[PLAT-13773] Add event type grouping to ASH

Summary: Support grouping by event type for ASH graphs Test Plan: unit tested + tested manually. Reviewers: rmadhavan Reviewed By: rmadhavan Subscribers: yugaware Differential Revision: https://phorge.dev.yugabyte.com/D34740
yugabyte · May 3, 2024 · 0b840d6 · 0b840d6
1 parent e13c915
commit 0b840d6
Show file tree

Hide file tree

Showing 18 changed files with 359 additions and 639 deletions.
diff --git a/...shoot/backend/src/main/java/com/yugabyte/troubleshoot/ts/models/ActiveSessionHistory.java b/...shoot/backend/src/main/java/com/yugabyte/troubleshoot/ts/models/ActiveSessionHistory.java
@@ -42,23 +42,26 @@ public class ActiveSessionHistory extends Model {
   private String waitEventClass;
 
   @CsvField(pos = 8)
+  private String waitEventType;
+
+  @CsvField(pos = 9)
   private String waitEvent;
 
-  @CsvField(pos = 9, converterType = UuidConverter.class)
+  @CsvField(pos = 10, converterType = UuidConverter.class)
   private UUID topLevelNodeId;
 
-  @CsvField(pos = 10)
+  @CsvField(pos = 11)
   private Long queryId;
 
-  @CsvField(pos = 11)
+  @CsvField(pos = 12)
   private Long ysqlSessionId;
 
-  @CsvField(pos = 12)
+  @CsvField(pos = 13)
   private String clientNodeIp;
 
-  @CsvField(pos = 13)
+  @CsvField(pos = 14)
   private String waitEventAux;
 
-  @CsvField(pos = 14)
+  @CsvField(pos = 15)
   private int sampleWeight;
 }
diff --git a/troubleshoot/backend/src/main/java/com/yugabyte/troubleshoot/ts/models/GraphData.java b/troubleshoot/backend/src/main/java/com/yugabyte/troubleshoot/ts/models/GraphData.java
@@ -32,6 +32,7 @@ public class GraphData {
   private String type;
   private String waitEventComponent;
   private String waitEventClass;
+  private String waitEventType;
   private String waitEvent;
   private Map<String, String> labels;
   @JsonIgnore private List<GraphPoint> points = new ArrayList<>();

diff --git a/troubleshoot/backend/src/main/java/com/yugabyte/troubleshoot/ts/models/GraphLabel.java b/troubleshoot/backend/src/main/java/com/yugabyte/troubleshoot/ts/models/GraphLabel.java
@@ -22,6 +22,7 @@ public enum GraphLabel {
   mountPoint("mountpoint"),
   waitEventComponent(""),
   waitEventClass(""),
+  waitEventType(""),
   waitEvent(""),
   clientNodeIp("");
 

diff --git a/...oot/backend/src/main/java/com/yugabyte/troubleshoot/ts/service/TsStorageGraphService.java b/...oot/backend/src/main/java/com/yugabyte/troubleshoot/ts/service/TsStorageGraphService.java
@@ -341,6 +341,9 @@ public GraphResponse getGraph(
                 graphData.setWaitEventClass(
                     key.getLabels().remove(GraphLabel.waitEventClass.name()));
               }
+              if (key.getLabels().containsKey(GraphLabel.waitEventType.name())) {
+                graphData.setWaitEventType(key.getLabels().remove(GraphLabel.waitEventType.name()));
+              }
               if (key.getLabels().containsKey(GraphLabel.waitEvent.name())) {
                 graphData.setWaitEvent(key.getLabels().remove(GraphLabel.waitEvent.name()));
               }

diff --git a/...ot/backend/src/main/java/com/yugabyte/troubleshoot/ts/task/ActiveSessionHistoryQuery.java b/...ot/backend/src/main/java/com/yugabyte/troubleshoot/ts/task/ActiveSessionHistoryQuery.java
@@ -49,8 +49,9 @@ public class ActiveSessionHistoryQuery {
 
