Add threadpool wait time metric (opensearch-project#9681)

Signed-off-by: Jay Deng <[email protected]>
kaushalmahi12 · Sep 12, 2023 · 1aad5a3 · 1aad5a3
1 parent 16b54a2
commit 1aad5a3
Show file tree

Hide file tree

Showing 12 changed files with 166 additions and 17 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -77,6 +77,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 
 ## [Unreleased 2.x]
 ### Added
+- Add metrics for thread_pool task wait time ([#9681](https://github.com/opensearch-project/OpenSearch/pull/9681))
 
 ### Dependencies
 - Bump `peter-evans/create-or-update-comment` from 2 to 3 ([#9575](https://github.com/opensearch-project/OpenSearch/pull/9575))

diff --git a/rest-api-spec/src/main/resources/rest-api-spec/test/cat.thread_pool/10_basic.yml b/rest-api-spec/src/main/resources/rest-api-spec/test/cat.thread_pool/10_basic.yml
@@ -1,3 +1,30 @@
+"Test cat thread_pool total_wait_time output":
+  - skip:
+      version: " - 3.0.0"
+      reason: thread_pool total_wait_time stats were introduced in V_3.0.0
+
+  - do:
+      cat.thread_pool: {}
+
+  - match:
+      $body: |
+        /  #node_name     name     active     queue     rejected
+        ^  (\S+       \s+ \S+  \s+ \d+    \s+ \d+   \s+ \d+      \n)+  $/
+
+  - do:
+      cat.thread_pool:
+        thread_pool_patterns: search,search_throttled,index_searcher,generic
+        h: name,total_wait_time,twt
+        v: true
+
+  - match:
+      $body: |
+        /^  id  \s+ name             \s+ total_wait_time    \s+ twt \n
+           (\S+ \s+ search           \s+ \d+s               \s+ \d+ \n
+            \S+ \s+ search_throttled \s+ \d+s               \s+ \d+ \n
+            \S+ \s+ index_searcher   \s+ \d+s               \s+ \d+ \n
+            \S+ \s+ generic          \s+ -1                 \s+ -1  \n)+  $/
+
 ---
 "Test cat thread_pool output":
   - skip:

diff --git a/server/src/internalClusterTest/java/org/opensearch/search/stats/ConcurrentSearchStatsIT.java b/server/src/internalClusterTest/java/org/opensearch/search/stats/ConcurrentSearchStatsIT.java
@@ -8,6 +8,8 @@
 
 package org.opensearch.search.stats;
 
+import org.opensearch.action.admin.cluster.node.stats.NodesStatsAction;
+import org.opensearch.action.admin.cluster.node.stats.NodesStatsRequestBuilder;
 import org.opensearch.action.admin.cluster.node.stats.NodesStatsResponse;
 import org.opensearch.action.admin.indices.stats.IndexStats;
 import org.opensearch.action.admin.indices.stats.IndicesStatsRequestBuilder;
@@ -26,6 +28,8 @@
 import org.opensearch.search.SearchService;
 import org.opensearch.test.InternalSettingsPlugin;
 import org.opensearch.test.OpenSearchIntegTestCase;
+import org.opensearch.threadpool.ThreadPool;
+import org.opensearch.threadpool.ThreadPoolStats;
 
 import java.util.Arrays;
 import java.util.Collection;
@@ -35,6 +39,7 @@
 import java.util.function.Function;
 
 import static org.opensearch.index.query.QueryBuilders.scriptQuery;
+import static org.opensearch.search.SearchService.CONCURRENT_SEGMENT_SEARCH_TARGET_MAX_SLICE_COUNT_SETTING;
 import static org.hamcrest.Matchers.greaterThan;
 import static org.hamcrest.Matchers.lessThan;
 
@@ -307,6 +312,54 @@ public void testAvgConcurrencyIndexLevel() throws InterruptedException {
         assertEquals(expectedConcurrency, stats.getTotal().getSearch().getTotal().getConcurrentAvgSliceCount(), 0);
     }
 
