Write properties metablock last in block-based tables (#4158)

Summary:
The properties meta-block should come at the end since we always need to
read it when opening a file, unlike index/filter/other meta-blocks, which
are sometimes read depending on the user's configuration. This ordering
will allow us to (in a future PR) do a small readahead on the end of the file
to read properties and meta-index blocks with one I/O.

The bulk of this PR is a refactoring of the `BlockBasedTableBuilder::Finish`
function. It was previously too large with inconsistent error handling, which
made it difficult to change. So I broke it up into one function per meta-block
write, and tried to make error handling consistent within those functions.
Then reordering the metablocks was trivial -- just reorder the calls to these
helper functions.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4158

Differential Revision: D8921705

Pulled By: ajkr

fbshipit-source-id: 96c9cc3182eb1adf11af46adab79dbeba7b12fcc

table/block_based_table_builder.cc

@@ -37,7 +37,6 @@
 #include "table/filter_block.h"
 #include "table/format.h"
 #include "table/full_filter_block.h"
-#include "table/meta_blocks.h"
 #include "table/table_builder.h"
 
 #include "util/string_util.h"
@@ -668,184 +667,204 @@ Status BlockBasedTableBuilder::InsertBlockInCache(const Slice& block_contents,
   return Status::OK();
 }
 
-Status BlockBasedTableBuilder::Finish() {
-  Rep* r = rep_;
-  bool empty_data_block = r->data_block.empty();
-  Flush();
-  assert(!r->closed);
-  r->closed = true;
-
-  // To make sure properties block is able to keep the accurate size of index
-  // block, we will finish writing all index entries here and flush them
-  // to storage after metaindex block is written.
-  if (ok() && !empty_data_block) {
-    r->index_builder->AddIndexEntry(
-        &r->last_key, nullptr /* no next data block */, r->pending_handle);
-  }
-
-  BlockHandle filter_block_handle, metaindex_block_handle, index_block_handle,
-      compression_dict_block_handle, range_del_block_handle;
-
-  // Write filter block
-  bool empty_filter_block = (r->filter_builder == nullptr ||
-                             r->filter_builder->NumAdded() == 0);
+void BlockBasedTableBuilder::WriteFilterBlock(
+    MetaIndexBuilder* meta_index_builder) {
+  BlockHandle filter_block_handle;
+  bool empty_filter_block = (rep_->filter_builder == nullptr ||
+                             rep_->filter_builder->NumAdded() == 0);
   if (ok() && !empty_filter_block) {
     Status s = Status::Incomplete();
-    while (s.IsIncomplete()) {
-      Slice filter_content = r->filter_builder->Finish(filter_block_handle, &s);
+    while (ok() && s.IsIncomplete()) {
+      Slice filter_content = rep_->filter_builder->Finish(filter_block_handle, &s);
       assert(s.ok() || s.IsIncomplete());
-      r->props.filter_size += filter_content.size();
+      rep_->props.filter_size += filter_content.size();
       WriteRawBlock(filter_content, kNoCompression, &filter_block_handle);
     }
   }
+  if (ok() && !empty_filter_block) {
+    // Add mapping from "<filter_block_prefix>.Name" to location
+    // of filter data.
+    std::string key;
+    if (rep_->filter_builder->IsBlockBased()) {
+      key = BlockBasedTable::kFilterBlockPrefix;
+    } else {
+      key = rep_->table_options.partition_filters
+                ? BlockBasedTable::kPartitionedFilterBlockPrefix
+                : BlockBasedTable::kFullFilterBlockPrefix;
+    }
+    key.append(rep_->table_options.filter_policy->Name());
+    meta_index_builder->Add(key, filter_block_handle);
+  }
+}
 