   static final String ASH_QUERY_NO_TIMESTAMP =
       "select sample_time, root_request_id, rpc_request_id, wait_event_component, wait_event_class,"
-          + " wait_event, top_level_node_id, query_id, ysql_session_id, client_node_ip,"
-          + " wait_event_aux, sample_weight from pg_catalog.yb_active_session_history";
+          + " wait_event_type, wait_event, top_level_node_id, query_id, ysql_session_id,"
+          + " client_node_ip, wait_event_aux, sample_weight"
+          + " from pg_catalog.yb_active_session_history";
 
   static final String ASH_SAMPLE_TIME_FILTER = " where sample_time >= ";
 
@@ -61,6 +62,7 @@ public class ActiveSessionHistoryQuery {
   private static final String PRC_REQUEST_ID = "rpc_request_id";
   private static final String WAIT_EVENT_COMPONENT = "wait_event_component";
   private static final String WAIT_EVENT_CLASS = "wait_event_class";
+  private static final String WAIT_EVENT_TYPE = "wait_event_type";
   private static final String WAIT_EVENT = "wait_event";
   private static final String TOP_LEVEL_NODE_ID = "top_level_node_id";
   private static final String QUERY_ID = "query_id";
@@ -232,6 +234,7 @@ private NodeProcessResult processNode(
                   .setRootRequestId(UUID.fromString(statsJson.get(ROOT_REQUEST_ID).asText()))
                   .setWaitEventComponent(statsJson.get(WAIT_EVENT_COMPONENT).asText())
                   .setWaitEventClass(statsJson.get(WAIT_EVENT_CLASS).asText())
+                  .setWaitEventType(statsJson.get(WAIT_EVENT_TYPE).asText())
                   .setWaitEvent(statsJson.get(WAIT_EVENT).asText())
                   .setTopLevelNodeId(UUID.fromString(statsJson.get(TOP_LEVEL_NODE_ID).asText()))
                   .setQueryId(statsJson.get(QUERY_ID).asLong())

diff --git a/troubleshoot/backend/src/main/resources/db/migration/V11__Add_event_type_to_ASH_table.sql b/troubleshoot/backend/src/main/resources/db/migration/V11__Add_event_type_to_ASH_table.sql
@@ -0,0 +1,2 @@
+ALTER TABLE active_session_history ADD COLUMN wait_event_type TEXT;
+
diff --git a/troubleshoot/backend/src/main/resources/graphs/ts_storage_graphs.yml b/troubleshoot/backend/src/main/resources/graphs/ts_storage_graphs.yml
@@ -107,9 +107,12 @@ active_session_history_tserver:
     waitEventClass:
       name: wait_event_class
       assumesGroupBy: ["waitEventComponent"]
+    waitEventType:
+      name: wait_event_type
+      defaultGroupBy: true
+      assumesGroupBy: ["waitEventComponent"]
     waitEvent:
       name: wait_event
-      defaultGroupBy: true
       assumesGroupBy: ["waitEventClass", "waitEventComponent"]
   additional_filter: "ysql_session_id != 0"
   data_columns:
@@ -123,6 +126,8 @@ active_session_history_tserver:
       type: "date"
     metadata:
       supportedGroupBy:
+        - label: waitEventType
+          name: Type
         - label: waitEvent
           name: Event
         - label: waitEventClass
@@ -158,9 +163,12 @@ active_session_history_ysql:
     waitEventClass:
       name: wait_event_class
       assumesGroupBy: ["waitEventComponent"]
+    waitEventType:
+      name: wait_event_type
+      defaultGroupBy: true
+      assumesGroupBy: ["waitEventComponent"]
     waitEvent:
       name: wait_event
-      defaultGroupBy: true
       assumesGroupBy: ["waitEventClass", "waitEventComponent"]
   additional_filter: "ysql_session_id != 0"
   data_columns:
@@ -174,6 +182,8 @@ active_session_history_ysql:
       type: "date"
     metadata:
       supportedGroupBy:
+        - label: waitEventType
+          name: Type
         - label: waitEvent
           name: Event
         - label: waitEventClass

