crimson/os/seastore: measure records write in journal

Signed-off-by: Yingxin Cheng <yingxin.cheng@intel.com>

diff --git a/src/crimson/os/seastore/journal.cc b/src/crimson/os/seastore/journal.cc
index 8e5b06f4ce457949b9af6142df9d65d3c3a568e7..f81e23801c6db50ce8e225e18ec1a1778dbd95f2 100644 (file)
--- a/src/crimson/os/seastore/journal.cc
+++ b/src/crimson/os/seastore/journal.cc
@@ -267,6 +267,27 @@ void Journal::register_metrics()
         },
         sm::description("total number of io depth")
       ),
+      sm::make_counter(
+        "record_group_padding_bytes",
+        [this] {
+          return record_submitter.get_record_group_padding_bytes();
+        },
+        sm::description("bytes of metadata padding when write record groups")
+      ),
+      sm::make_counter(
+        "record_group_metadata_bytes",
+        [this] {
+          return record_submitter.get_record_group_metadata_bytes();
+        },
+        sm::description("bytes of raw metadata when write record groups")
+      ),
+      sm::make_counter(
+        "record_group_data_bytes",
+        [this] {
+          return record_submitter.get_record_group_data_bytes();
+        },
+        sm::description("bytes of data when write record groups")
+      ),
     }
   );
 }
@@ -454,7 +475,8 @@ Journal::RecordBatch::add_pending(
   });
 }
 
-ceph::bufferlist Journal::RecordBatch::encode_batch(
+std::pair<ceph::bufferlist, record_group_size_t>
+Journal::RecordBatch::encode_batch(
   const journal_seq_t& committed_to,
   segment_nonce_t segment_nonce)
 {
@@ -468,12 +490,13 @@ ceph::bufferlist Journal::RecordBatch::encode_batch(
 
   state = state_t::SUBMITTING;
   submitting_size = pending.get_size();
-  submitting_length = pending.size.get_encoded_length();
-  submitting_mdlength = pending.size.get_mdlength();
+  auto gsize = pending.size;
+  submitting_length = gsize.get_encoded_length();
+  submitting_mdlength = gsize.get_mdlength();
   auto bl = encode_records(pending, committed_to, segment_nonce);
   // Note: pending is cleared here
   assert(bl.length() == (std::size_t)submitting_length);
-  return bl;
+  return std::make_pair(bl, gsize);
 }
 
 void Journal::RecordBatch::set_result(
@@ -604,9 +627,10 @@ void Journal::RecordSubmitter::flush_current_batch()
   pop_free_batch();
 
   increment_io();
-  ceph::bufferlist to_write = p_batch->encode_batch(
+  auto [to_write, sizes] = p_batch->encode_batch(
     journal_segment_manager.get_committed_to(),
     journal_segment_manager.get_nonce());
+  account_submission(sizes);
   std::ignore = journal_segment_manager.write(to_write
   ).safe_then([this, p_batch](auto write_result) {
     finish_submit_batch(p_batch, write_result);
@@ -632,7 +656,7 @@ Journal::RecordSubmitter::submit_pending(
   bool flush)
 {
   assert(!p_current_batch->is_submitting());
-  record_batch_stats.increment(
+  stats.record_batch_stats.increment(
       p_current_batch->get_num_records() + 1);
   auto write_fut = [this, flush, record=std::move(record)]() mutable {
     if (flush && p_current_batch->is_empty()) {
@@ -643,6 +667,7 @@ Journal::RecordSubmitter::submit_pending(
         journal_segment_manager.get_block_size(),
         journal_segment_manager.get_committed_to(),
         journal_segment_manager.get_nonce());
+      account_submission(sizes);
       return journal_segment_manager.write(to_write
       ).safe_then([mdlength = sizes.get_mdlength()](auto write_result) {
         return record_locator_t{

diff --git a/src/crimson/os/seastore/journal.h b/src/crimson/os/seastore/journal.h
index 60484afce613aeda41bd1fed7978acf6106d364b..7f783e2938e47a71f4b772655d868b3bcc797327 100644 (file)
--- a/src/crimson/os/seastore/journal.h
+++ b/src/crimson/os/seastore/journal.h
@@ -290,7 +290,7 @@ private:
         extent_len_t block_size);
 
     // Encode the batched records for write.
-    ceph::bufferlist encode_batch(
+    std::pair<ceph::bufferlist, record_group_size_t> encode_batch(
         const journal_seq_t& committed_to,
         segment_nonce_t segment_nonce);
 
@@ -362,16 +362,27 @@ private:
                     JournalSegmentManager&);
 
     grouped_io_stats get_record_batch_stats() const {
-      return record_batch_stats;
+      return stats.record_batch_stats;
     }
 
     grouped_io_stats get_io_depth_stats() const {
-      return io_depth_stats;
+      return stats.io_depth_stats;
+    }
+
+    uint64_t get_record_group_padding_bytes() const {
+      return stats.record_group_padding_bytes;
+    }
+
+    uint64_t get_record_group_metadata_bytes() const {
+      return stats.record_group_metadata_bytes;
+    }
+
+    uint64_t get_record_group_data_bytes() const {
+      return stats.record_group_data_bytes;
     }
 
     void reset_stats() {
-      record_batch_stats = {};
-      io_depth_stats = {};
+      stats = {};
     }
 
     void set_write_pipeline(WritePipeline *_write_pipeline) {
@@ -386,7 +397,7 @@ private:
 
     void increment_io() {
       ++num_outstanding_io;
-      io_depth_stats.increment(num_outstanding_io);
+      stats.io_depth_stats.increment(num_outstanding_io);
       update_state();
     }
 
@@ -418,6 +429,13 @@ private:
       free_batch_ptrs.pop_front();
     }
 
+    void account_submission(const record_group_size_t& size) {
+      stats.record_group_padding_bytes +=
+        (size.get_mdlength() - size.get_raw_mdlength());
+      stats.record_group_metadata_bytes += size.get_raw_mdlength();
+      stats.record_group_data_bytes += size.dlength;
+    }
+
     using maybe_result_t = RecordBatch::maybe_result_t;
     void finish_submit_batch(RecordBatch*, maybe_result_t);
 
@@ -446,8 +464,13 @@ private:
     seastar::circular_buffer<RecordBatch*> free_batch_ptrs;
     std::optional<seastar::promise<> > wait_submit_promise;
 
-    grouped_io_stats record_batch_stats;
-    grouped_io_stats io_depth_stats;
+    struct {
+      grouped_io_stats record_batch_stats;
+      grouped_io_stats io_depth_stats;
+      uint64_t record_group_padding_bytes = 0;
+      uint64_t record_group_metadata_bytes = 0;
+      uint64_t record_group_data_bytes = 0;
+    } stats;
   };
 
   SegmentProvider* segment_provider = nullptr;