Add executor pool size to S3 configs

velox/connectors/hive/storage_adapters/s3fs/S3Config.h

@@ -77,6 +77,7 @@ class S3Config {
     kUseProxyFromEnv,
     kCredentialsProvider,
     kIMDSEnabled,
+    kExecutorPoolSize,
     kEnd
   };
 
@@ -116,6 +117,8 @@ class S3Config {
             {Keys::kCredentialsProvider,
              std::make_pair("aws-credentials-provider", std::nullopt)},
             {Keys::kIMDSEnabled, std::make_pair("aws-imds-enabled", "true")},
+            {Keys::kExecutorPoolSize,
+             std::make_pair("executor-pool-size", std::nullopt)},
         };
     return config;
   }
@@ -251,6 +254,15 @@ class S3Config {
     return folly::to<bool>(value);
   }
 
+  /// Executor pool size for async operations.
+  std::optional<uint32_t> executorPoolSize() const {
+    auto val = config_.find(Keys::kExecutorPoolSize)->second;
+    if (val.has_value()) {
+      return folly::to<uint32_t>(val.value());
+    }
+    return std::optional<uint32_t>();
+  }
+
  private:
   std::unordered_map<Keys, std::optional<std::string>> config_;
   std::string payloadSigningPolicy_;

velox/connectors/hive/storage_adapters/s3fs/S3FileSystem.cpp

@@ -34,6 +34,7 @@
 #include <aws/core/auth/AWSCredentialsProviderChain.h>
 #include <aws/core/client/AdaptiveRetryStrategy.h>
 #include <aws/core/client/DefaultRetryStrategy.h>
+#include <aws/core/utils/threading/PooledThreadExecutor.h>
 #include <aws/identity-management/auth/STSAssumeRoleCredentialsProvider.h>
 #include <aws/s3/S3Client.h>
 #include <aws/s3/model/CopyObjectRequest.h>
@@ -299,6 +300,9 @@ class S3FileSystem::Impl {
         inferPayloadSign(s3Config.payloadSigningPolicy());
 
     auto credentialsProvider = getCredentialsProvider(s3Config);
+    if (s3Config.executorPoolSize().has_value() && s3Config.executorPoolSize().value()>0) {
+      clientConfig.executor = Aws::MakeShared<Aws::Utils::Threading::PooledThreadExecutor>(nullptr, s3Config.executorPoolSize().value());
+    }
 
     client_ = std::make_shared<Aws::S3::S3Client>(
         credentialsProvider, nullptr /* endpointProvider */, clientConfig);

velox/docs/configs.rst

@@ -1054,6 +1054,10 @@ Each query can override the config by setting corresponding query session proper
      - true
      - AWS Instance Metadata Service (IMDS) is an AWS EC2 instance component used by applications to securely access metadata.
        We must disable it on other instances to avoid high first-time read latency from S3 compatible object storages.
+   * - hive.s3.executor-pool-size
+     - integer
+     -
+     - Define the executor size in the AWS thread pool which is used to perform S3 requests. 0 means disabled. 
 
 Bucket Level Configuration
 """"""""""""""""""""""""""