Expose AWS HTTP Client stats via JMX

trinodb · Jan 5, 2021 · 39a9945 · 39a9945
1 parent 2161800
commit 39a9945
Show file tree

Hide file tree

Showing 4 changed files with 120 additions and 0 deletions.
diff --git a/plugin/trino-hive/src/main/java/io/trino/plugin/hive/aws/AbstractSdkMetricsCollector.java b/plugin/trino-hive/src/main/java/io/trino/plugin/hive/aws/AbstractSdkMetricsCollector.java
@@ -24,6 +24,9 @@
 import java.util.function.Consumer;
 
 import static com.amazonaws.util.AWSRequestMetrics.Field.ClientExecuteTime;
+import static com.amazonaws.util.AWSRequestMetrics.Field.HttpClientPoolAvailableCount;
+import static com.amazonaws.util.AWSRequestMetrics.Field.HttpClientPoolLeasedCount;
+import static com.amazonaws.util.AWSRequestMetrics.Field.HttpClientPoolPendingCount;
 import static com.amazonaws.util.AWSRequestMetrics.Field.HttpClientRetryCount;
 import static com.amazonaws.util.AWSRequestMetrics.Field.HttpRequestTime;
 import static com.amazonaws.util.AWSRequestMetrics.Field.RequestCount;
@@ -54,6 +57,21 @@ public final void collectMetrics(Request<?> request, Response<?> response)
             recordThrottleExceptionCount(throttleExceptions.longValue());
         }
 
+        Number httpClientPoolAvailableCount = timingInfo.getCounter(HttpClientPoolAvailableCount.name());
+        if (httpClientPoolAvailableCount != null) {
+            recordHttpClientPoolAvailableCount(httpClientPoolAvailableCount.longValue());
+        }
+
+        Number httpClientPoolLeasedCount = timingInfo.getCounter(HttpClientPoolLeasedCount.name());
+        if (httpClientPoolLeasedCount != null) {
+            recordHttpClientPoolLeasedCount(httpClientPoolLeasedCount.longValue());
+        }
+
+        Number httpClientPoolPendingCount = timingInfo.getCounter(HttpClientPoolPendingCount.name());
+        if (httpClientPoolPendingCount != null) {
+            recordHttpClientPoolPendingCount(httpClientPoolPendingCount.longValue());
+        }
+
         recordSubTimingDurations(timingInfo, HttpRequestTime, this::recordHttpRequestTime);
         recordSubTimingDurations(timingInfo, ClientExecuteTime, this::recordClientExecutionTime);
         recordSubTimingDurations(timingInfo, RetryPauseTime, this::recordRetryPauseTime);
@@ -71,6 +89,12 @@ public final void collectMetrics(Request<?> request, Response<?> response)
 
     protected abstract void recordRetryPauseTime(Duration duration);
 
+    protected abstract void recordHttpClientPoolAvailableCount(long count);
+
+    protected abstract void recordHttpClientPoolLeasedCount(long count);
+
+    protected abstract void recordHttpClientPoolPendingCount(long count);
+
     private static void recordSubTimingDurations(TimingInfo timingInfo, AWSRequestMetrics.Field field, Consumer<Duration> consumer)
     {
         List<TimingInfo> subTimings = timingInfo.getAllSubMeasurements(field.name());

diff --git a/plugin/trino-hive/src/main/java/io/trino/plugin/hive/metastore/glue/GlueMetastoreStats.java b/plugin/trino-hive/src/main/java/io/trino/plugin/hive/metastore/glue/GlueMetastoreStats.java
@@ -20,6 +20,8 @@
 import org.weakref.jmx.Managed;
 import org.weakref.jmx.Nested;
 
+import java.util.concurrent.atomic.AtomicLong;
+
 import static java.util.Objects.requireNonNull;
 import static java.util.concurrent.TimeUnit.MILLISECONDS;
 
@@ -51,6 +53,9 @@ public class GlueMetastoreStats
     private final TimeStat awsRequestTime = new TimeStat(MILLISECONDS);
     private final TimeStat awsClientExecuteTime = new TimeStat(MILLISECONDS);
     private final TimeStat awsClientRetryPauseTime = new TimeStat(MILLISECONDS);
+    private final AtomicLong awsHttpClientPoolAvailableCount = new AtomicLong();
+    private final AtomicLong awsHttpClientPoolLeasedCount = new AtomicLong();
+    private final AtomicLong awsHttpClientPoolPendingCount = new AtomicLong();
 
     @Managed
     @Nested
@@ -220,6 +225,24 @@ public TimeStat getAwsClientRetryPauseTime()
         return awsClientRetryPauseTime;
     }
 
