grix-connector 3.22.0 → 3.24.0

package/scripts/upgrade-guardian.sh

@@ -59,32 +59,49 @@ log "started: from=$FROM_VERSION target=$TARGET_VERSION crash_count=$CRASH_COUNT
 # Wait for old daemon to exit (SIGTERM + shutdown + restart)
 sleep 15
 
-# Monitor loop: wait for healthz to return 200
-ELAPSED=0
-while [ $ELAPSED -lt $HEALTH_TIMEOUT ]; do
+# 前向激活：用已装好的目标版本 CLI start 刷新受管 runtime 副本、服务定义并触发重启。
+# 放在 guardian（detached 独立进程）而非旧 daemon 进程内执行：旧 daemon 的激活子
+# 进程在 systemd 默认 KillMode=control-group 下会随 restart 一起被杀，且旧 daemon
+# 在 await 期间被杀后 pending 的失败处理（回滚/上报）执行不到，会与 guardian 脱同步。
+# Linux 无受管 runtime 副本（服务直接跑全局包），start 只做服务定义刷新与重启，同样安全。
+# 激活失败视同新版本启动失败，落入下方崩溃计数/回滚路径，不删 pending。
+TARGET_CLI="$(npm root -g 2>> "$LOG_FILE")/$NPM_PACKAGE/dist/grix.js"
+ACTIVATED=0
+if [ -f "$TARGET_CLI" ] && node "$TARGET_CLI" start >> "$LOG_FILE" 2>&1; then
+    ACTIVATED=1
+    log "forward activation succeeded: managed runtime switched to $TARGET_VERSION"
+else
+    log "forward activation failed for target $TARGET_VERSION; treating as startup failure"
+fi
+
+# Monitor loop: wait for healthz to return 200 (only after successful activation)
+if [ $ACTIVATED -eq 1 ]; then
+    ELAPSED=0
+    while [ $ELAPSED -lt $HEALTH_TIMEOUT ]; do
+        if curl -sf "$HEALTH_URL" > /dev/null 2>&1; then
+            # healthz 通过=新版健康。这里不删 pending：删了会抢在新进程的
+            # handlePendingOnStartup 之前,使其读不到 pending、success 回执发不出去
+            # (后端只见 installed 不见 success)。交由新进程上报 success 后自行删 pending,
+            # guardian 只负责崩溃回滚。
+            log "healthz passed, upgrade healthy; leaving pending for in-process success report"
+            exit 0
+        fi
+        sleep 3
+        ELAPSED=$((ELAPSED + 3))
+    done
+
+    # Extra grace period for slow systems
+    sleep 15
     if curl -sf "$HEALTH_URL" > /dev/null 2>&1; then
-        # healthz 通过=新版健康。这里不删 pending：删了会抢在新进程的
-        # handlePendingOnStartup 之前,使其读不到 pending、success 回执发不出去
-        # (后端只见 installed 不见 success)。交由新进程上报 success 后自行删 pending,
-        # guardian 只负责崩溃回滚。
-        log "healthz passed, upgrade healthy; leaving pending for in-process success report"
+        # 同上:不删 pending,交由进程内 handlePendingOnStartup 上报 success 并删除。
+        log "healthz passed (delayed), upgrade healthy; leaving pending for in-process success report"
         exit 0
     fi
-    sleep 3
-    ELAPSED=$((ELAPSED + 3))
-done
-
-# Extra grace period for slow systems
-sleep 15
-if curl -sf "$HEALTH_URL" > /dev/null 2>&1; then
-    # 同上:不删 pending,交由进程内 handlePendingOnStartup 上报 success 并删除。
-    log "healthz passed (delayed), upgrade healthy; leaving pending for in-process success report"
-    exit 0
 fi
 
-# Health check failed
+# Health check failed (or forward activation failed)
 CRASH_COUNT=$((CRASH_COUNT + 1))
-log "healthz timeout, crash_count=$CRASH_COUNT"
+log "healthz timeout or forward activation failed, crash_count=$CRASH_COUNT"
 
 if [ $CRASH_COUNT -ge $MAX_CRASHES ]; then
     # 新版升级器会把完整旧包保留在同文件系统的固定备份目录。优先原子恢复，
@@ -98,13 +115,23 @@ if [ $CRASH_COUNT -ge $MAX_CRASHES ]; then
                 if { [ ! -e "$PACKAGE_DIR" ] || mv "$PACKAGE_DIR" "$FAILED_DIR"; } \
                    && mv "$BACKUP_DIR" "$PACKAGE_DIR"; then
                     rm -rf "$FAILED_DIR"
-                    rm -f "$PENDING_FILE"
-                    log "atomic rollback succeeded"
-                    exit 0
-                fi
-                log "atomic rollback failed; restoring current package before npm fallback"
-                if [ ! -e "$PACKAGE_DIR" ] && [ -e "$FAILED_DIR" ]; then
-                    mv "$FAILED_DIR" "$PACKAGE_DIR" 2>/dev/null || true
+                    # 原子恢复只换回全局包；macOS 服务跑的是受管 runtime 副本，必须用
+                    # 旧包 CLI start 刷新副本并触发重启，激活成功才删 pending。
+                    NPM_ROOT="$(npm root -g 2>> "$LOG_FILE")"
+                    ROLLBACK_CLI="$NPM_ROOT/$NPM_PACKAGE/dist/grix.js"
+                    if [ -n "$NPM_ROOT" ] && [ -f "$ROLLBACK_CLI" ] \
+                       && node "$ROLLBACK_CLI" start >> "$LOG_FILE" 2>&1; then
+                        rm -f "$PENDING_FILE"
+                        log "atomic rollback succeeded and managed runtime activated"
+                        exit 0
+                    fi
+                    # 激活失败不退出：落到下面的 npm 回滚路径再试一次（npm install 旧版幂等）。
+                    log "atomic rollback package restored but managed runtime activation failed; falling through to npm rollback"
+                else
+                    log "atomic rollback failed; restoring current package before npm fallback"
+                    if [ ! -e "$PACKAGE_DIR" ] && [ -e "$FAILED_DIR" ]; then
+                        mv "$FAILED_DIR" "$PACKAGE_DIR" 2>/dev/null || true
+                    fi
                 fi
             fi
             ;;
@@ -115,8 +142,16 @@ if [ $CRASH_COUNT -ge $MAX_CRASHES ]; then
     if npm install -g "${NPM_PACKAGE}@${FROM_VERSION}" --prefer-online --no-audit --no-fund >> "$LOG_FILE" 2>&1 \
        || { log "rollback via default registry failed, retrying official npm"; \
             npm install -g "${NPM_PACKAGE}@${FROM_VERSION}" --registry https://registry.npmjs.org --prefer-online --no-audit --no-fund >> "$LOG_FILE" 2>&1; }; then
-        log "rollback succeeded"
-        rm -f "$PENDING_FILE"
+        log "rollback package installed; activating managed runtime"
+        NPM_ROOT="$(npm root -g 2>> "$LOG_FILE")"
+        ROLLBACK_CLI="$NPM_ROOT/$NPM_PACKAGE/dist/grix.js"
+        if [ -n "$NPM_ROOT" ] && [ -f "$ROLLBACK_CLI" ] \
+           && node "$ROLLBACK_CLI" start >> "$LOG_FILE" 2>&1; then
+            log "rollback succeeded and managed runtime activated"
+            rm -f "$PENDING_FILE"
+        else
+            log "ROLLBACK ACTIVATION FAILED — package restored but managed runtime was not switched"
+        fi
     else
         log "ROLLBACK FAILED — manual intervention required"
     fi