[Backport 2.x]Added search backpressure stats API

Added search backpressure stats to the existing node/stats API to describe: 1. the number of cancellations (currently for SearchShardTask only) 2. the current state of TaskResourceUsageTracker Signed-off-by: Ketan Verma <ketan9495@gmail.com>
opensearch-project · Nov 3, 2022 · da4b84e · da4b84e
1 parent 72cafde
commit da4b84e
Show file tree

Hide file tree

Showing 26 changed files with 670 additions and 56 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -28,6 +28,7 @@ Inspired from [Keep a Changelog](https://keepachangelog.com/en/1.0.0/)
 - Use getParameterCount instead of getParameterTypes ([#4821](https://github.com/opensearch-project/OpenSearch/pull/4821))
 - Added in-flight cancellation of SearchShardTask based on resource consumption ([#4565](https://github.com/opensearch-project/OpenSearch/pull/4565))
 - Added resource usage trackers for in-flight cancellation of SearchShardTask ([#4805](https://github.com/opensearch-project/OpenSearch/pull/4805))
+- Added search backpressure stats API ([#4932](https://github.com/opensearch-project/OpenSearch/pull/4932))
 
 ### Dependencies
 - Bumps `com.diffplug.spotless` from 6.9.1 to 6.10.0

diff --git a/server/src/main/java/org/opensearch/action/admin/cluster/node/stats/NodeStats.java b/server/src/main/java/org/opensearch/action/admin/cluster/node/stats/NodeStats.java
@@ -56,6 +56,7 @@
 import org.opensearch.node.AdaptiveSelectionStats;
 import org.opensearch.script.ScriptCacheStats;
 import org.opensearch.script.ScriptStats;
+import org.opensearch.search.backpressure.stats.SearchBackpressureStats;
 import org.opensearch.threadpool.ThreadPoolStats;
 import org.opensearch.transport.TransportStats;
 
@@ -119,6 +120,9 @@ public class NodeStats extends BaseNodeResponse implements ToXContentFragment {
     @Nullable
     private ShardIndexingPressureStats shardIndexingPressureStats;
 
+    @Nullable
+    private SearchBackpressureStats searchBackpressureStats;
+
     public NodeStats(StreamInput in) throws IOException {
         super(in);
         timestamp = in.readVLong();
@@ -156,6 +160,11 @@ public NodeStats(StreamInput in) throws IOException {
             shardIndexingPressureStats = null;
         }
 
+        if (in.getVersion().onOrAfter(Version.V_2_4_0)) {
+            searchBackpressureStats = in.readOptionalWriteable(SearchBackpressureStats::new);
+        } else {
+            searchBackpressureStats = null;
+        }
     }
 
     public NodeStats(
@@ -176,7 +185,8 @@ public NodeStats(
         @Nullable AdaptiveSelectionStats adaptiveSelectionStats,
         @Nullable ScriptCacheStats scriptCacheStats,
         @Nullable IndexingPressureStats indexingPressureStats,
-        @Nullable ShardIndexingPressureStats shardIndexingPressureStats
+        @Nullable ShardIndexingPressureStats shardIndexingPressureStats,
+        @Nullable SearchBackpressureStats searchBackpressureStats
     ) {
         super(node);
         this.timestamp = timestamp;
@@ -196,6 +206,7 @@ public NodeStats(
         this.scriptCacheStats = scriptCacheStats;
         this.indexingPressureStats = indexingPressureStats;
         this.shardIndexingPressureStats = shardIndexingPressureStats;
+        this.searchBackpressureStats = searchBackpressureStats;
     }
 
     public long getTimestamp() {
@@ -305,6 +316,11 @@ public ShardIndexingPressureStats getShardIndexingPressureStats() {
         return shardIndexingPressureStats;
     }
 
+    @Nullable
+    public SearchBackpressureStats getSearchBackpressureStats() {
+        return searchBackpressureStats;
+    }
+
     @Override
     public void writeTo(StreamOutput out) throws IOException {
         super.writeTo(out);
@@ -336,6 +352,9 @@ public void writeTo(StreamOutput out) throws IOException {
         if (out.getVersion().onOrAfter(Version.V_1_2_0)) {
             out.writeOptionalWriteable(shardIndexingPressureStats);
         }
+        if (out.getVersion().onOrAfter(Version.V_2_4_0)) {
+            out.writeOptionalWriteable(searchBackpressureStats);
+        }
     }
 
