Implement CCS telemetry export as part of _cluster/stats (#112310)

smalyshev · web-flow · commit 9081a951d501 · 2024-09-10T09:31:06.000-06:00
* Implement CCS telemetry export as part of _cluster/stats
diff --git a/docs/reference/cluster/stats.asciidoc b/docs/reference/cluster/stats.asciidoc
@@ -1307,6 +1307,142 @@ Each repository type may also include other statistics about the repositories of
 
 ====
 
+`ccs`::
+(object) Contains information relating to <<modules-cross-cluster-search, {ccs}>> settings and activity in the cluster.
++
+.Properties of `ccs`
+[%collapsible%open]
+=====
+
+
+`_search`:::
+(object) Contains the telemetry information about the <<modules-cross-cluster-search, {ccs}>> usage in the cluster.
++
+.Properties of `_search`
+[%collapsible%open]
+======
+`total`:::
+(integer) The total number of {ccs} requests that have been executed by the cluster.
+
+`success`:::
+(integer) The total number of {ccs} requests that have been successfully executed by the cluster.
+
+`skipped`:::
+(integer) The total number of {ccs} requests (successful or failed) that had at least one remote cluster skipped.
+
+`took`:::
+(object) Contains statistics about the time taken to execute {ccs} requests.
++
+.Properties of `took`
+[%collapsible%open]
+=======
+`max`:::
+(integer) The maximum time taken to execute a {ccs} request, in milliseconds.
+
+`avg`:::
+(integer) The median time taken to execute a {ccs} request, in milliseconds.
+
+`p90`:::
+(integer) The 90th percentile of the time taken to execute {ccs} requests, in milliseconds.
+=======
+
+`took_mrt_true`::
+(object) Contains statistics about the time taken to execute {ccs} requests for which the
+<<ccs-minimize-roundtrips,`ccs_minimize_roundtrips`>> setting was set to `true`.
++
+.Properties of `took_mrt_true`
+[%collapsible%open]
+=======
+`max`:::
+(integer) The maximum time taken to execute a {ccs} request, in milliseconds.
+
+`avg`:::
+(integer) The median time taken to execute a {ccs} request, in milliseconds.
+
+`p90`:::
+(integer) The 90th percentile of the time taken to execute {ccs} requests, in milliseconds.
+=======
+
+`took_mrt_false`::
+(object) Contains statistics about the time taken to execute {ccs} requests for which the
+<<ccs-minimize-roundtrips,`ccs_minimize_roundtrips`>> setting was set to `false`.
++
+.Properties of `took_mrt_false`
+[%collapsible%open]
+=======
+`max`:::
+(integer) The maximum time taken to execute a {ccs} request, in milliseconds.
+
+`avg`:::
+(integer) The median time taken to execute a {ccs} request, in milliseconds.
+
+`p90`:::
+(integer) The 90th percentile of the time taken to execute {ccs} requests, in milliseconds.
+=======
+
+`remotes_per_search_max`::
+(integer) The maximum number of remote clusters that were queried in a single {ccs} request.
+
+`remotes_per_search_avg`::
+(float) The average number of remote clusters that were queried in a single {ccs} request.
+
+`failure_reasons`::
+(object) Contains statistics about the reasons for {ccs} request failures.
+The keys are the failure reason names and the values are the number of requests that failed for that reason.
+
+`features`::
+(object) Contains statistics about the features used in {ccs} requests. The keys are the names of the search feature,
+and the values are the number of requests that used that feature. Single request can use more than one feature
+(e.g. both `async` and `wildcard`). Known features are:
+
+* `async` - <<async-search, Async search>>
+
+* `mrt` - <<ccs-minimize-roundtrips,`ccs_minimize_roundtrips`>> setting was set to `true`.
+
+* `wildcard` - <<api-multi-index,Multi-target syntax>> for indices with wildcards was used in the search request.
+
+`clients`::
+(object) Contains statistics about the clients that executed {ccs} requests.
+The keys are the names of the clients, and the values are the number of requests that were executed by that client.
+Only known clients (such as `kibana` or `elasticsearch`) are counted.
+
+`clusters`::
+(object) Contains statistics about the clusters that were queried in {ccs} requests.
+The keys are cluster names, and the values are per-cluster telemetry data.
+This also includes the local cluster itself, which uses the name `(local)`.
++
+.Properties of per-cluster data:
+[%collapsible%open]
+=======
+`total`:::
+(integer) The total number of successful (not skipped) {ccs} requests that were executed against this cluster.
+This may include requests where partial results were returned, but not requests in which the cluster has been skipped entirely.
+
+`skipped`:::
+(integer) The total number of {ccs} requests for which this cluster was skipped.
+
+`took`:::
+(object) Contains statistics about the time taken to execute requests against this cluster.
++
+.Properties of `took`
+[%collapsible%open]
+========
+`max`:::
+(integer) The maximum time taken to execute a {ccs} request, in milliseconds.
+
+`avg`:::
+(integer) The median time taken to execute a {ccs} request, in milliseconds.
+
+`p90`:::
+(integer) The 90th percentile of the time taken to execute {ccs} requests, in milliseconds.
+========
+
+=======
+
+======
+
+=====
+
 [[cluster-stats-api-example]]
 ==== {api-examples-title}
 
