fix: recover stalled update requests

README.md

@@ -56,7 +56,7 @@ PostgreSQL 18 兼容工作继续冻结。
 先在宝塔面板停止并删除当前错误的 Java 项目，再执行：
 
 ```bash
-curl -fsSL https://git.awaioi.com/ERP-Team/kaidi/releases/download/v1.0.0-preview.28/install.sh | sudo env KAIDI_APP_PORT=18080 bash
+curl -fsSL https://git.awaioi.com/ERP-Team/kaidi/releases/download/v1.0.0-preview.29/install.sh | sudo env KAIDI_APP_PORT=18080 bash
 ```
 
 该命令只安装程序运行所需的 systemd 单元，自动创建 `kaidi` 用户并检测现有 Java 17（包括

deploy/systemd/kaidi-update.service

@@ -18,5 +18,5 @@ IOSchedulingPriority=6
 PrivateTmp=true
 ProtectHome=true
 ProtectSystem=full
-ReadWritePaths=/opt/kaidi /www/wwwroot/kaidi /var/lib/kaidi /var/lib/kaidi-update /var/log/kaidi /etc/systemd/system
+ReadWritePaths=/opt/kaidi -/www/wwwroot/kaidi /var/lib/kaidi /var/lib/kaidi-update /var/log/kaidi /etc/systemd/system
 UMask=0077

deploy/update.sh

@@ -62,6 +62,9 @@ case "$RUNTIME_ENV_FILE" in
 esac
 
 bootstrap_die() {
+  if [ -f "$PROCESSING_FILE" ] && [ ! -L "$PROCESSING_FILE" ]; then
+    fail "$1"
+  fi
   printf '%s\n' "$1" >&2
   exit 1
 }
@@ -174,6 +177,15 @@ prepare_update_layout() {
     || bootstrap_die "Private update directories could not be prepared"
 }
 
+prepare_queue_layout() {
+  if [ -d "$PROCESSING_DIR" ] && [ ! -L "$PROCESSING_DIR" ] \
+    && [ -d "$FAILED_REQUEST_ROOT" ] && [ ! -L "$FAILED_REQUEST_ROOT" ]; then
+    return 0
+  fi
+  install -d -o root -g root -m 0700 "$PROCESSING_DIR" "$FAILED_REQUEST_ROOT" \
+    || bootstrap_die "Private update queue directories could not be prepared"
+}
+
 secure_release_tree() {
   release_dir=$1
   chown -R root:"$SERVICE_GROUP" "$release_dir" || return 1
@@ -197,9 +209,6 @@ verify_release_access() {
   fi
 }
 
-prepare_update_layout
-load_runtime_database_env
-
 # shellcheck disable=SC2329 # Invoked by the EXIT trap below.
 cleanup() {
   rc=$?
@@ -616,6 +625,7 @@ validate_release_operations() {
     "$WORK_DIR/extracted/ops/kaidi-update.path" >/dev/null || return 1
 }
 
+prepare_queue_layout
 exec 9>"$LOCK_FILE"
 flock -n 9 || exit 0
 recover_interrupted_transaction
@@ -650,6 +660,8 @@ TARGET_VERSION=$REQUESTED_VERSION
 REQUEST_ACTION=$(jq -er '(.action // "INSTALL") | strings | ascii_upcase
   | select(. == "DOWNLOAD" or . == "INSTALL")' "$PROCESSING_FILE") \
   || fail "Update request action is invalid"
+prepare_update_layout
+load_runtime_database_env
 [ -z "$RELEASE_TOKEN" ] || { [ "${#RELEASE_TOKEN}" -le 512 ] \
   && ! printf '%s' "$RELEASE_TOKEN" | grep -q '[[:cntrl:]]'; } \
   || fail "UPDATE_RELEASE_TOKEN is invalid"

scripts/test-install-fixture.sh

@@ -319,6 +319,8 @@ grep -Fqx 'StartLimitBurst=3' "$ROOT/deploy/systemd/kaidi-finance.service" \
   || fail 'application service no longer has a bounded restart burst'
 grep -Fqx 'Restart=no' "$ROOT/deploy/systemd/kaidi-update.service" \
   || fail 'update service can automatically repeat a failed switching transaction'
+grep -Fq -- '-/www/wwwroot/kaidi' "$ROOT/deploy/systemd/kaidi-update.service" \
+  || fail 'update service requires the Baota application path on a systemd-only installation'
 ! grep -Fq '/var/lib/kaidi-update/processing' "$ROOT/deploy/systemd/kaidi-update.path" \
   || fail 'update path can automatically repeat a claimed switching transaction'
 # shellcheck disable=SC2016 # Match literal installer source.

scripts/test-update-fixture.sh

@@ -26,18 +26,35 @@ for version in 01.2.3 1.02.3 1.2.03 1.2.3-01 1.2.3-alpha..1; do
 done
 
 missing_service_user="kaidi-fixture-missing-$$"
+mkdir -p "$WORK/bootstrap/app" "$WORK/bootstrap/state/inbox" \
+  "$WORK/bootstrap/state/processing" "$WORK/bootstrap/state/failed" "$WORK/bootstrap/log" \
+  "$WORK/bootstrap/bin"
+cat > "$WORK/bootstrap/bin/flock" <<'SH'
+#!/bin/sh
+exit 0
+SH
+chmod 0755 "$WORK/bootstrap/bin/flock"
+jq -n \
+  '{action:"DOWNLOAD",version:"1.0.0-preview.2",reason:"bootstrap fixture"}' \
+  > "$WORK/bootstrap/state/inbox/request.json"
 if KAIDI_APP_ROOT="$WORK/bootstrap/app" \
   KAIDI_UPDATE_STATE_ROOT="$WORK/bootstrap/state" \
   KAIDI_LOG_ROOT="$WORK/bootstrap/log" \
   KAIDI_SERVICE_USER="$missing_service_user" \
   KAIDI_SERVICE_GROUP="$missing_service_user" \
+  PATH="$WORK/bootstrap/bin:$PATH" \
   sh "$ROOT/deploy/update.sh" > "$WORK/bootstrap.log" 2>&1; then
   fail 'updater accepted a missing service identity during bootstrap'
 fi
 grep -Fq "Service user $missing_service_user is missing" "$WORK/bootstrap.log" \
   || fail 'updater bootstrap failure did not preserve its diagnostic'
-! grep -Eq 'No such file|nonexistent directory|cannot create' "$WORK/bootstrap.log" \
-  || fail 'updater bootstrap failure was masked by an unavailable status directory'
+[ "$(jq -r '.state' "$WORK/bootstrap/state/status.json")" = FAILED ] \
+  || fail 'updater bootstrap failure did not persist FAILED'
+[ ! -e "$WORK/bootstrap/state/inbox/request.json" ] \
+  && [ ! -e "$WORK/bootstrap/state/processing/request.json" ] \
+  || fail 'updater bootstrap failure left a request permanently queued'
+find "$WORK/bootstrap/state/failed" -type f -name 'request-*.json' -print -quit | grep -q . \
+  || fail 'updater bootstrap failure did not archive the claimed request'
 
 write_mock_commands() {
   local mock_bin=$1