Flush Reporting

PR awslabs#102
pfifer · May 15, 2017 · 55ca106 · 55ca106
1 parent 2722572
commit 55ca106
Show file tree

Hide file tree

Showing 9 changed files with 322 additions and 21 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -69,7 +69,9 @@ set(SOURCE_FILES
         aws/utils/time_sensitive_queue.h
         aws/utils/token_bucket.h
         aws/utils/utils.cc
-        aws/utils/utils.h)
+        aws/utils/utils.h
+        aws/utils/processing_statistics_logger.cc
+        aws/utils/processing_statistics_logger.h)
 
 
 

diff --git a/aws/kinesis/core/aggregator.h b/aws/kinesis/core/aggregator.h
@@ -24,6 +24,7 @@
 #include <aws/kinesis/core/configuration.h>
 #include <aws/utils/concurrent_hash_map.h>
 #include <aws/utils/executor.h>
+#include <aws/utils/processing_statistics_logger.h>
 
 namespace aws {
 namespace kinesis {
@@ -41,12 +42,14 @@ class Aggregator : boost::noncopyable {
       const std::shared_ptr<ShardMap> shard_map,
       const DeadlineCallback& deadline_callback,
       const std::shared_ptr<aws::kinesis::core::Configuration>& config,
+      aws::utils::flush_statistics_aggregator& flush_stats,
       const std::shared_ptr<aws::metrics::MetricsManager>& metrics_manager =
           std::make_shared<aws::metrics::NullMetricsManager>())
       : executor_(executor),
         shard_map_(shard_map),
         deadline_callback_(deadline_callback),
         config_(config),
+        flush_stats_(flush_stats),
         metrics_manager_(metrics_manager),
         reducers_([this](auto) { return this->make_reducer(); }) {}
 
@@ -79,14 +82,17 @@ class Aggregator : boost::noncopyable {
         executor_,
         deadline_callback_,
         config_->aggregation_max_size(),
-        config_->aggregation_max_count());
+        config_->aggregation_max_count(),
+        flush_stats_
+    );
   }
 
   std::shared_ptr<aws::utils::Executor> executor_;
   std::shared_ptr<ShardMap> shard_map_;
   DeadlineCallback deadline_callback_;
   std::shared_ptr<aws::kinesis::core::Configuration> config_;
   std::shared_ptr<aws::metrics::MetricsManager> metrics_manager_;
+  aws::utils::flush_statistics_aggregator& flush_stats_;
   ReducerMap reducers_;
 };
 

diff --git a/aws/kinesis/core/collector.h b/aws/kinesis/core/collector.h
@@ -18,6 +18,7 @@
 #include <aws/kinesis/core/reducer.h>
 #include <aws/kinesis/core/configuration.h>
 #include <aws/utils/concurrent_hash_map.h>
+#include <aws/utils/processing_statistics_logger.h>
 
 namespace aws {
 namespace kinesis {
@@ -32,13 +33,15 @@ class Collector : boost::noncopyable {
       const std::shared_ptr<aws::utils::Executor>& executor,
       const FlushCallback& flush_callback,
       const std::shared_ptr<aws::kinesis::core::Configuration>& config,
+      aws::utils::flush_statistics_aggregator& flush_stats,
       const std::shared_ptr<aws::metrics::MetricsManager>& metrics_manager =
           std::make_shared<aws::metrics::NullMetricsManager>())
       : flush_callback_(flush_callback),
         reducer_(executor,
                  [this](auto prr) { this->handle_flush(std::move(prr)); },
                  config->collection_max_size(),
                  config->collection_max_count(),
+                 flush_stats,
                  [this](auto kr) { return this->should_flush(kr); }),
         buffered_data_([](auto) { return new std::atomic<size_t>(0); }) {}
 

diff --git a/aws/kinesis/core/pipeline.h b/aws/kinesis/core/pipeline.h
@@ -15,6 +15,7 @@
 #define AWS_KINESIS_CORE_PIPELINE_H_
 