+    public void testThreadPoolWaitTime() throws Exception {
+        int NUM_SHARDS = 1;
+        String INDEX = "test-" + randomAlphaOfLength(5).toLowerCase(Locale.ROOT);
+        createIndex(
+            INDEX,
+            Settings.builder()
+                .put(indexSettings())
+                .put(IndexMetadata.SETTING_NUMBER_OF_SHARDS, NUM_SHARDS)
+                .put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, 0)
+                .build()
+        );
+
+        ensureGreen();
+
+        for (int i = 0; i < 10; i++) {
+            client().prepareIndex(INDEX).setId(Integer.toString(i)).setSource("field", "value" + i).get();
+            refresh();
+        }
+        client().admin()
+            .cluster()
+            .prepareUpdateSettings()
+            .setTransientSettings(Settings.builder().put(CONCURRENT_SEGMENT_SEARCH_TARGET_MAX_SLICE_COUNT_SETTING.getKey(), 10))
+            .execute()
+            .actionGet();
+
+        client().prepareSearch(INDEX).execute().actionGet();
+
+        NodesStatsRequestBuilder nodesStatsRequestBuilder = new NodesStatsRequestBuilder(
+            client().admin().cluster(),
+            NodesStatsAction.INSTANCE
+        ).setNodesIds().all();
+        NodesStatsResponse nodesStatsResponse = nodesStatsRequestBuilder.execute().actionGet();
+        ThreadPoolStats threadPoolStats = nodesStatsResponse.getNodes().get(0).getThreadPool();
+
+        for (ThreadPoolStats.Stats stats : threadPoolStats) {
+            if (stats.getName().equals(ThreadPool.Names.INDEX_SEARCHER)) {
+                assertThat(stats.getWaitTime().micros(), greaterThan(0L));
+            }
+        }
+
+        client().admin()
+            .cluster()
+            .prepareUpdateSettings()
+            .setTransientSettings(Settings.builder().put(CONCURRENT_SEGMENT_SEARCH_TARGET_MAX_SLICE_COUNT_SETTING.getKey(), 2))
+            .execute()
+            .actionGet();
+    }
+
     public static class ScriptedDelayedPlugin extends MockScriptPlugin {
         static final String SCRIPT_NAME = "search_timeout";
 

diff --git a/server/src/main/java/org/opensearch/common/util/concurrent/OpenSearchThreadPoolExecutor.java b/server/src/main/java/org/opensearch/common/util/concurrent/OpenSearchThreadPoolExecutor.java
@@ -205,4 +205,15 @@ protected Runnable wrapRunnable(Runnable command) {
     protected Runnable unwrap(Runnable runnable) {
         return contextHolder.unwrap(runnable);
     }
+
+    /**
+     * Returns the cumulative wait time of the ThreadPool. If the ThreadPool does not support tracking the cumulative pool wait time
+     * then this should return -1 which will prevent the value from showing up in {@link org.opensearch.threadpool.ThreadPoolStats}.
+     * ThreadPools that do support this metric should override this method. For example, {@link QueueResizingOpenSearchThreadPoolExecutor}
+     * does so using the {@link TimedRunnable} to get the difference between Runnable creation and execution.
+     *
+     */
+    public long getPoolWaitTimeNanos() {
+        return -1;
+    }
 }
diff --git a/...ava/org/opensearch/common/util/concurrent/QueueResizableOpenSearchThreadPoolExecutor.java b/...ava/org/opensearch/common/util/concurrent/QueueResizableOpenSearchThreadPoolExecutor.java
@@ -9,6 +9,7 @@
 package org.opensearch.common.util.concurrent;
 
 import org.opensearch.common.ExponentiallyWeightedMovingAverage;
+import org.opensearch.common.metrics.CounterMetric;
 
 import java.util.concurrent.ThreadFactory;
 import java.util.concurrent.TimeUnit;
@@ -27,6 +28,7 @@ public final class QueueResizableOpenSearchThreadPoolExecutor extends OpenSearch
     private final ResizableBlockingQueue<Runnable> workQueue;
     private final Function<Runnable, WrappedRunnable> runnableWrapper;
     private final ExponentiallyWeightedMovingAverage executionEWMA;
+    private final CounterMetric poolWaitTime;
 