+    @Managed
+    public long getAwsHttpClientPoolAvailableCount()
+    {
+        return awsHttpClientPoolAvailableCount.get();
+    }
+
+    @Managed
+    public long getAwsHttpClientPoolLeasedCount()
+    {
+        return awsHttpClientPoolLeasedCount.get();
+    }
+
+    @Managed
+    public long getAwsHttpClientPoolPendingCount()
+    {
+        return awsHttpClientPoolPendingCount.get();
+    }
+
     public GlueSdkClientMetricsCollector newRequestMetricsCollector()
     {
         return new GlueSdkClientMetricsCollector(this);
@@ -270,5 +293,23 @@ protected void recordRetryPauseTime(Duration duration)
         {
             stats.awsClientRetryPauseTime.add(duration);
         }
+
+        @Override
+        protected void recordHttpClientPoolAvailableCount(long count)
+        {
+            stats.awsHttpClientPoolAvailableCount.set(count);
+        }
+
+        @Override
+        protected void recordHttpClientPoolLeasedCount(long count)
+        {
+            stats.awsHttpClientPoolLeasedCount.set(count);
+        }
+
+        @Override
+        protected void recordHttpClientPoolPendingCount(long count)
+        {
+            stats.awsHttpClientPoolPendingCount.set(count);
+        }
     }
 }
diff --git a/...in/trino-hive/src/main/java/io/trino/plugin/hive/s3/TrinoS3FileSystemMetricCollector.java b/...in/trino-hive/src/main/java/io/trino/plugin/hive/s3/TrinoS3FileSystemMetricCollector.java
@@ -63,4 +63,22 @@ protected void recordRetryPauseTime(Duration duration)
     {
         stats.addAwsClientRetryPauseTime(duration);
     }
+
+    @Override
+    protected void recordHttpClientPoolAvailableCount(long count)
+    {
+        stats.setAwsHttpClientPoolAvailableCount(count);
+    }
+
+    @Override
+    protected void recordHttpClientPoolLeasedCount(long count)
+    {
+        stats.setAwsHttpClientPoolLeasedCount(count);
+    }
+
+    @Override
+    protected void recordHttpClientPoolPendingCount(long count)
+    {
+        stats.setAwsHttpClientPoolPendingCount(count);
+    }
 }
diff --git a/plugin/trino-hive/src/main/java/io/trino/plugin/hive/s3/TrinoS3FileSystemStats.java b/plugin/trino-hive/src/main/java/io/trino/plugin/hive/s3/TrinoS3FileSystemStats.java
@@ -22,6 +22,7 @@
 
 import java.net.SocketException;
 import java.net.SocketTimeoutException;
+import java.util.concurrent.atomic.AtomicLong;
 
 import static java.util.concurrent.TimeUnit.MILLISECONDS;
 
@@ -52,6 +53,9 @@ public class TrinoS3FileSystemStats
     private final TimeStat awsRequestTime = new TimeStat(MILLISECONDS);
     private final TimeStat awsClientExecuteTime = new TimeStat(MILLISECONDS);
     private final TimeStat awsClientRetryPauseTime = new TimeStat(MILLISECONDS);
+    private final AtomicLong awsHttpClientPoolAvailableCount = new AtomicLong();
+    private final AtomicLong awsHttpClientPoolLeasedCount = new AtomicLong();
+    private final AtomicLong awsHttpClientPoolPendingCount = new AtomicLong();
 
     @Managed
     @Nested
@@ -193,6 +197,24 @@ public TimeStat getAwsClientRetryPauseTime()
         return awsClientRetryPauseTime;
     }
 
+    @Managed
+    public long getAwsHttpClientPoolAvailableCount()
+    {
+        return awsHttpClientPoolAvailableCount.get();
+    }
+
+    @Managed
+    public long getAwsHttpClientPoolLeasedCount()
+    {
+        return awsHttpClientPoolLeasedCount.get();
+    }
+
+    @Managed
+    public long getAwsHttpClientPoolPendingCount()
+    {
+        return awsHttpClientPoolPendingCount.get();
+    }
+
     @Managed
     @Nested
     public CounterStat getGetObjectRetries()
@@ -315,6 +337,21 @@ public void addAwsClientRetryPauseTime(Duration duration)
         awsClientRetryPauseTime.add(duration);
     }
 
+    public void setAwsHttpClientPoolAvailableCount(long count)
+    {
+        this.awsHttpClientPoolAvailableCount.set(count);
+    }
+
+    public void setAwsHttpClientPoolLeasedCount(long count)
+    {
+        this.awsHttpClientPoolLeasedCount.set(count);
+    }
+
+    public void setAwsHttpClientPoolPendingCount(long count)
+    {
+        this.awsHttpClientPoolPendingCount.set(count);
+    }
+
     public void newGetObjectRetry()
     {
         getObjectRetries.update(1);