 #include <boost/format.hpp>
+#include <iomanip>
 
 #include <aws/kinesis/core/aggregator.h>
 #include <aws/kinesis/core/collector.h>
@@ -25,6 +26,7 @@
 #include <aws/kinesis/core/retrier.h>
 #include <aws/kinesis/KinesisClient.h>
 #include <aws/metrics/metrics_manager.h>
+#include <aws/utils/processing_statistics_logger.h>
 
 namespace aws {
 namespace kinesis {
@@ -46,6 +48,7 @@ class Pipeline : boost::noncopyable {
       : stream_(std::move(stream)),
         region_(std::move(region)),
         config_(std::move(config)),
+        stats_logger_(stream_, config_->record_max_buffered_time()),
         executor_(std::move(executor)),
         kinesis_client_(std::move(kinesis_client)),
         metrics_manager_(std::move(metrics_manager)),
@@ -58,11 +61,12 @@ class Pipeline : boost::noncopyable {
                 metrics_manager_)),
         aggregator_(
             std::make_shared<Aggregator>(
-                executor_,
-                shard_map_,
-                [this](auto kr) { this->limiter_put(kr); },
-                config_,
-                metrics_manager_)),
+                    executor_,
+                    shard_map_,
+                    [this](auto kr) { this->limiter_put(kr); },
+                    config_,
+                    stats_logger_.stage1(),
+                    metrics_manager_)),
         limiter_(
             std::make_shared<Limiter>(
                 executor_,
@@ -71,10 +75,11 @@ class Pipeline : boost::noncopyable {
                 config_)),
         collector_(
             std::make_shared<Collector>(
-                executor_,
-                [this](auto prr) { this->send_put_records_request(prr); },
-                config_,
-                metrics_manager_)),
+                    executor_,
+                    [this](auto prr) { this->send_put_records_request(prr); },
+                    config_,
+                    stats_logger_.stage2(),
+                    metrics_manager_)),
         retrier_(
             std::make_shared<Retrier>(
                 config_,
@@ -111,6 +116,7 @@ class Pipeline : boost::noncopyable {
   }
 
  private:
+
   void aggregator_put(const std::shared_ptr<UserRecord>& ur) {
     auto kr = aggregator_->put(ur);
     if (kr) {
@@ -148,6 +154,7 @@ class Pipeline : boost::noncopyable {
                   sdk_ctx));
           ctx->set_end(std::chrono::steady_clock::now());
           ctx->set_outcome(outcome);
+          this->request_completed(ctx);
           // At the time of writing, the SDK can spawn a large number of
           // threads in order to achieve request parallelism. These threads will
           // later put items into the IPC manager after they finish the logic in
@@ -161,6 +168,10 @@ class Pipeline : boost::noncopyable {
         prc);
   }
 
+  void request_completed(std::shared_ptr<PutRecordsContext> context) {
+    stats_logger_.request_complete(context);
+  }
+
   void retrier_put_kr(const std::shared_ptr<KinesisRecord>& kr) {
     executor_->submit([=] {
       retrier_->put(kr,
@@ -172,6 +183,7 @@ class Pipeline : boost::noncopyable {
   std::string stream_;
   std::string region_;
   std::shared_ptr<Configuration> config_;
+  aws::utils::processing_statistics_logger stats_logger_;
   std::shared_ptr<aws::utils::Executor> executor_;
   std::shared_ptr<Aws::Kinesis::KinesisClient> kinesis_client_;
   std::shared_ptr<aws::metrics::MetricsManager> metrics_manager_;
@@ -185,6 +197,8 @@ class Pipeline : boost::noncopyable {
 
   std::shared_ptr<aws::metrics::Metric> user_records_rcvd_metric_;
   std::atomic<uint64_t> outstanding_user_records_;
+
+
 };
 
 } //namespace core

diff --git a/aws/kinesis/core/reducer.h b/aws/kinesis/core/reducer.h
@@ -17,8 +17,8 @@
 #include <mutex>
 
 #include <aws/utils/logging.h>
-
 #include <aws/utils/executor.h>
+#include <aws/utils/processing_statistics_logger.h>
 #include <aws/mutex.h>
 
 namespace aws {
@@ -57,11 +57,13 @@ class Reducer : boost::noncopyable {
       const std::function<void (std::shared_ptr<U>)>& flush_callback,
       size_t size_limit,
       size_t count_limit,
+      aws::utils::flush_statistics_aggregator& flush_stats,
       FlushPredicate flush_predicate = [](auto) { return false; })
       : executor_(executor),
         flush_callback_(flush_callback),
         size_limit_(size_limit),
         count_limit_(count_limit),
+        flush_stats_(flush_stats),
         flush_predicate_(flush_predicate),
         container_(std::make_shared<U>()),
         scheduled_callback_(
@@ -79,11 +81,13 @@ class Reducer : boost::noncopyable {
     auto size = container_->size();
     auto estimated_size = container_->estimated_size();
     auto flush_predicate_result = flush_predicate_(input);
-    if (size >= count_limit_ ||
-        estimated_size >= size_limit_ ||
-        flush_predicate_result) {
 