@@ -1607,7 +1743,35 @@ The API returns the following response:
    },
    "repositories": {
      ...
-   }
+   },
+   "ccs": {
+     "_search": {
+        "total": 7,
+        "success": 7,
+        "skipped": 0,
+        "took": {
+            "max": 36,
+            "avg": 20,
+            "p90": 33
+        },
+        "took_mrt_true": {
+            "max": 33,
+            "avg": 15,
+            "p90": 33
+        },
+        "took_mrt_false": {
+            "max": 36,
+            "avg": 26,
+            "p90": 36
+        },
+        "remotes_per_search_max": 3,
+        "remotes_per_search_avg": 2.0,
+        "failure_reasons": { ... },
+        "features": { ... },
+        "clients": { ... },
+        "clusters": { ... }
+     }
+  }
 }
 --------------------------------------------------
 // TESTRESPONSE[s/"plugins": \[[^\]]*\]/"plugins": $body.$_path/]
@@ -1618,10 +1782,15 @@ The API returns the following response:
 // TESTRESPONSE[s/"packaging_types": \[[^\]]*\]/"packaging_types": $body.$_path/]
 // TESTRESPONSE[s/"snapshots": \{[^\}]*\}/"snapshots": $body.$_path/]
 // TESTRESPONSE[s/"repositories": \{[^\}]*\}/"repositories": $body.$_path/]
+// TESTRESPONSE[s/"clusters": \{[^\}]*\}/"clusters": $body.$_path/]
+// TESTRESPONSE[s/"features": \{[^\}]*\}/"features": $body.$_path/]
+// TESTRESPONSE[s/"clients": \{[^\}]*\}/"clients": $body.$_path/]
+// TESTRESPONSE[s/"failure_reasons": \{[^\}]*\}/"failure_reasons": $body.$_path/]
 // TESTRESPONSE[s/"field_types": \[[^\]]*\]/"field_types": $body.$_path/]
 // TESTRESPONSE[s/"runtime_field_types": \[[^\]]*\]/"runtime_field_types": $body.$_path/]
 // TESTRESPONSE[s/"search": \{[^\}]*\}/"search": $body.$_path/]
-// TESTRESPONSE[s/: true|false/: $body.$_path/]
+// TESTRESPONSE[s/"remotes_per_search_avg": [.0-9]+/"remotes_per_search_avg": $body.$_path/]
+// TESTRESPONSE[s/: (true|false)/: $body.$_path/]
 // TESTRESPONSE[s/: (\-)?[0-9]+/: $body.$_path/]
 // TESTRESPONSE[s/: "[^"]*"/: $body.$_path/]
 // These replacements do a few things:
diff --git a/server/src/main/java/org/elasticsearch/TransportVersions.java b/server/src/main/java/org/elasticsearch/TransportVersions.java
@@ -207,6 +207,7 @@ static TransportVersion def(int id) {
     public static final TransportVersion UNASSIGNED_PRIMARY_COUNT_ON_CLUSTER_HEALTH = def(8_737_00_0);
     public static final TransportVersion ESQL_AGGREGATE_EXEC_TRACKS_INTERMEDIATE_ATTRS = def(8_738_00_0);
 