     @Override
@@ -408,6 +427,9 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws
         if (getShardIndexingPressureStats() != null) {
             getShardIndexingPressureStats().toXContent(builder, params);
         }
+        if (getSearchBackpressureStats() != null) {
+            getSearchBackpressureStats().toXContent(builder, params);
+        }
         return builder;
     }
 }
diff --git a/server/src/main/java/org/opensearch/action/admin/cluster/node/stats/NodesStatsRequest.java b/server/src/main/java/org/opensearch/action/admin/cluster/node/stats/NodesStatsRequest.java
@@ -237,7 +237,8 @@ public enum Metric {
         ADAPTIVE_SELECTION("adaptive_selection"),
         SCRIPT_CACHE("script_cache"),
         INDEXING_PRESSURE("indexing_pressure"),
-        SHARD_INDEXING_PRESSURE("shard_indexing_pressure");
+        SHARD_INDEXING_PRESSURE("shard_indexing_pressure"),
+        SEARCH_BACKPRESSURE("search_backpressure");
 
         private String metricName;
 

diff --git a/...c/main/java/org/opensearch/action/admin/cluster/node/stats/TransportNodesStatsAction.java b/...c/main/java/org/opensearch/action/admin/cluster/node/stats/TransportNodesStatsAction.java
@@ -118,7 +118,8 @@ protected NodeStats nodeOperation(NodeStatsRequest nodeStatsRequest) {
             NodesStatsRequest.Metric.ADAPTIVE_SELECTION.containedIn(metrics),
             NodesStatsRequest.Metric.SCRIPT_CACHE.containedIn(metrics),
             NodesStatsRequest.Metric.INDEXING_PRESSURE.containedIn(metrics),
-            NodesStatsRequest.Metric.SHARD_INDEXING_PRESSURE.containedIn(metrics)
+            NodesStatsRequest.Metric.SHARD_INDEXING_PRESSURE.containedIn(metrics),
+            NodesStatsRequest.Metric.SEARCH_BACKPRESSURE.containedIn(metrics)
         );
     }
 

diff --git a/.../src/main/java/org/opensearch/action/admin/cluster/stats/TransportClusterStatsAction.java b/.../src/main/java/org/opensearch/action/admin/cluster/stats/TransportClusterStatsAction.java
@@ -162,6 +162,7 @@ protected ClusterStatsNodeResponse nodeOperation(ClusterStatsNodeRequest nodeReq
             false,
             false,
             false,
+            false,
             false
         );
         List<ShardStats> shardsStats = new ArrayList<>();

diff --git a/server/src/main/java/org/opensearch/common/settings/ClusterSettings.java b/server/src/main/java/org/opensearch/common/settings/ClusterSettings.java
@@ -591,8 +591,7 @@ public void apply(Settings value, Settings current, Settings previous) {
                 TaskManager.TASK_RESOURCE_CONSUMERS_ENABLED,
 
                 // Settings related to search backpressure
-                SearchBackpressureSettings.SETTING_ENABLED,
-                SearchBackpressureSettings.SETTING_ENFORCED,
+                SearchBackpressureSettings.SETTING_MODE,
                 SearchBackpressureSettings.SETTING_CANCELLATION_RATIO,
                 SearchBackpressureSettings.SETTING_CANCELLATION_RATE,
                 SearchBackpressureSettings.SETTING_CANCELLATION_BURST,