     /**
      * Create new resizable at runtime thread pool executor
@@ -101,6 +103,7 @@ public final class QueueResizableOpenSearchThreadPoolExecutor extends OpenSearch
         this.workQueue = workQueue;
         this.runnableWrapper = runnableWrapper;
         this.executionEWMA = new ExponentiallyWeightedMovingAverage(ewmaAlpha, 0);
+        this.poolWaitTime = new CounterMetric();
     }
 
     @Override
@@ -156,6 +159,7 @@ protected void afterExecute(Runnable r, Throwable t) {
             // taskExecutionNanos may be -1 if the task threw an exception
             executionEWMA.addValue(taskExecutionNanos);
         }
+        poolWaitTime.inc(timedRunnable.getWaitTimeNanos());
     }
 
     /**
@@ -173,4 +177,9 @@ public synchronized int resize(int capacity) {
             capacity
         );
     }
+
+    @Override
+    public long getPoolWaitTimeNanos() {
+        return poolWaitTime.count();
+    }
 }
diff --git a/...java/org/opensearch/common/util/concurrent/QueueResizingOpenSearchThreadPoolExecutor.java b/...java/org/opensearch/common/util/concurrent/QueueResizingOpenSearchThreadPoolExecutor.java
@@ -36,6 +36,7 @@
 import org.apache.logging.log4j.Logger;
 import org.apache.logging.log4j.message.ParameterizedMessage;
 import org.opensearch.common.ExponentiallyWeightedMovingAverage;
+import org.opensearch.common.metrics.CounterMetric;
 import org.opensearch.common.unit.TimeValue;
 
 import java.util.Locale;
@@ -66,6 +67,7 @@ public final class QueueResizingOpenSearchThreadPoolExecutor extends OpenSearchT
     private final int maxQueueSize;
     private final long targetedResponseTimeNanos;
     private final ExponentiallyWeightedMovingAverage executionEWMA;
+    private final CounterMetric poolWaitTime;
 
     private final AtomicLong totalTaskNanos = new AtomicLong(0);
     private final AtomicInteger taskCount = new AtomicInteger(0);
@@ -97,6 +99,7 @@ public final class QueueResizingOpenSearchThreadPoolExecutor extends OpenSearchT
         this.maxQueueSize = maxQueueSize;
         this.targetedResponseTimeNanos = targetedResponseTime.getNanos();
         this.executionEWMA = new ExponentiallyWeightedMovingAverage(EWMA_ALPHA, 0);
+        this.poolWaitTime = new CounterMetric();
         logger.debug(
             "thread pool [{}] will adjust queue by [{}] when determining automatic queue size",
             getName(),
@@ -190,6 +193,7 @@ protected void afterExecute(Runnable r, Throwable t) {
             // taskExecutionNanos may be -1 if the task threw an exception
             executionEWMA.addValue(taskExecutionNanos);
         }
+        poolWaitTime.inc(timedRunnable.getWaitTimeNanos());
 
         if (taskCount.incrementAndGet() == this.tasksPerFrame) {
             final long endTimeNs = System.nanoTime();
@@ -290,4 +294,8 @@ protected void appendThreadPoolExecutorDetails(StringBuilder sb) {
         sb.append("adjustment amount = ").append(QUEUE_ADJUSTMENT_AMOUNT).append(", ");
     }
 