+void BlockBasedTableBuilder::WriteIndexBlock(
+    MetaIndexBuilder* meta_index_builder, BlockHandle* index_block_handle) {
   IndexBuilder::IndexBlocks index_blocks;
-  auto index_builder_status = r->index_builder->Finish(&index_blocks);
+  auto index_builder_status = rep_->index_builder->Finish(&index_blocks);
   if (index_builder_status.IsIncomplete()) {
     // We we have more than one index partition then meta_blocks are not
     // supported for the index. Currently meta_blocks are used only by
     // HashIndexBuilder which is not multi-partition.
     assert(index_blocks.meta_blocks.empty());
-  } else if (!index_builder_status.ok()) {
-    return index_builder_status;
+  } else if (ok() && !index_builder_status.ok()) {
+    rep_->status = index_builder_status;
   }
-
-  // Write meta blocks and metaindex block with the following order.
-  //    1. [meta block: filter]
-  //    2. [meta block: properties]
-  //    3. [meta block: compression dictionary]
-  //    4. [meta block: range deletion tombstone]
-  //    5. [metaindex block]
-  // write meta blocks
-  MetaIndexBuilder meta_index_builder;
-  for (const auto& item : index_blocks.meta_blocks) {
-    BlockHandle block_handle;
-    WriteBlock(item.second, &block_handle, false /* is_data_block */);
-    meta_index_builder.Add(item.first, block_handle);
-  }
-
   if (ok()) {
-    if (!empty_filter_block) {
-      // Add mapping from "<filter_block_prefix>.Name" to location
-      // of filter data.
-      std::string key;
-      if (r->filter_builder->IsBlockBased()) {
-        key = BlockBasedTable::kFilterBlockPrefix;
-      } else {
-        key = r->table_options.partition_filters
-                  ? BlockBasedTable::kPartitionedFilterBlockPrefix
-                  : BlockBasedTable::kFullFilterBlockPrefix;
+    for (const auto& item : index_blocks.meta_blocks) {
+      BlockHandle block_handle;
+      WriteBlock(item.second, &block_handle, false /* is_data_block */);
+      if (!ok()) {
+        break;
       }
-      key.append(r->table_options.filter_policy->Name());
-      meta_index_builder.Add(key, filter_block_handle);
+      meta_index_builder->Add(item.first, block_handle);
+    }
+  }
+  if (ok()) {
+    if (rep_->table_options.enable_index_compression) {
+      WriteBlock(index_blocks.index_block_contents, index_block_handle, false);
+    } else {
+      WriteRawBlock(index_blocks.index_block_contents, kNoCompression,
+                    index_block_handle);
     }
+  }
+  // If there are more index partitions, finish them and write them out
+  Status s = index_builder_status;
+  while (ok() && s.IsIncomplete()) {
+    s = rep_->index_builder->Finish(&index_blocks, *index_block_handle);
+    if (!s.ok() && !s.IsIncomplete()) {
+      rep_->status = s;
+      return;
+    }
+    if (rep_->table_options.enable_index_compression) {
+      WriteBlock(index_blocks.index_block_contents, index_block_handle, false);
+    } else {
+      WriteRawBlock(index_blocks.index_block_contents, kNoCompression,
+                    index_block_handle);
+    }
+    // The last index_block_handle will be for the partition index block
+  }
+}
 