diff --git a/server/src/main/java/org/opensearch/node/NodeService.java b/server/src/main/java/org/opensearch/node/NodeService.java
@@ -173,7 +173,8 @@ public NodeStats stats(
         boolean adaptiveSelection,
         boolean scriptCache,
         boolean indexingPressure,
-        boolean shardIndexingPressure
+        boolean shardIndexingPressure,
+        boolean searchBackpressure
     ) {
         // for indices stats we want to include previous allocated shards stats as well (it will
         // only be applied to the sensible ones to use, like refresh/merge/flush/indexing stats)
@@ -195,7 +196,8 @@ public NodeStats stats(
             adaptiveSelection ? responseCollectorService.getAdaptiveStats(searchTransportService.getPendingSearchRequests()) : null,
             scriptCache ? scriptService.cacheStats() : null,
             indexingPressure ? this.indexingPressureService.nodeStats() : null,
-            shardIndexingPressure ? this.indexingPressureService.shardStats(indices) : null
+            shardIndexingPressure ? this.indexingPressureService.shardStats(indices) : null,
+            searchBackpressure ? this.searchBackpressureService.nodeStats() : null
         );
     }
 

diff --git a/server/src/main/java/org/opensearch/search/backpressure/SearchBackpressureService.java b/server/src/main/java/org/opensearch/search/backpressure/SearchBackpressureService.java
@@ -16,12 +16,16 @@
 import org.opensearch.common.util.TokenBucket;
 import org.opensearch.monitor.jvm.JvmStats;
 import org.opensearch.monitor.process.ProcessProbe;
+import org.opensearch.search.backpressure.settings.SearchBackpressureMode;
 import org.opensearch.search.backpressure.settings.SearchBackpressureSettings;
+import org.opensearch.search.backpressure.stats.SearchBackpressureStats;
+import org.opensearch.search.backpressure.stats.SearchShardTaskStats;
 import org.opensearch.search.backpressure.trackers.CpuUsageTracker;
 import org.opensearch.search.backpressure.trackers.ElapsedTimeTracker;
 import org.opensearch.search.backpressure.trackers.HeapUsageTracker;
 import org.opensearch.search.backpressure.trackers.NodeDuressTracker;
 import org.opensearch.search.backpressure.trackers.TaskResourceUsageTracker;
+import org.opensearch.search.backpressure.trackers.TaskResourceUsageTrackerType;
 import org.opensearch.tasks.CancellableTask;
 import org.opensearch.tasks.Task;
 import org.opensearch.tasks.TaskCancellation;
@@ -117,7 +121,8 @@ public SearchBackpressureService(
     }
 
     void doRun() {
-        if (getSettings().isEnabled() == false) {
+        SearchBackpressureMode mode = getSettings().getMode();
+        if (mode == SearchBackpressureMode.DISABLED) {
             return;
         }
 
@@ -126,7 +131,7 @@ void doRun() {
         }
 
         // We are only targeting in-flight cancellation of SearchShardTask for now.
-        List<CancellableTask> searchShardTasks = getSearchShardTasks();
+        List<SearchShardTask> searchShardTasks = getSearchShardTasks();
 
         // Force-refresh usage stats of these tasks before making a cancellation decision.
         taskResourceTrackingService.refreshResourceStats(searchShardTasks.toArray(new Task[0]));
@@ -138,12 +143,13 @@ void doRun() {
 
         for (TaskCancellation taskCancellation : getTaskCancellations(searchShardTasks)) {
             logger.debug(
-                "cancelling task [{}] due to high resource consumption [{}]",
+                "[{} mode] cancelling task [{}] due to high resource consumption [{}]",
+                mode.getName(),
                 taskCancellation.getTask().getId(),
                 taskCancellation.getReasonString()
             );
 
-            if (getSettings().isEnforced() == false) {
+            if (mode != SearchBackpressureMode.ENFORCED) {
                 continue;
             }
 
@@ -159,7 +165,6 @@ void doRun() {
             }
 
             taskCancellation.cancel();
-            state.incrementCancellationCount();
         }
     }
 