+    @Override
+    public long getPoolWaitTimeNanos() {
+        return poolWaitTime.count();
+    }
 }
diff --git a/server/src/main/java/org/opensearch/common/util/concurrent/TimedRunnable.java b/server/src/main/java/org/opensearch/common/util/concurrent/TimedRunnable.java
@@ -107,6 +107,14 @@ long getTotalExecutionNanos() {
         return Math.max(finishTimeNanos - startTimeNanos, 1);
     }
 
+    long getWaitTimeNanos() {
+        if (startTimeNanos == -1) {
+            // There must have been an exception thrown, the total time is unknown (-1)
+            return -1;
+        }
+        return Math.max(startTimeNanos - creationTimeNanos, 1);
+    }
+
     /**
      * If the task was failed or rejected, return true.
      * Otherwise, false.

diff --git a/server/src/main/java/org/opensearch/rest/action/cat/RestThreadPoolAction.java b/server/src/main/java/org/opensearch/rest/action/cat/RestThreadPoolAction.java
@@ -163,6 +163,10 @@ protected Table getTableWithHeader(final RestRequest request) {
         table.addCell("rejected", "alias:r;default:true;text-align:right;desc:number of rejected tasks");
         table.addCell("largest", "alias:l;default:false;text-align:right;desc:highest number of seen active threads");
         table.addCell("completed", "alias:c;default:false;text-align:right;desc:number of completed tasks");
+        table.addCell(
+            "total_wait_time",
+            "alias:twt;default:false;text-align:right;desc:total time tasks spent waiting in thread_pool queue"
+        );
         table.addCell("core", "alias:cr;default:false;text-align:right;desc:core number of threads in a scaling thread pool");
         table.addCell("max", "alias:mx;default:false;text-align:right;desc:maximum number of threads in a scaling thread pool");
         table.addCell("size", "alias:sz;default:false;text-align:right;desc:number of threads in a fixed thread pool");
@@ -267,6 +271,7 @@ private Table buildTable(RestRequest req, ClusterStateResponse state, NodesInfoR
                 table.addCell(poolStats == null ? null : poolStats.getRejected());
                 table.addCell(poolStats == null ? null : poolStats.getLargest());
                 table.addCell(poolStats == null ? null : poolStats.getCompleted());
+                table.addCell(poolStats == null ? null : poolStats.getWaitTime());
                 table.addCell(core);
                 table.addCell(max);
                 table.addCell(size);

diff --git a/server/src/main/java/org/opensearch/threadpool/ThreadPool.java b/server/src/main/java/org/opensearch/threadpool/ThreadPool.java
@@ -383,19 +383,21 @@ public ThreadPoolStats stats() {
             long rejected = -1;
             int largest = -1;
             long completed = -1;
-            if (holder.executor() instanceof ThreadPoolExecutor) {
-                ThreadPoolExecutor threadPoolExecutor = (ThreadPoolExecutor) holder.executor();
+            long waitTimeNanos = -1;
+            if (holder.executor() instanceof OpenSearchThreadPoolExecutor) {
+                OpenSearchThreadPoolExecutor threadPoolExecutor = (OpenSearchThreadPoolExecutor) holder.executor();
                 threads = threadPoolExecutor.getPoolSize();
                 queue = threadPoolExecutor.getQueue().size();
                 active = threadPoolExecutor.getActiveCount();
                 largest = threadPoolExecutor.getLargestPoolSize();
                 completed = threadPoolExecutor.getCompletedTaskCount();
+                waitTimeNanos = threadPoolExecutor.getPoolWaitTimeNanos();
                 RejectedExecutionHandler rejectedExecutionHandler = threadPoolExecutor.getRejectedExecutionHandler();
                 if (rejectedExecutionHandler instanceof XRejectedExecutionHandler) {
                     rejected = ((XRejectedExecutionHandler) rejectedExecutionHandler).rejected();
                 }
             }
-            stats.add(new ThreadPoolStats.Stats(name, threads, queue, active, rejected, largest, completed));
+            stats.add(new ThreadPoolStats.Stats(name, threads, queue, active, rejected, largest, completed, waitTimeNanos));
         }
         return new ThreadPoolStats(stats);
     }

diff --git a/server/src/main/java/org/opensearch/threadpool/ThreadPoolStats.java b/server/src/main/java/org/opensearch/threadpool/ThreadPoolStats.java
@@ -32,6 +32,8 @@
 
 package org.opensearch.threadpool;
 
