!122 支持配置等待静态主的时间

Merge pull request !122 from 薛蒙恩/wait_static_primary_time

src/cm_ctl/ctl_param_check.cpp

@@ -127,6 +127,9 @@ const char *g_cmsParamInfo[] = {
     "cms_enable_failover_on2nodes|bool|0,0|NULL|NULL|",
     "cms_enable_db_crash_recovery|bool|0,0|NULL|NULL|",
     "cms_network_isolation_timeout|int|10,2147483647|NULL|NULL|",
+#ifndef ENABLE_PRIVATEGAUSS
+    "wait_static_primary_times|int|5,2147483647|NULL|NULL|",
+#endif
 };
 
 const char *g_valueTypeStr[] = {

src/cm_server/cm_server.centralized.conf.sample

@@ -87,4 +87,6 @@ cms_enable_db_crash_recovery = false	 # used in 2 nodes cluster. when network re
 cms_network_isolation_timeout = 20		 # cms judges the network is isolated when it finds ddb cluster is not sync with each other nodes,
 										 # after cms_network_isolation_timeout times.
 										 # default 20
+wait_static_primary_times = 6            # Time to wait for the primary recovery after the primary stopped unexpectedly.
+                                         # default value is 6
 ############### must leave a new line at the end ###################

src/cm_server/cms_arbitrate_datanode_pms.cpp

@@ -2248,7 +2248,7 @@ static void InitDnArbCond(DnArbCtx *ctx)
     ctx->cond.maxMemArbiTime = 0;
     ctx->cond.instMainta = IsMaintance(ctx->maintaMode);
     ctx->cond.switchoverIdx = INVALID_INDEX;
-    ctx->cond.arbitInterval = g_clusterStarting ? g_clusterStartingArbitDelay : DATANODE_ARBITE_DELAY;
+    ctx->cond.arbitInterval = g_clusterStarting ? g_clusterStartingArbitDelay : g_waitStaticPrimaryTimes;
     ctx->cond.arbitStaticInterval = 5;
     ctx->cond.setOffline = SetOfflineNode();
     ctx->cond.snameAzDnCount = 0;

src/cm_server/cms_arbitrate_datanode_pms_utils.cpp

@@ -554,7 +554,7 @@ uint32 GetDnArbitateDelayTime(const DnArbCtx *ctx)
     /* if static primary has finished redo, not need to wait for 180s */
     cm_local_replconninfo *status = &(ctx->dnReport[cond->staticPriIdx].local_status);
     if (status->local_role == INSTANCE_ROLE_STANDBY && status->disconn_mode == PROHIBIT_CONNECTION) {
-        return DATANODE_ARBITE_DELAY;
+        return g_waitStaticPrimaryTimes;
     }
     return cond->arbitInterval;
 }

src/cm_server/cms_common.cpp

@@ -615,6 +615,12 @@ void get_parameters_from_configfile()
     g_diskTimeout = get_uint32_value_from_config(configDir, "disk_timeout", 200);
     g_agentNetworkTimeout = get_uint32_value_from_config(configDir, "agent_network_timeout", 6);
     GetDnArbitrateMode();
+#ifndef ENABLE_PRIVATEGAUSS
+    g_waitStaticPrimaryTimes = get_uint32_value_from_config(configDir, "wait_static_primary_times", 6);
+    if (g_waitStaticPrimaryTimes < 5) {
+        g_waitStaticPrimaryTimes = 5;
+    }
+#endif
 }
 
 void clean_init_cluster_state()

src/cm_server/cms_global_params.cpp

@@ -328,6 +328,7 @@ uint32 g_delayArbiTime = 0;
 int32 g_clusterArbiTime = 300;
 bool g_isPauseArbitration = false;
 char g_cmManualPausePath[MAX_PATH_LEN] = {0};
+uint32 g_waitStaticPrimaryTimes = 6;
 
 bool isLargerNode()
 {

src/cm_server/cms_monitor_main.cpp

@@ -344,11 +344,18 @@ static void ReloadParametersFromConfigfile()
     g_diskTimeout = get_uint32_value_from_config(configDir, "disk_timeout", 200);
     g_agentNetworkTimeout = get_uint32_value_from_config(configDir, "agent_network_timeout", 6);
     GetDnArbitrateMode();
+#ifndef ENABLE_PRIVATEGAUSS
+    g_waitStaticPrimaryTimes = get_uint32_value_from_config(configDir, "wait_static_primary_times", 6);
+    if (g_waitStaticPrimaryTimes < 5) {
+        g_waitStaticPrimaryTimes = 5;
+    }
+#endif
 
     if (g_cm_server_num == CMS_ONE_PRIMARY_ONE_STANDBY) {
         GetTwoNodesArbitrateParams();
     }
 
+
 #ifdef ENABLE_MULTIPLE_NODES
     write_runlog(LOG,
         "reload cm_server parameters:\n"
@@ -379,13 +386,13 @@ static void ReloadParametersFromConfigfile()
         "datastorage_threshold_check_interval=%d,\n"
         "  max_datastorage_threshold_check=%d, enableSetReadOnly=%s, enableSetReadOnlyThreshold=%u, "
         "switch_rto=%d, force_promote=%d, cluster_starting_aribt_delay=%u, enable_e2e_rto=%u, "
-        "g_delayArbiTime=%u, g_clusterArbiTime=%d.\n",
+        "g_delayArbiTime=%u, g_clusterArbiTime=%d, wait_static_primary_times=%u.\n",
         log_min_messages, maxLogFileSize, sys_log_path, g_alarmComponentPath, g_alarmReportInterval,
         instance_heartbeat_timeout, g_ddbArbicfg.haHeartBeatTimeOut, cmserver_self_vote_timeout,
         g_ddbArbicfg.haStatusInterval, cmserver_ha_connect_timeout, instance_failover_delay_timeout,
         datastorage_threshold_check_interval, max_datastorage_threshold_check, g_enableSetReadOnly,
         g_readOnlyThreshold, switch_rto, force_promote, g_clusterStartingArbitDelay,
-        g_enableE2ERto, g_delayArbiTime, g_clusterArbiTime);
+        g_enableE2ERto, g_delayArbiTime, g_clusterArbiTime, g_waitStaticPrimaryTimes);
 #endif
 }
 

src/include/cm/cm_server/cms_arbitrate_datanode_pms_utils.h

@@ -44,7 +44,6 @@ typedef struct DnArbitInfo_t {
     uint32 maxTerm;
 } DnArbitInfo;
 
-const uint32 DATANODE_ARBITE_DELAY = 6;
 
 extern bool CheckPotentialTermRollback();
 extern void GroupStatusShow(const char *str, const uint32 groupIndex, const uint32 instanceId,

src/include/cm/cm_server/cms_global_params.h

@@ -475,6 +475,7 @@ extern uint32 g_delayArbiTime;
 extern int32 g_clusterArbiTime;
 extern bool g_isPauseArbitration;
 extern char g_cmManualPausePath[MAX_PATH_LEN];
+extern uint32 g_waitStaticPrimaryTimes;
 
 extern void clean_init_cluster_state();
 extern void instance_delay_arbitrate_time_out_direct_clean(uint32 group_index, int member_index,