-    // Write properties and compression dictionary blocks.
-    {
-      PropertyBlockBuilder property_block_builder;
-      r->props.column_family_id = r->column_family_id;
-      r->props.column_family_name = r->column_family_name;
-      r->props.filter_policy_name = r->table_options.filter_policy != nullptr ?
-          r->table_options.filter_policy->Name() : "";
-      r->props.index_size =
-          r->index_builder->EstimatedSize() + kBlockTrailerSize;
-      r->props.comparator_name = r->ioptions.user_comparator != nullptr
-                                     ? r->ioptions.user_comparator->Name()
-                                     : "nullptr";
-      r->props.merge_operator_name = r->ioptions.merge_operator != nullptr
-                                         ? r->ioptions.merge_operator->Name()
+void BlockBasedTableBuilder::WritePropertiesBlock(
+    MetaIndexBuilder* meta_index_builder) {
+  BlockHandle properties_block_handle;
+  if (ok()) {
+    PropertyBlockBuilder property_block_builder;
+    rep_->props.column_family_id = rep_->column_family_id;
+    rep_->props.column_family_name = rep_->column_family_name;
+    rep_->props.filter_policy_name = rep_->table_options.filter_policy != nullptr
+                                      ? rep_->table_options.filter_policy->Name()
+                                      : "";
+    rep_->props.index_size = rep_->index_builder->EstimatedSize() + kBlockTrailerSize;
+    rep_->props.comparator_name = rep_->ioptions.user_comparator != nullptr
+                                   ? rep_->ioptions.user_comparator->Name()
+                                   : "nullptr";
+    rep_->props.merge_operator_name = rep_->ioptions.merge_operator != nullptr
+                                       ? rep_->ioptions.merge_operator->Name()
+                                       : "nullptr";
+    rep_->props.compression_name =
+        CompressionTypeToString(rep_->compression_ctx.type());
+    rep_->props.prefix_extractor_name = rep_->moptions.prefix_extractor != nullptr
+                                         ? rep_->moptions.prefix_extractor->Name()
                                          : "nullptr";
-      r->props.compression_name =
-          CompressionTypeToString(r->compression_ctx.type());
-      r->props.prefix_extractor_name =
-          r->moptions.prefix_extractor != nullptr
-              ? r->moptions.prefix_extractor->Name()
-              : "nullptr";
-
-      std::string property_collectors_names = "[";
-      for (size_t i = 0;
-           i < r->ioptions.table_properties_collector_factories.size(); ++i) {
-        if (i != 0) {
-          property_collectors_names += ",";
-        }
-        property_collectors_names +=
-            r->ioptions.table_properties_collector_factories[i]->Name();
-      }
-      property_collectors_names += "]";
-      r->props.property_collectors_names = property_collectors_names;
-      if (r->table_options.index_type ==
-          BlockBasedTableOptions::kTwoLevelIndexSearch) {
-        assert(r->p_index_builder_ != nullptr);
-        r->props.index_partitions = r->p_index_builder_->NumPartitions();
-        r->props.top_level_index_size =
-            r->p_index_builder_->EstimateTopLevelIndexSize(r->offset);
-      }
-      r->props.index_key_is_user_key =
-          !r->index_builder->seperator_is_key_plus_seq();
-      r->props.creation_time = r->creation_time;
-      r->props.oldest_key_time = r->oldest_key_time;
-
-      // Add basic properties
-      property_block_builder.AddTableProperty(r->props);
-
-      // Add use collected properties
-      NotifyCollectTableCollectorsOnFinish(r->table_properties_collectors,
-                                           r->ioptions.info_log,
-                                           &property_block_builder);
-
-      BlockHandle properties_block_handle;
-      WriteRawBlock(
-          property_block_builder.Finish(),
-          kNoCompression,
-          &properties_block_handle
-      );
-      meta_index_builder.Add(kPropertiesBlock, properties_block_handle);
-
-      // Write compression dictionary block
-      if (r->compression_dict && r->compression_dict->size()) {
-        WriteRawBlock(*r->compression_dict, kNoCompression,
-                      &compression_dict_block_handle);
-        meta_index_builder.Add(kCompressionDictBlock,
-                               compression_dict_block_handle);
+
+    std::string property_collectors_names = "[";
+    for (size_t i = 0;
+         i < rep_->ioptions.table_properties_collector_factories.size(); ++i) {
+      if (i != 0) {
+        property_collectors_names += ",";
       }
-    }  // end of properties/compression dictionary block writing
+      property_collectors_names +=
+          rep_->ioptions.table_properties_collector_factories[i]->Name();
+    }
+    property_collectors_names += "]";
+    rep_->props.property_collectors_names = property_collectors_names;
+    if (rep_->table_options.index_type ==
+        BlockBasedTableOptions::kTwoLevelIndexSearch) {
+      assert(rep_->p_index_builder_ != nullptr);
+      rep_->props.index_partitions = rep_->p_index_builder_->NumPartitions();
+      rep_->props.top_level_index_size =
+          rep_->p_index_builder_->EstimateTopLevelIndexSize(rep_->offset);
+    }
+    rep_->props.index_key_is_user_key =
+        !rep_->index_builder->seperator_is_key_plus_seq();
+    rep_->props.creation_time = rep_->creation_time;
+    rep_->props.oldest_key_time = rep_->oldest_key_time;
+
+    // Add basic properties
+    property_block_builder.AddTableProperty(rep_->props);
 