@@ -182,7 +187,7 @@ boolean isNodeInDuress() {
     /**
      * Returns true if the increase in heap usage is due to search requests.
      */
-    boolean isHeapUsageDominatedBySearch(List<CancellableTask> searchShardTasks) {
+    boolean isHeapUsageDominatedBySearch(List<SearchShardTask> searchShardTasks) {
         long usage = searchShardTasks.stream().mapToLong(task -> task.getTotalResourceStats().getMemoryInBytes()).sum();
         long threshold = getSettings().getSearchShardTaskSettings().getTotalHeapBytesThreshold();
         if (usage < threshold) {
@@ -196,12 +201,12 @@ boolean isHeapUsageDominatedBySearch(List<CancellableTask> searchShardTasks) {
     /**
      * Filters and returns the list of currently running SearchShardTasks.
      */
-    List<CancellableTask> getSearchShardTasks() {
+    List<SearchShardTask> getSearchShardTasks() {
         return taskResourceTrackingService.getResourceAwareTasks()
             .values()
             .stream()
             .filter(task -> task instanceof SearchShardTask)
-            .map(task -> (CancellableTask) task)
+            .map(task -> (SearchShardTask) task)
             .collect(Collectors.toUnmodifiableList());
     }
 
@@ -222,13 +227,17 @@ TaskCancellation getTaskCancellation(CancellableTask task) {
             }
         }
 
+        if (task instanceof SearchShardTask) {
+            callbacks.add(state::incrementCancellationCount);
+        }
+
         return new TaskCancellation(task, reasons, callbacks);
     }
 
     /**
      * Returns a list of TaskCancellations sorted by descending order of their cancellation scores.
      */
-    List<TaskCancellation> getTaskCancellations(List<CancellableTask> tasks) {
+    List<TaskCancellation> getTaskCancellations(List<? extends CancellableTask> tasks) {
         return tasks.stream()
             .map(this::getTaskCancellation)
             .filter(TaskCancellation::isEligibleForCancellation)
@@ -246,7 +255,7 @@ SearchBackpressureState getState() {
 
     @Override
     public void onTaskCompleted(Task task) {
-        if (getSettings().isEnabled() == false) {
+        if (getSettings().getMode() == SearchBackpressureMode.DISABLED) {
             return;
         }
 
@@ -310,4 +319,17 @@ protected void doStop() {
 
     @Override
     protected void doClose() throws IOException {}
+
+    public SearchBackpressureStats nodeStats() {
+        List<SearchShardTask> searchShardTasks = getSearchShardTasks();
+
+        SearchShardTaskStats searchShardTaskStats = new SearchShardTaskStats(
+            state.getCancellationCount(),
+            state.getLimitReachedCount(),
+            taskResourceUsageTrackers.stream()
+                .collect(Collectors.toUnmodifiableMap(t -> TaskResourceUsageTrackerType.fromName(t.name()), t -> t.stats(searchShardTasks)))
+        );
+
+        return new SearchBackpressureStats(searchShardTaskStats, getSettings().getMode());
+    }
 }
diff --git a/server/src/main/java/org/opensearch/search/backpressure/settings/SearchBackpressureMode.java b/server/src/main/java/org/opensearch/search/backpressure/settings/SearchBackpressureMode.java
@@ -0,0 +1,52 @@
+/*
+ * SPDX-License-Identifier: Apache-2.0
+ *
+ * The OpenSearch Contributors require contributions made to
+ * this file be licensed under the Apache-2.0 license or a
+ * compatible open source license.
+ */
+
+package org.opensearch.search.backpressure.settings;
+
+/**
+ * Defines the search backpressure mode.
+ */
+public enum SearchBackpressureMode {
+    /**
+     * SearchBackpressureService is completely disabled.
+     */
+    DISABLED("disabled"),
+
+    /**
+     * SearchBackpressureService only monitors the resource usage of running tasks.
+     */
+    MONITOR_ONLY("monitor_only"),
+
+    /**
+     * SearchBackpressureService monitors and rejects tasks that exceed resource usage thresholds.
+     */
+    ENFORCED("enforced");
+
+    private final String name;
+
+    SearchBackpressureMode(String name) {
+        this.name = name;
+    }
+
+    public String getName() {
+        return name;
+    }
+
+    public static SearchBackpressureMode fromName(String name) {
+        switch (name) {
+            case "disabled":
+                return DISABLED;
+            case "monitor_only":
+                return MONITOR_ONLY;
+            case "enforced":
+                return ENFORCED;
+        }
+
+        throw new IllegalArgumentException("Invalid SearchBackpressureMode: " + name);
+    }
+}