diff --git a/...bleshoot/backend/src/test/java/com/yugabyte/troubleshoot/ts/service/GraphServiceTest.java b/...bleshoot/backend/src/test/java/com/yugabyte/troubleshoot/ts/service/GraphServiceTest.java
@@ -60,17 +60,23 @@ public void setUp() {
     pgStatStatementsService.save(createStatements("node2", 1L, 30));
     pgStatStatementsService.save(createStatements("node3", 1L, 30));
     activeSessionHistoryService.save(
-        createAshEntries("node1", "TServer", "Consensus", "WAL_Sync", 1L, "0.0.0.0", 50));
+        createAshEntries(
+            "node1", "TServer", "Consensus", "Network", "WAL_Sync", 1L, "0.0.0.0", 50));
     activeSessionHistoryService.save(
-        createAshEntries("node1", "YSQL", "YSQLQuery", "QueryProcessing", 200L, "1.1.1.1", 30));
+        createAshEntries(
+            "node1", "YSQL", "YSQLQuery", "Network", "QueryProcessing", 200L, "1.1.1.1", 30));
     activeSessionHistoryService.save(
-        createAshEntries("node1", "YSQL", "TServerWait", "CatalogRead", 200L, "1.1.1.1", 50));
+        createAshEntries(
+            "node1", "YSQL", "TServerWait", "Network", "CatalogRead", 200L, "1.1.1.1", 50));
     activeSessionHistoryService.save(
-        createAshEntries("node2", "TServer", "Consensus", "WAL_Sync", 1L, "0.0.0.0", 30));
+        createAshEntries(
+            "node2", "TServer", "Consensus", "Network", "WAL_Sync", 1L, "0.0.0.0", 30));
     activeSessionHistoryService.save(
-        createAshEntries("node2", "YSQL", "YSQLQuery", "QueryProcessing", 200L, "2.2.2.2", 50));
+        createAshEntries(
+            "node2", "YSQL", "YSQLQuery", "Network", "QueryProcessing", 200L, "2.2.2.2", 50));
     activeSessionHistoryService.save(
-        createAshEntries("node2", "YSQL", "TServerWait", "CatalogRead", 300L, "2.2.2.2", 50));
+        createAshEntries(
+            "node2", "YSQL", "TServerWait", "Network", "CatalogRead", 300L, "2.2.2.2", 50));
 
     server = MockRestServiceServer.createServer(prometheusClientTemplate);
   }
@@ -484,6 +490,7 @@ private List<ActiveSessionHistory> createAshEntries(
       String nodeName,
       String waitEventComponent,
       String waitEventClass,
+      String waitEventType,
       String waitEvent,
       long queryId,
       String clientNodeIp,
@@ -496,6 +503,7 @@ private List<ActiveSessionHistory> createAshEntries(
                     nodeName,
                     waitEventComponent,
                     waitEventClass,
+                    waitEventType,
                     waitEvent,
                     queryId,
                     clientNodeIp))
@@ -507,6 +515,7 @@ private ActiveSessionHistory createAshEntry(
       String nodeName,
       String waitEventComponent,
       String waitEventClass,
+      String waitEventType,
       String waitEvent,
       long queryId,
       String clientNodeIp) {
@@ -518,6 +527,7 @@ private ActiveSessionHistory createAshEntry(
         .setRpcRequestId(new Random().nextLong())
         .setWaitEventComponent(waitEventComponent)
         .setWaitEventClass(waitEventClass)
+        .setWaitEventType(waitEventType)
         .setWaitEvent(waitEvent)
         .setTopLevelNodeId(UUID.randomUUID())
         .setQueryId(queryId)