-    if (ok() && !r->range_del_block.empty()) {
-      WriteRawBlock(r->range_del_block.Finish(), kNoCompression,
-                    &range_del_block_handle);
-      meta_index_builder.Add(kRangeDelBlock, range_del_block_handle);
-    }  // range deletion tombstone meta block
-  }    // meta blocks
+    // Add use collected properties
+    NotifyCollectTableCollectorsOnFinish(rep_->table_properties_collectors,
+                                         rep_->ioptions.info_log,
+                                         &property_block_builder);
 
-  // Write index block
+    WriteRawBlock(property_block_builder.Finish(), kNoCompression,
+                  &properties_block_handle);
+  }
   if (ok()) {
-    // flush the meta index block
-    WriteRawBlock(meta_index_builder.Finish(), kNoCompression,
-                  &metaindex_block_handle);
+    meta_index_builder->Add(kPropertiesBlock, properties_block_handle);
+  }
+}
 
-    if (r->table_options.enable_index_compression) {
-      WriteBlock(index_blocks.index_block_contents, &index_block_handle, false);
-    } else {
-      WriteRawBlock(index_blocks.index_block_contents, kNoCompression,
-                    &index_block_handle);
+void BlockBasedTableBuilder::WriteCompressionDictBlock(
+    MetaIndexBuilder* meta_index_builder) {
+  if (rep_->compression_dict && rep_->compression_dict->size()) {
+    BlockHandle compression_dict_block_handle;
+    if (ok()) {
+      WriteRawBlock(*rep_->compression_dict, kNoCompression,
+                    &compression_dict_block_handle);
     }
-    // If there are more index partitions, finish them and write them out
-    Status& s = index_builder_status;
-    while (s.IsIncomplete()) {
-      s = r->index_builder->Finish(&index_blocks, index_block_handle);
-      if (!s.ok() && !s.IsIncomplete()) {
-        return s;
-      }
-      if (r->table_options.enable_index_compression) {
-        WriteBlock(index_blocks.index_block_contents, &index_block_handle,
-                   false);
-      } else {
-        WriteRawBlock(index_blocks.index_block_contents, kNoCompression,
-                      &index_block_handle);
-      }
-      // The last index_block_handle will be for the partition index block
+    if (ok()) {
+      meta_index_builder->Add(kCompressionDictBlock,
+                              compression_dict_block_handle);
     }
   }
+}
+
+void BlockBasedTableBuilder::WriteRangeDelBlock(
+    MetaIndexBuilder* meta_index_builder) {
+  if (ok() && !rep_->range_del_block.empty()) {
+    BlockHandle range_del_block_handle;
+    WriteRawBlock(rep_->range_del_block.Finish(), kNoCompression,
+                  &range_del_block_handle);
+    meta_index_builder->Add(kRangeDelBlock, range_del_block_handle);
+  }
+}
+
+Status BlockBasedTableBuilder::Finish() {
+  Rep* r = rep_;
+  bool empty_data_block = r->data_block.empty();
+  Flush();
+  assert(!r->closed);
+  r->closed = true;
+
+  // To make sure properties block is able to keep the accurate size of index
+  // block, we will finish writing all index entries first.
+  if (ok() && !empty_data_block) {
+    r->index_builder->AddIndexEntry(
+        &r->last_key, nullptr /* no next data block */, r->pending_handle);
+  }
+
+  // Write meta blocks and metaindex block with the following order.
+  //    1. [meta block: filter]
+  //    2. [meta block: index]
+  //    3. [meta block: compression dictionary]
+  //    4. [meta block: range deletion tombstone]
+  //    5. [meta block: properties]
+  //    6. [metaindex block]
+  BlockHandle metaindex_block_handle, index_block_handle;
+  MetaIndexBuilder meta_index_builder;
+  WriteFilterBlock(&meta_index_builder);
+  WriteIndexBlock(&meta_index_builder, &index_block_handle);
+  WriteCompressionDictBlock(&meta_index_builder);
+  WriteRangeDelBlock(&meta_index_builder);
+  WritePropertiesBlock(&meta_index_builder);
+  if (ok()) {
+    // flush the meta index block
+    WriteRawBlock(meta_index_builder.Finish(), kNoCompression,
+                  &metaindex_block_handle);
+  }
 