+    public static final TransportVersion CCS_TELEMETRY_STATS = def(8_739_00_0);
     /*
      * STOP! READ THIS FIRST! No, really,
      *        ____ _____ ___  ____  _        ____  _____    _    ____    _____ _   _ ___ ____    _____ ___ ____  ____ _____ _
diff --git a/server/src/main/java/org/elasticsearch/action/admin/cluster/stats/CCSTelemetrySnapshot.java b/server/src/main/java/org/elasticsearch/action/admin/cluster/stats/CCSTelemetrySnapshot.java
@@ -277,7 +277,7 @@ public int hashCode() {
      */
     public void add(CCSTelemetrySnapshot stats) {
         // This should be called in ClusterStatsResponse ctor only, so we don't need to worry about concurrency
-        if (stats.totalCount == 0) {
+        if (stats == null || stats.totalCount == 0) {
             // Just ignore the empty stats.
             // This could happen if the node is brand new or if the stats are not available, e.g. because it runs an old version.
             return;
@@ -315,7 +315,7 @@ public void add(CCSTelemetrySnapshot stats) {
      *      "p90": 2570
      * }
      */
-    public static void publishLatency(XContentBuilder builder, String name, LongMetricValue took) throws IOException {
+    private static void publishLatency(XContentBuilder builder, String name, LongMetricValue took) throws IOException {
         builder.startObject(name);
         {
             builder.field("max", took.max());
diff --git a/server/src/main/java/org/elasticsearch/action/admin/cluster/stats/CCSUsageTelemetry.java b/server/src/main/java/org/elasticsearch/action/admin/cluster/stats/CCSUsageTelemetry.java
@@ -175,7 +175,7 @@ public static class PerClusterCCSTelemetry {
         // The number of successful (not skipped) requests to this cluster.
         private final LongAdder count;
         private final LongAdder skippedCount;
-        // This is only over the successful requetss, skipped ones do not count here.
+        // This is only over the successful requests, skipped ones do not count here.
         private final LongMetric took;
 
         PerClusterCCSTelemetry(String clusterAlias) {
diff --git a/server/src/main/java/org/elasticsearch/action/admin/cluster/stats/ClusterStatsNodeResponse.java b/server/src/main/java/org/elasticsearch/action/admin/cluster/stats/ClusterStatsNodeResponse.java
@@ -30,6 +30,7 @@ public class ClusterStatsNodeResponse extends BaseNodeResponse {
     private final ClusterHealthStatus clusterStatus;
     private final SearchUsageStats searchUsageStats;
     private final RepositoryUsageStats repositoryUsageStats;
+    private final CCSTelemetrySnapshot ccsMetrics;
 
     public ClusterStatsNodeResponse(StreamInput in) throws IOException {
         super(in);
@@ -47,6 +48,11 @@ public ClusterStatsNodeResponse(StreamInput in) throws IOException {
         } else {
             repositoryUsageStats = RepositoryUsageStats.EMPTY;
         }
+        if (in.getTransportVersion().onOrAfter(TransportVersions.CCS_TELEMETRY_STATS)) {
+            ccsMetrics = new CCSTelemetrySnapshot(in);
+        } else {
+            ccsMetrics = new CCSTelemetrySnapshot();
+        }
     }
 
