0 }
tolower($0) ~ /^set-cookie:/ {
found++
if (tolower($0) !~ /; secure([;[:space:]]|$)/) insecure = 1
}
END { exit(found == 0 || insecure) }
' "$LOGIN_HEADERS" || fail "login cookie Secure check failed"
}
pre_cutover_check
"$NGINX_BIN" -t
if [[ "$ACTIVE_SHA256" == "$CANDIDATE_SHA256" ]]; then
post_cutover_checks
printf 'The reviewed Host Nginx configuration is already active and healthy.\n'
exit 0
fi
backup="${ACTIVE}.before-gitea-$("$DATE_BIN" +%Y%m%d%H%M%S)"
[[ ! -e "$backup" && ! -L "$backup" ]] || fail "backup path already exists: ${backup}"
printf '\nActive SHA-256: %s\n' "$ACTIVE_SHA256"
printf 'Candidate SHA-256: %s\n' "$CANDIDATE_SHA256"
printf 'Planned backup: %s\n' "$backup"
printf 'Type APPLY to replace the Host Nginx site: '
read -r confirmation
[[ "$confirmation" == "APPLY" ]] || fail "cancelled"
"$INSTALL_BIN" -o root -g root -m 0644 "$ACTIVE" "$backup"
[[ "$("$SHA256SUM_BIN" "$backup" | "$AWK_BIN" '{print $1}')" == "$ACTIVE_SHA256" ]] || \
fail "backup digest mismatch"
rollback_armed=1
"$INSTALL_BIN" -o root -g root -m 0644 "$CANDIDATE_SNAPSHOT" "$ACTIVE"
[[ "$("$SHA256SUM_BIN" "$ACTIVE" | "$AWK_BIN" '{print $1}')" == "$CANDIDATE_SHA256" ]] || \
fail "installed digest mismatch"
"$NGINX_BIN" -t
"$SYSTEMCTL_BIN" reload nginx
"$SYSTEMCTL_BIN" is-active --quiet nginx
post_cutover_checks
rollback_armed=0
printf '\nCUTOVER SUCCESS\n'
printf 'Backup: %s\n' "$backup"
printf 'Candidate SHA-256: %s\n' "$CANDIDATE_SHA256"
printf 'Local HTTPS health, redirect, Secure cookie, Keycloak hold, and unknown-host rejection: PASS\n'
printf 'Run the public HTTPS and Git clone/push checks from a separate client next.\n'