-      auto output = flush(lock);
+    aws::utils::flush_statistics_context flush_reason;
+    flush_reason.record_count(size >= count_limit_).data_size(estimated_size >= size_limit_)
+            .predicate_match(flush_predicate_result);
+
+    if (flush_reason.flush_required()) {
+      auto output = flush(lock, flush_reason);
       if (output && output->size() > 0) {
         return output;
       }
@@ -96,7 +100,9 @@ class Reducer : boost::noncopyable {
 
   // Manually trigger a flush, as though a deadline has been reached
   void flush() {
-    deadline_reached();
+    aws::utils::flush_statistics_context flush_reason;
+    flush_reason.manual(true);
+    trigger_flush(flush_reason);
   }
 
   // Records in the process of being flushed won't be counted
@@ -116,7 +122,7 @@ class Reducer : boost::noncopyable {
   using Mutex = aws::mutex;
   using Lock = aws::unique_lock<Mutex>;
 
-  std::shared_ptr<U> flush(Lock& lock) {
+  std::shared_ptr<U> flush(Lock &lock, aws::utils::flush_statistics_context &flush_reason) {
     if (!lock) {
       lock.lock();
     }
@@ -154,6 +160,7 @@ class Reducer : boost::noncopyable {
     }
 
     set_deadline();
+    flush_stats_.merge(flush_reason, flush_container->size());
 
     return flush_container;
   }
@@ -171,8 +178,14 @@ class Reducer : boost::noncopyable {
   }
 
   void deadline_reached() {
+    aws::utils::flush_statistics_context flush_reason;
+    flush_reason.timed(true);
+    trigger_flush(flush_reason);
+  }
+
+  void trigger_flush(aws::utils::flush_statistics_context &reason) {
     Lock lock(lock_);
-    auto r = flush(lock);
+    auto r = flush(lock, reason);
     lock.unlock();
     if (r && r->size() > 0) {
       flush_callback_(r);
@@ -187,6 +200,7 @@ class Reducer : boost::noncopyable {
   Mutex lock_;
   std::shared_ptr<U> container_;
   std::shared_ptr<aws::utils::ScheduledCallback> scheduled_callback_;
+  aws::utils::flush_statistics_aggregator& flush_stats_;
 };
 
 } //namespace core

diff --git a/aws/kinesis/core/test/aggregator_test.cc b/aws/kinesis/core/test/aggregator_test.cc
@@ -17,6 +17,7 @@
 #include <aws/kinesis/core/test/test_utils.h>
 #include <aws/utils/io_service_executor.h>
 #include <aws/utils/utils.h>
+#include <aws/utils/processing_statistics_logger.h>
 
 namespace {
 
@@ -67,6 +68,8 @@ class MockShardMap : public aws::kinesis::core::ShardMap {
   std::vector<boost::multiprecision::uint128_t> limits_;
 };
 
+  aws::utils::flush_statistics_aggregator flush_stats("Test", "InRecords", "OutRecords");
+
 auto make_aggregator(
     bool shard_map_down = false,
     FlushCallback cb = [](auto) {},
@@ -81,7 +84,8 @@ auto make_aggregator(
           std::make_shared<aws::utils::IoServiceExecutor>(4),
           std::make_shared<MockShardMap>(shard_map_down),
           cb,
-          config);
+          config,
+          flush_stats);
 }
 
 } //namespace

diff --git a/aws/kinesis/core/test/reducer_test.cc b/aws/kinesis/core/test/reducer_test.cc
@@ -17,6 +17,7 @@
 #include <aws/utils/utils.h>
 #include <aws/utils/io_service_executor.h>
 #include <aws/kinesis/core/test/test_utils.h>
+#include <aws/utils/processing_statistics_logger.h>
 
 namespace {
 
@@ -27,11 +28,13 @@ using Reducer =
 using FlushCallback =
     std::function<void (std::shared_ptr<aws::kinesis::core::KinesisRecord>)>;
 
+  aws::utils::flush_statistics_aggregator flush_stats("Test", "TestRecords", "TestRecords2");
+
 std::shared_ptr<Reducer> make_reducer(size_t size_limit = 256 * 1024,
                                       size_t count_limit = 1000,
                                       FlushCallback cb = [](auto) {}) {
   auto executor = std::make_shared<aws::utils::IoServiceExecutor>(8);
-  return std::make_shared<Reducer>(executor, cb, size_limit, count_limit);
+  return std::make_shared<Reducer>(executor, cb, size_limit, count_limit, flush_stats);
 }
 
 template <typename T>