     public ClusterStatsNodeResponse(
@@ -56,7 +62,8 @@ public ClusterStatsNodeResponse(
         NodeStats nodeStats,
         ShardStats[] shardsStats,
         SearchUsageStats searchUsageStats,
-        RepositoryUsageStats repositoryUsageStats
+        RepositoryUsageStats repositoryUsageStats,
+        CCSTelemetrySnapshot ccsTelemetrySnapshot
     ) {
         super(node);
         this.nodeInfo = nodeInfo;
@@ -65,6 +72,7 @@ public ClusterStatsNodeResponse(
         this.clusterStatus = clusterStatus;
         this.searchUsageStats = Objects.requireNonNull(searchUsageStats);
         this.repositoryUsageStats = Objects.requireNonNull(repositoryUsageStats);
+        this.ccsMetrics = ccsTelemetrySnapshot;
     }
 
     public NodeInfo nodeInfo() {
@@ -95,6 +103,10 @@ public RepositoryUsageStats repositoryUsageStats() {
         return repositoryUsageStats;
     }
 
+    public CCSTelemetrySnapshot getCcsMetrics() {
+        return ccsMetrics;
+    }
+
     @Override
     public void writeTo(StreamOutput out) throws IOException {
         super.writeTo(out);
@@ -108,5 +120,9 @@ public void writeTo(StreamOutput out) throws IOException {
         if (out.getTransportVersion().onOrAfter(TransportVersions.REPOSITORIES_TELEMETRY)) {
             repositoryUsageStats.writeTo(out);
         } // else just drop these stats, ok for bwc
+        if (out.getTransportVersion().onOrAfter(TransportVersions.CCS_TELEMETRY_STATS)) {
+            ccsMetrics.writeTo(out);
+        }
     }
+
 }
diff --git a/server/src/main/java/org/elasticsearch/action/admin/cluster/stats/ClusterStatsResponse.java b/server/src/main/java/org/elasticsearch/action/admin/cluster/stats/ClusterStatsResponse.java
@@ -24,13 +24,17 @@
 import java.util.List;
 import java.util.Locale;
 
+import static org.elasticsearch.action.search.TransportSearchAction.CCS_TELEMETRY_FEATURE_FLAG;
+
 public class ClusterStatsResponse extends BaseNodesResponse<ClusterStatsNodeResponse> implements ToXContentFragment {
 
     final ClusterStatsNodes nodesStats;
     final ClusterStatsIndices indicesStats;
     final ClusterHealthStatus status;
     final ClusterSnapshotStats clusterSnapshotStats;
     final RepositoryUsageStats repositoryUsageStats;
+
+    final CCSTelemetrySnapshot ccsMetrics;
     final long timestamp;
     final String clusterUUID;
 
@@ -50,6 +54,7 @@ public ClusterStatsResponse(
         this.timestamp = timestamp;
         nodesStats = new ClusterStatsNodes(nodes);
         indicesStats = new ClusterStatsIndices(nodes, mappingStats, analysisStats, versionStats);
+        ccsMetrics = new CCSTelemetrySnapshot();
         ClusterHealthStatus status = null;
         for (ClusterStatsNodeResponse response : nodes) {
             // only the master node populates the status
@@ -58,6 +63,7 @@ public ClusterStatsResponse(
                 break;
             }
         }
+        nodes.forEach(node -> ccsMetrics.add(node.getCcsMetrics()));
         this.status = status;
         this.clusterSnapshotStats = clusterSnapshotStats;
 
@@ -90,6 +96,10 @@ public ClusterStatsIndices getIndicesStats() {
         return indicesStats;
     }
 
+    public CCSTelemetrySnapshot getCcsMetrics() {
+        return ccsMetrics;
+    }
+
     @Override
     public void writeTo(StreamOutput out) throws IOException {
         TransportAction.localOnly();
@@ -125,6 +135,12 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws
         builder.field("repositories");
         repositoryUsageStats.toXContent(builder, params);
 
