use NetKeepAlive for px check server alive

src/sql/dtl/ob_dtl_basic_channel.cpp

@@ -30,8 +30,8 @@
 #include "sql/dtl/ob_dtl_interm_result_manager.h"
 #include "sql/dtl/ob_dtl_channel_loop.h"
 #include "sql/dtl/ob_dtl_channel_watcher.h"
-#include "share/ob_server_blacklist.h"
 #include "observer/omt/ob_th_worker.h"
+#include "sql/engine/px/ob_px_util.h"
 #include "sql/session/ob_sql_session_info.h"
 
 using namespace oceanbase::common;
@@ -850,8 +850,7 @@ int ObDtlBasicChannel::wait_unblocking()
                 LOG_WARN("worker interrupt", K(tmp_ret), K(ret));
                 break;
               }
-              if (OB_UNLIKELY(share::ObServerBlacklist::get_instance().is_in_blacklist(
+              if (OB_UNLIKELY(ObPxCheckAlive::is_in_blacklist(peer_,
-                    share::ObCascadMember(peer_, GCONF.cluster_id), true,
                               channel_loop_->get_process_query_time()))) {
                 ret = OB_RPC_CONNECT_ERROR;
                 LOG_WARN("peer no in communication, maybe crashed", K(ret), K(peer_),

src/sql/dtl/ob_dtl_channel_loop.cpp

@@ -15,9 +15,9 @@
 #include "share/interrupt/ob_global_interrupt_call.h"
 #include "observer/omt/ob_tenant_config_mgr.h"
 #include "share/diagnosis/ob_sql_monitor_statname.h"
-#include "share/ob_server_blacklist.h"
 #include "observer/omt/ob_th_worker.h"
 #include "share/ob_occam_time_guard.h"
+#include "sql/engine/px/ob_px_util.h"
 
 using namespace oceanbase::common;
 
@@ -359,9 +359,7 @@ int ObDtlChannelLoop::process_channels(ObIDltChannelLoopPred *pred, int64_t &nth
       LOG_WARN("unexpect next idx", K(next_idx_), K(chan_cnt), K(ret));
     } else {
       chan = chans_[next_idx_];
-      if (OB_UNLIKELY(share::ObServerBlacklist::get_instance().is_in_blacklist(
+      if (OB_UNLIKELY(ObPxCheckAlive::is_in_blacklist(chan->get_peer(), get_process_query_time()))) {
-            share::ObCascadMember(chan->get_peer(), GCONF.cluster_id), true,
-            get_process_query_time()))) {
         ret = OB_RPC_CONNECT_ERROR;
         LOG_WARN("peer no in communication, maybe crashed", K(ret), K(chan->get_peer()),
                   K(static_cast<int64_t>(GCONF.cluster_id)));
@@ -415,8 +413,7 @@ int ObDtlChannelLoop::process_channel(int64_t &nth_channel)
   }
   ObDtlChannel *ch = sentinel_node_.next_link_;
   while (OB_EAGAIN == ret && ch != &sentinel_node_) {
-    if (OB_UNLIKELY(share::ObServerBlacklist::get_instance().is_in_blacklist(
+    if (OB_UNLIKELY(ObPxCheckAlive::is_in_blacklist(ch->get_peer(),
-          share::ObCascadMember(ch->get_peer(), GCONF.cluster_id), true,
           get_process_query_time()))) {
       ret = OB_RPC_CONNECT_ERROR;
       LOG_WARN("peer no in communication, maybe crashed", K(ret), K(ch->get_peer()),

src/sql/engine/px/ob_dfo_scheduler.cpp

@@ -21,7 +21,6 @@
 #include "sql/engine/px/ob_px_dtl_msg.h"
 #include "sql/engine/px/ob_px_rpc_processor.h"
 #include "sql/engine/px/ob_px_sqc_async_proxy.h"
-#include "share/ob_server_blacklist.h"
 
 using namespace oceanbase::common;
 using namespace oceanbase::share;
@@ -489,9 +488,7 @@ int ObSerialDfoScheduler::dispatch_sqcs(ObExecContext &exec_ctx,
     ObPxSqcMeta &sqc = *sqcs.at(idx);
     const ObAddr &addr = sqc.get_exec_addr();
     auto proxy = coord_info_.rpc_proxy_.to(addr);
-    if (OB_UNLIKELY(share::ObServerBlacklist::get_instance().is_in_blacklist(
+    if (OB_UNLIKELY(ObPxCheckAlive::is_in_blacklist(addr, session->get_process_query_time()))) {
-                        share::ObCascadMember(addr, cluster_id), true /* add_server */,
-                        session->get_process_query_time()))) {
       if (!ignore_vtable_error) {
         ret = OB_RPC_CONNECT_ERROR;
         LOG_WARN("peer no in communication, maybe crashed", K(ret), K(sqc), K(cluster_id),

src/sql/engine/px/ob_px_data_ch_provider.cpp

@@ -14,7 +14,6 @@
 
 #include "ob_px_data_ch_provider.h"
 #include "sql/engine/px/ob_px_util.h"
-#include "share/ob_server_blacklist.h"
 
