maintenance: record p99 put/multiput/get/multiget rpc byte size (#323)

apache · May 6, 2019 · c9488b7 · c9488b7
1 parent c0341cd
commit c9488b7
Show file tree

Hide file tree

Showing 6 changed files with 30 additions and 9 deletions.
diff --git a/rdsn b/rdsn
diff --git a/scripts/falcon_screen.json b/scripts/falcon_screen.json
@@ -700,6 +700,19 @@
       "graph_type": "a",
       "method": "",
       "timespan": 86400
+    },
+    {
+      "title": "各节点 P99 RPC 报文长度",
+      "endpoints": ["cluster=${cluster.name} job=replica service=pegasus"],
+      "counters": [
+        "zion*profiler*RPC_RRDB_RRDB_PUT.size.request.server/cluster=${cluster.name},job=replica,port=${replica.port},service=pegasus",
+        "zion*profiler*RPC_RRDB_RRDB_MULTI_PUT.size.request.server/cluster=${cluster.name},job=replica,port=${replica.port},service=pegasus",
+        "zion*profiler*RPC_RRDB_RRDB_GET.size.response.server/cluster=${cluster.name},job=replica,port=${replica.port},service=pegasus",
+        "zion*profiler*RPC_RRDB_RRDB_MULTI_GET.size.response.server/cluster=${cluster.name},job=replica,port=${replica.port},service=pegasus"
+      ],
+      "graph_type": "a",
+      "method": "",
+      "timespan": 86400
     }
   ]
 }
diff --git a/src/reporter/pegasus_counter_reporter.cpp b/src/reporter/pegasus_counter_reporter.cpp
@@ -154,9 +154,9 @@ void pegasus_counter_reporter::update()
         oss << "logging perf counter(name, type, value):" << std::endl;
         oss << std::fixed << std::setprecision(2);
         perf_counters::instance().iterate_snapshot(
-            [&oss](const dsn::perf_counter_ptr &ptr, double val) {
-                oss << "[" << ptr->full_name() << ", " << dsn_counter_type_to_string(ptr->type())
-                    << ", " << val << "]" << std::endl;
+            [&oss](const dsn::perf_counters::counter_snapshot &cs) {
+                oss << "[" << cs.name << ", " << dsn_counter_type_to_string(cs.type) << ", "
+                    << cs.value << "]" << std::endl;
             });
         ddebug("%s", oss.str().c_str());
     }
@@ -168,9 +168,9 @@ void pegasus_counter_reporter::update()
         bool first_append = true;
         _falcon_metric.timestamp = timestamp;
         perf_counters::instance().iterate_snapshot(
-            [&oss, &first_append, this](const dsn::perf_counter_ptr &ptr, double val) {
-                _falcon_metric.metric = ptr->full_name();
-                _falcon_metric.value = val;
+            [&oss, &first_append, this](const dsn::perf_counters::counter_snapshot &cs) {
+                _falcon_metric.metric = cs.name;
+                _falcon_metric.value = cs.value;
                 _falcon_metric.counterType = "GAUGE";
                 if (!first_append)
                     oss << ",";

diff --git a/src/server/brief_stat.cpp b/src/server/brief_stat.cpp
@@ -47,10 +47,10 @@ std::string get_brief_stat()
     oss << std::fixed << std::setprecision(0);
     bool first_item = true;
     dsn::perf_counters::snapshot_iterator iter =
-        [&oss, &first_item](const dsn::perf_counter_ptr &counter, double value) mutable {
+        [&oss, &first_item](const dsn::perf_counters::counter_snapshot &cs) mutable {
             if (!first_item)
                 oss << ", ";
-            oss << s_brief_stat_map.find(counter->full_name())->second << "=" << value;
+            oss << s_brief_stat_map.find(cs.name)->second << "=" << cs.value;
             first_item = false;
         };
     std::vector<bool> match_result;

diff --git a/src/server/config-server.ini b/src/server/config-server.ini
@@ -435,6 +435,7 @@ profiler::inqueue = false
 ;profiler::qps = false
 profiler::cancelled = false
 ;profiler::latency.server = false
+profiler::size.request.server = true
 
 [task.RPC_RRDB_RRDB_PUT_ACK]
 is_profile = true
@@ -449,6 +450,7 @@ profiler::inqueue = false
 ;profiler::qps = false
 profiler::cancelled = false
 ;profiler::latency.server = false
+profiler::size.request.server = true
 