+        if (CCS_TELEMETRY_FEATURE_FLAG.isEnabled()) {
+            builder.startObject("ccs");
+            ccsMetrics.toXContent(builder, params);
+            builder.endObject();
+        }
+
         return builder;
     }
 
diff --git a/server/src/main/java/org/elasticsearch/action/admin/cluster/stats/TransportClusterStatsAction.java b/server/src/main/java/org/elasticsearch/action/admin/cluster/stats/TransportClusterStatsAction.java
@@ -81,6 +81,7 @@ public class TransportClusterStatsAction extends TransportNodesAction<
     private final IndicesService indicesService;
     private final RepositoriesService repositoriesService;
     private final SearchUsageHolder searchUsageHolder;
+    private final CCSUsageTelemetry ccsUsageHolder;
 
     private final MetadataStatsCache<MappingStats> mappingStatsCache;
     private final MetadataStatsCache<AnalysisStats> analysisStatsCache;
@@ -108,6 +109,7 @@ public TransportClusterStatsAction(
         this.indicesService = indicesService;
         this.repositoriesService = repositoriesService;
         this.searchUsageHolder = usageService.getSearchUsageHolder();
+        this.ccsUsageHolder = usageService.getCcsUsageHolder();
         this.mappingStatsCache = new MetadataStatsCache<>(threadPool.getThreadContext(), MappingStats::of);
         this.analysisStatsCache = new MetadataStatsCache<>(threadPool.getThreadContext(), AnalysisStats::of);
     }
@@ -249,6 +251,7 @@ protected ClusterStatsNodeResponse nodeOperation(ClusterStatsNodeRequest nodeReq
         final SearchUsageStats searchUsageStats = searchUsageHolder.getSearchUsageStats();
 
         final RepositoryUsageStats repositoryUsageStats = repositoriesService.getUsageStats();
+        final CCSTelemetrySnapshot ccsTelemetry = ccsUsageHolder.getCCSTelemetrySnapshot();
 
         return new ClusterStatsNodeResponse(
             nodeInfo.getNode(),
@@ -257,7 +260,8 @@ protected ClusterStatsNodeResponse nodeOperation(ClusterStatsNodeRequest nodeReq
             nodeStats,
             shardsStats.toArray(new ShardStats[shardsStats.size()]),
             searchUsageStats,
-            repositoryUsageStats
+            repositoryUsageStats,
+            ccsTelemetry
         );
     }
 
diff --git a/server/src/main/java/org/elasticsearch/action/search/TransportSearchAction.java b/server/src/main/java/org/elasticsearch/action/search/TransportSearchAction.java
@@ -127,7 +127,7 @@ public class TransportSearchAction extends HandledTransportAction<SearchRequest,
     public static final String FROZEN_INDICES_DEPRECATION_MESSAGE = "Searching frozen indices [{}] is deprecated."
         + " Consider cold or frozen tiers in place of frozen indices. The frozen feature will be removed in a feature release.";
 
-    private static final FeatureFlag CCS_TELEMETRY_FEATURE_FLAG = new FeatureFlag("ccs_telemetry");
+    public static final FeatureFlag CCS_TELEMETRY_FEATURE_FLAG = new FeatureFlag("ccs_telemetry");
 
     /** The maximum number of shards for a single search request. */
     public static final Setting<Long> SHARD_COUNT_LIMIT_SETTING = Setting.longSetting(
diff --git a/server/src/test/java/org/elasticsearch/action/admin/cluster/stats/VersionStatsTests.java b/server/src/test/java/org/elasticsearch/action/admin/cluster/stats/VersionStatsTests.java
diff --git a/x-pack/plugin/monitoring/src/test/java/org/elasticsearch/xpack/monitoring/collector/cluster/ClusterStatsMonitoringDocTests.java b/x-pack/plugin/monitoring/src/test/java/org/elasticsearch/xpack/monitoring/collector/cluster/ClusterStatsMonitoringDocTests.java
