fix: improve shell get_app_stat and ls_nodes & fix scripts (#382)

scripts/clusters_show.sh

@@ -11,6 +11,7 @@ if [ $# -lt 2 ]; then
   exit 1
 fi
 
+PID=$$
 clusters_file=$1
 format=$2
 if [ "$format" != "table" -a "$format" != "csv" ]; then
@@ -49,8 +50,8 @@ while read cluster
 do
   tmp_file="/tmp/$UID.$PID.pegasus.clusters_status.cluster_info"
   echo "cluster_info" | ./run.sh shell -n $cluster &>$tmp_file
-  cluster_info_ok=`grep "succeed" $tmp_file | wc -l`
-  if [ $cluster_info_ok -ne 1 ]; then
+  cluster_info_fail=`grep "\<failed\>" $tmp_file | wc -l`
+  if [ $cluster_info_fail -eq 1 ]; then
     echo "ERROR: get cluster info failed, refer error to $tmp_file"
     exit 1
   fi
@@ -71,8 +72,13 @@ do
   echo "app_stat -o $app_stat_result" | ./run.sh shell -n $cluster &>$tmp_file
   app_stat_fail=`grep "\<failed\>" $tmp_file | wc -l`
   if [ $app_stat_fail -eq 1 ]; then
-    echo "ERROR: app stat failed, refer error to $tmp_file"
-    exit 1
+    sleep 1
+    echo "app_stat -o $app_stat_result" | ./run.sh shell -n $cluster &>$tmp_file
+    app_stat_fail=`grep "\<failed\>" $tmp_file | wc -l`
+    if [ $app_stat_fail -eq 1 ]; then
+      echo "ERROR: app stat failed, refer error to $tmp_file"
+      exit 1
+    fi
   fi
   app_count=`cat $app_stat_result | wc -l`
   app_count=$((app_count-2))

scripts/pegasus_manual_compact.sh

@@ -58,8 +58,8 @@ function get_env()
 
     log_file="/tmp/$UID.$PID.pegasus.get_app_envs.${app_name}"
     echo -e "use ${app_name}\n get_app_envs" | ./run.sh shell --cluster ${cluster} &>${log_file}
-    get_ok=`grep 'get app envs succeed' ${log_file} | wc -l`
-    if [ ${get_ok} -ne 1 ]; then
+    get_fail=`grep 'get app env failed' ${log_file} | wc -l`
+    if [ ${get_fail} -eq 1 ]; then
         echo "ERROR: get app envs failed, refer to ${log_file}"
         exit 1
     fi
@@ -77,8 +77,8 @@ function set_env()
     echo "set_app_envs ${key}=${value}"
     log_file="/tmp/$UID.$PID.pegasus.set_app_envs.${app_name}"
     echo -e "use ${app_name}\n set_app_envs ${key} ${value}" | ./run.sh shell --cluster ${cluster} &>${log_file}
-    set_ok=`grep 'set app envs succeed' ${log_file} | wc -l`
-    if [ ${set_ok} -ne 1 ]; then
+    set_fail=`grep 'set app env failed' ${log_file} | wc -l`
+    if [ ${set_fail} -eq 1 ]; then
         echo "ERROR: set app envs failed, refer to ${log_file}"
         exit 1
     fi

scripts/pegasus_replica_thread.sh

@@ -27,8 +27,8 @@ all_start_time=$((`date +%s`))
 echo
 
 echo "app $app_name -d" | ./run.sh shell --cluster $cluster &>/tmp/$UID.$PID.pegasus.app.$app_name
-list_ok=`grep "list app $app_name succeed" /tmp/$UID.$PID.pegasus.app.$app_name | wc -l`
-if [ $list_ok -ne 1 ]; then
+list_fail=`grep "\<failed\>" /tmp/$UID.$PID.pegasus.app.$app_name | wc -l`
+if [ $list_fail -eq 1 ]; then
   grep ERR /tmp/$UID.$PID.pegasus.app.$app_name
   echo "ERROR: list app failed, refer to /tmp/$UID.$PID.pegasus.app.$app_name"
   exit 1

scripts/pegasus_set_usage_scenario.sh

@@ -33,8 +33,8 @@ all_start_time=$((`date +%s`))
 echo
 
 echo -e "use $app_name\nset_app_envs $scenario_key $scenario" | ./run.sh shell --cluster $cluster &>/tmp/$UID.$PID.pegasus.set_app_envs
-set_ok=`grep 'set app envs succeed' /tmp/$UID.$PID.pegasus.set_app_envs | wc -l`
-if [ $set_ok -ne 1 ]; then
+set_fail=`grep 'set app env failed' /tmp/$UID.$PID.pegasus.set_app_envs | wc -l`
+if [ $set_fail -eq 1 ]; then
   grep ERR /tmp/$UID.$PID.pegasus.set_app_envs
   echo "ERROR: set app envs failed, refer to /tmp/$UID.$PID.pegasus.set_app_envs"
   exit 1

src/shell/command_helper.h

@@ -680,12 +680,13 @@ get_app_stat(shell_context *sc, const std::string &app_name, std::vector<row_dat
     }
 
     ::dsn::command command;
-    command.cmd = "perf-counters-by-substr";
+    command.cmd = "perf-counters";
     char tmp[256];
-    if (app_name.empty())
-        sprintf(tmp, "@");
-    else
-        sprintf(tmp, "@%d.", app_info->app_id);
+    if (app_name.empty()) {
+        sprintf(tmp, ".*@.*");
+    } else {
+        sprintf(tmp, ".*@%d\\..*", app_info->app_id);
+    }
     command.arguments.emplace_back(tmp);
     std::vector<std::pair<bool, std::string>> results;
     call_remote_command(sc, nodes, command, results);

src/shell/commands/node_management.cpp

@@ -171,7 +171,7 @@ bool ls_nodes(command_executor *e, shell_context *sc, arguments args)
     if (resource_usage) {
         std::vector<node_desc> nodes;
         if (!fill_nodes(sc, "replica-server", nodes)) {
-            derror("get replica server node list failed");
+            std::cout << "get replica server node list failed" << std::endl;
             return true;
         }
 
@@ -192,21 +192,20 @@ bool ls_nodes(command_executor *e, shell_context *sc, arguments args)
             if (tmp_it == tmp_map.end())
                 continue;
             if (!results[i].first) {
-                derror("query perf counter info from node %s failed", node_addr.to_string());
+                std::cout << "query perf counter info from node " << node_addr.to_string()
+                          << " failed" << std::endl;
                 return true;
             }
             dsn::perf_counter_info info;
             dsn::blob bb(results[i].second.data(), 0, results[i].second.size());
             if (!dsn::json::json_forwarder<dsn::perf_counter_info>::decode(bb, info)) {
-                derror("decode perf counter info from node %s failed, result = %s",
-                       node_addr.to_string(),
-                       results[i].second.c_str());
+                std::cout << "decode perf counter info from node " << node_addr.to_string()
+                          << " failed, result = " << results[i].second << std::endl;
                 return true;
             }
             if (info.result != "OK") {
-                derror("query perf counter info from node %s returns error, error = %s",
-                       node_addr.to_string(),
-                       info.result.c_str());
+                std::cout << "query perf counter info from node " << node_addr.to_string()
+                          << " returns error, error = " << info.result << std::endl;
                 return true;
             }
             list_nodes_helper &h = tmp_it->second;