[Fix](checker) Fixed infinite loop after internal error in checker. (#…

…44479)


When the checker encounters an internal error, such as a transaction
conflict, the return value will be less than 0 and the function will
return immediately, but the related instance will not be removed from
the map. Additionally, if the return value of the do check is not 0, the
inverted check will not be performed. This PR fixes both of these
issues.

cloud/src/recycler/checker.cpp

@@ -168,25 +168,17 @@ int Checker::start() {
             auto ctime_ms =
                     duration_cast<milliseconds>(system_clock::now().time_since_epoch()).count();
             g_bvar_checker_enqueue_cost_s.put(instance_id, ctime_ms / 1000 - enqueue_time_s);
-            ret = checker->do_check();
+            int ret1 = checker->do_check();
 
+            int ret2 = 0;
             if (config::enable_inverted_check) {
-                if (ret == 0) {
-                    ret = checker->do_inverted_check();
-                }
-            }
-
-            if (ret < 0) {
-                // If ret < 0, it means that a temporary error occurred during the check process.
-                // The check job should not be considered finished, and the next round of check job
-                // should be retried as soon as possible.
-                return;
+                ret2 = checker->do_inverted_check();
             }
 
             // If instance checker has been aborted, don't finish this job
             if (!checker->stopped()) {
                 finish_instance_recycle_job(txn_kv_.get(), check_job_key, instance.instance_id(),
-                                            ip_port_, ret == 0, ctime_ms);
+                                            ip_port_, ret1 == 0 && ret2 == 0, ctime_ms);
             }
             {
                 std::lock_guard lock(mtx_);