 
 using namespace oceanbase::common;
@@ -426,9 +425,7 @@ int ObPxChProviderUtil::check_status(int64_t timeout_ts, const ObAddr &qc_addr,
   } else if (timeout_ts <= ObTimeUtility::current_time()) {
     ret = OB_TIMEOUT;
     LOG_WARN("timeout and abort", K(timeout_ts), K(ret));
-  } else if (OB_UNLIKELY(share::ObServerBlacklist::get_instance().is_in_blacklist(
+  } else if (OB_UNLIKELY(ObPxCheckAlive::is_in_blacklist(qc_addr, query_start_time))) {
-                    share::ObCascadMember(qc_addr, GCONF.cluster_id), true,
-                    query_start_time))) {
     ret = OB_RPC_CONNECT_ERROR;
     LOG_WARN("peer no in communication, maybe crashed", K(ret), K(qc_addr),
               K(static_cast<int64_t>(GCONF.cluster_id)));

src/sql/engine/px/ob_px_sqc_async_proxy.cpp

@@ -13,7 +13,7 @@
 #define USING_LOG_PREFIX SQL_ENG
 
 #include "sql/engine/px/ob_px_sqc_async_proxy.h"
-#include "share/ob_server_blacklist.h"
+#include "sql/engine/px/ob_px_util.h"
 
 namespace oceanbase {
 using namespace common;
@@ -68,9 +68,7 @@ int ObPxSqcAsyncProxy::launch_all_rpc_request() {
         args.enable_serialize_cache();
       }
       ARRAY_FOREACH_X(sqcs_, idx, count, OB_SUCC(ret)) {
-        if (OB_UNLIKELY(share::ObServerBlacklist::get_instance().is_in_blacklist(
+        if (OB_UNLIKELY(ObPxCheckAlive::is_in_blacklist(sqcs_.at(idx)->get_exec_addr(),
-                          share::ObCascadMember(sqcs_.at(idx)->get_exec_addr(), cluster_id),
-                          true /* add_server */,
                         session_->get_process_query_time()))) {
           ret = OB_RPC_CONNECT_ERROR;
           LOG_WARN("peer no in communication, maybe crashed", K(ret),

src/sql/engine/px/ob_px_util.cpp

@@ -28,9 +28,9 @@
 #include "share/schema/ob_part_mgr_util.h"
 #include "sql/engine/dml/ob_table_insert_op.h"
 #include "sql/session/ob_sql_session_info.h"
-#include "share/ob_server_blacklist.h"
 #include "common/ob_smart_call.h"
 #include "storage/ob_locality_manager.h"
+#include "rpc/obrpc/ob_net_keepalive.h"
 
 using namespace oceanbase::common;
 using namespace oceanbase::sql;
@@ -3480,14 +3480,12 @@ int ObExtraServerAliveCheck::do_check() const
     if (OB_FAIL(dfo_mgr_->get_running_dfos(dfos))) {
       LOG_WARN("fail find dfo", K(ret));
     } else {
-      share::ObServerBlacklist &server_black_list = share::ObServerBlacklist::get_instance();
       // need check all sqc because we set sqc need_report = false here and don't need wait sqc finish msg.
       for (int64_t i = 0; i < dfos.count(); i++) {
         ObIArray<ObPxSqcMeta> &sqcs = dfos.at(i)->get_sqcs();
         for (int64_t j = 0; j < sqcs.count(); j++) {
           if (sqcs.at(j).need_report()) {
-            if (OB_UNLIKELY(server_black_list.is_in_blacklist(
+            if (OB_UNLIKELY(ObPxCheckAlive::is_in_blacklist(sqcs.at(j).get_exec_addr(),
-                share::ObCascadMember(sqcs.at(j).get_exec_addr(), cluster_id_), true,
                 query_start_time_))) {
               sqcs.at(j).set_need_report(false);
               sqcs.at(j).set_thread_finish(true);
@@ -3503,8 +3501,7 @@ int ObExtraServerAliveCheck::do_check() const
       }
     }
   } else if (OB_LIKELY(qc_addr_.is_valid())) {
-    if (OB_UNLIKELY(share::ObServerBlacklist::get_instance().is_in_blacklist(share::ObCascadMember(
+    if (OB_UNLIKELY(ObPxCheckAlive::is_in_blacklist(qc_addr_, query_start_time_))) {
-          qc_addr_, cluster_id_), true, query_start_time_))) {
       ret = OB_RPC_CONNECT_ERROR;
       LOG_WARN("qc not in communication, maybe crashed", K(ret), K(qc_addr_));
     }
@@ -3535,3 +3532,19 @@ bool ObVirtualTableErrorWhitelist::should_ignore_vtable_error(int error_code)
   }
   return should_ignore;
 }
+
+bool ObPxCheckAlive::is_in_blacklist(const common::ObAddr &addr, int64_t server_start_time)
+{
+  int ret = OB_SUCCESS;
+  bool in_blacklist = false;
+  obrpc::ObNetKeepAliveData alive_data;
+  if (OB_FAIL(ObNetKeepAlive::get_instance().in_black(addr, in_blacklist, &alive_data))) {
+    LOG_WARN("check in black failed", K(ret));
+  } else if (!in_blacklist && server_start_time > 0) {
+    in_blacklist = alive_data.start_service_time_ >= server_start_time;
+  }
+  if (in_blacklist) {
+    LOG_WARN("server in blacklist", K(addr), K(server_start_time), K(alive_data.start_service_time_));
+  }
+  return in_blacklist;
+}

src/sql/engine/px/ob_px_util.h

@@ -548,6 +548,12 @@ public:
   static bool should_ignore_vtable_error(int error_code);
 };
 
+class ObPxCheckAlive
+{
+public:
+  static bool is_in_blacklist(const common::ObAddr &addr, int64_t server_start_time);
+};
+
 }
 }