+import org.opensearch.Version;
+import org.opensearch.common.unit.TimeValue;
 import org.opensearch.core.common.io.stream.StreamInput;
 import org.opensearch.core.common.io.stream.StreamOutput;
 import org.opensearch.core.common.io.stream.Writeable;
@@ -65,15 +67,17 @@ public static class Stats implements Writeable, ToXContentFragment, Comparable<S
         private final long rejected;
         private final int largest;
         private final long completed;
+        private final long waitTimeNanos;
 
-        public Stats(String name, int threads, int queue, int active, long rejected, int largest, long completed) {
+        public Stats(String name, int threads, int queue, int active, long rejected, int largest, long completed, long waitTimeNanos) {
             this.name = name;
             this.threads = threads;
             this.queue = queue;
             this.active = active;
             this.rejected = rejected;
             this.largest = largest;
             this.completed = completed;
+            this.waitTimeNanos = waitTimeNanos;
         }
 
         public Stats(StreamInput in) throws IOException {
@@ -84,6 +88,7 @@ public Stats(StreamInput in) throws IOException {
             rejected = in.readLong();
             largest = in.readInt();
             completed = in.readLong();
+            waitTimeNanos = in.getVersion().onOrAfter(Version.V_3_0_0) ? in.readLong() : -1;
         }
 
         @Override
@@ -95,6 +100,9 @@ public void writeTo(StreamOutput out) throws IOException {
             out.writeLong(rejected);
             out.writeInt(largest);
             out.writeLong(completed);
+            if (out.getVersion().onOrAfter(Version.V_3_0_0)) {
+                out.writeLong(waitTimeNanos);
+            }
         }
 
         public String getName() {
@@ -125,6 +133,14 @@ public long getCompleted() {
             return this.completed;
         }
 
+        public TimeValue getWaitTime() {
+            return TimeValue.timeValueNanos(waitTimeNanos);
+        }
+
+        public long getWaitTimeNanos() {
+            return waitTimeNanos;
+        }
+
         @Override
         public XContentBuilder toXContent(XContentBuilder builder, Params params) throws IOException {
             builder.startObject(name);
@@ -146,6 +162,12 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws
             if (completed != -1) {
                 builder.field(Fields.COMPLETED, completed);
             }
+            if (waitTimeNanos != -1) {
+                if (builder.humanReadable()) {
+                    builder.field(Fields.WAIT_TIME, getWaitTime());
+                }
+                builder.field(Fields.WAIT_TIME_NANOS, getWaitTimeNanos());
+            }
             builder.endObject();
             return builder;
         }
@@ -197,6 +219,8 @@ static final class Fields {
         static final String REJECTED = "rejected";
         static final String LARGEST = "largest";
         static final String COMPLETED = "completed";
+        static final String WAIT_TIME = "total_wait_time";
+        static final String WAIT_TIME_NANOS = "total_wait_time_in_nanos";
     }
 
     @Override

diff --git a/server/src/test/java/org/opensearch/action/admin/cluster/node/stats/NodeStatsTests.java b/server/src/test/java/org/opensearch/action/admin/cluster/node/stats/NodeStatsTests.java
@@ -602,7 +602,8 @@ public static NodeStats createNodeStats(boolean remoteStoreStats) {
                         randomIntBetween(1, 1000),
                         randomNonNegativeLong(),
                         randomIntBetween(1, 1000),
-                        randomIntBetween(1, 1000)
+                        randomIntBetween(1, 1000),
+                        randomIntBetween(-1, 10)
                     )
                 );
             }