 [task.RPC_RRDB_RRDB_MULTI_PUT_ACK]
 is_profile = true
@@ -533,6 +535,7 @@ profiler::inqueue = false
 ;profiler::qps = false
 profiler::cancelled = false
 ;profiler::latency.server = false
+profiler::size.response.server = true
 
 [task.RPC_RRDB_RRDB_GET_ACK]
 is_profile = true
@@ -541,6 +544,7 @@ is_profile = true
 rpc_request_throttling_mode = TM_DELAY
 rpc_request_delays_milliseconds = 1000, 1000, 1000, 1000, 1000, 10000
 is_profile = true
+profiler::size.response.server = true
 
 [task.RPC_RRDB_RRDB_SORTKEY_COUNT]
 rpc_request_throttling_mode = TM_DELAY

diff --git a/src/server/config.ini b/src/server/config.ini
@@ -465,6 +465,7 @@
   rpc_request_throttling_mode = TM_DELAY
   rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
+  profiler::size.request.server = true
 
 [task.RPC_RRDB_RRDB_PUT_ACK]
   is_profile = true
@@ -473,6 +474,7 @@
   rpc_request_throttling_mode = TM_DELAY
   rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
+  profiler::size.request.server = true
 
 [task.RPC_RRDB_RRDB_MULTI_PUT_ACK]
   is_profile = true
@@ -521,6 +523,7 @@
   rpc_request_throttling_mode = TM_DELAY
   rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
+  profiler::size.response.server = true
 
 [task.RPC_RRDB_RRDB_GET_ACK]
   is_profile = true
@@ -529,6 +532,7 @@
   rpc_request_throttling_mode = TM_DELAY
   rpc_request_delays_milliseconds = 50, 50, 50, 50, 50, 100
   is_profile = true
+  profiler::size.response.server = true
 
 [task.RPC_RRDB_RRDB_MULTI_GET_ACK]
   is_profile = true
+12 −2		include/dsn/cpp/serialization_helper/dsn.layer2_types.h
+12 −16		include/dsn/dist/replication/replication_types.h
+2 −3		include/dsn/perf_counter/perf_counter.h
+23 −20		include/dsn/perf_counter/perf_counters.h
+3 −11		include/dsn/toollet/profiler.h
+27 −0		src/core/core/dsn.layer2_types.cpp
+28 −28		src/core/perf_counter/perf_counters.cpp
+0 −0		src/core/tests/perf_counter_test.cpp
+9 −8		src/core/tests/perf_counters_test.cpp
+57 −26		src/core/tools/common/profiler.cpp
+11 −11		src/core/tools/common/profiler_command.cpp
+9 −19		src/core/tools/common/profiler_header.h
+4 −4		src/core/tools/common/profiler_output.cpp
+1 −3		src/dist/replication/common/duplication_common.cpp
+23 −24		src/dist/replication/common/replication_types.cpp
+2 −2		src/dist/replication/ddl_lib/replication_ddl_client.cpp
+14 −1		src/dist/replication/lib/replica_stub.cpp
+2 −0		src/dist/replication/lib/replica_stub.h
+5 −1		src/dist/replication/meta_server/CMakeLists.txt
+183 −0		src/dist/replication/meta_server/duplication/duplication_info.cpp
+218 −0		src/dist/replication/meta_server/duplication/duplication_info.h
+172 −0		src/dist/replication/meta_server/duplication/meta_duplication_service.cpp
+96 −0		src/dist/replication/meta_server/duplication/meta_duplication_service.h
+4 −0		src/dist/replication/meta_server/meta_data.h
+51 −10		src/dist/replication/meta_server/meta_service.cpp
+17 −2		src/dist/replication/meta_server/meta_service.h
+5 −2		src/dist/replication/meta_server/server_state.h
+2 −2		src/dist/replication/replication.thrift
+43 −3		src/dist/replication/test/meta_test/misc/misc.h
+4 −1		src/dist/replication/test/meta_test/unit_test/CMakeLists.txt
+7 −0		src/dist/replication/test/meta_test/unit_test/config-test.ini
+0 −17		src/dist/replication/test/meta_test/unit_test/data_definition_test.cpp
+179 −0		src/dist/replication/test/meta_test/unit_test/duplication_info_test.cpp
+83 −43		src/dist/replication/test/meta_test/unit_test/meta_service_test_app.h
+6 −1		src/dsn.layer2.thrift