   // Write footer
   if (ok()) {

table/block_based_table_builder.h

@@ -18,6 +18,7 @@
 #include "rocksdb/listener.h"
 #include "rocksdb/options.h"
 #include "rocksdb/status.h"
+#include "table/meta_blocks.h"
 #include "table/table_builder.h"
 #include "util/compression.h"
 
@@ -106,6 +107,14 @@ class BlockBasedTableBuilder : public TableBuilder {
   Status InsertBlockInCache(const Slice& block_contents,
                             const CompressionType type,
                             const BlockHandle* handle);
+
+  void WriteFilterBlock(MetaIndexBuilder* meta_index_builder);
+  void WriteIndexBlock(MetaIndexBuilder* meta_index_builder,
+                       BlockHandle* index_block_handle);
+  void WritePropertiesBlock(MetaIndexBuilder* meta_index_builder);
+  void WriteCompressionDictBlock(MetaIndexBuilder* meta_index_builder);
+  void WriteRangeDelBlock(MetaIndexBuilder* meta_index_builder);
+
   struct Rep;
   class BlockBasedTablePropertiesCollectorFactory;
   class BlockBasedTablePropertiesCollector;

table/table_test.cc

@@ -352,19 +352,19 @@ class TableConstructor: public Constructor {
     file_writer_->Flush();
     EXPECT_TRUE(s.ok()) << s.ToString();
 
-    EXPECT_EQ(GetSink()->contents().size(), builder->FileSize());
+    EXPECT_EQ(TEST_GetSink()->contents().size(), builder->FileSize());
 
     // Open the table
     uniq_id_ = cur_uniq_id_++;
     file_reader_.reset(test::GetRandomAccessFileReader(new test::StringSource(
-        GetSink()->contents(), uniq_id_, ioptions.allow_mmap_reads)));
+        TEST_GetSink()->contents(), uniq_id_, ioptions.allow_mmap_reads)));
     const bool kSkipFilters = true;
     const bool kImmortal = true;
     return ioptions.table_factory->NewTableReader(
         TableReaderOptions(ioptions, moptions.prefix_extractor.get(), soptions,
                            internal_comparator, !kSkipFilters, !kImmortal,
                            level_),
-        std::move(file_reader_), GetSink()->contents().size(), &table_reader_);
+        std::move(file_reader_), TEST_GetSink()->contents().size(), &table_reader_);
   }
 
   virtual InternalIterator* NewIterator(
@@ -390,11 +390,11 @@ class TableConstructor: public Constructor {
   virtual Status Reopen(const ImmutableCFOptions& ioptions,
                         const MutableCFOptions& moptions) {
     file_reader_.reset(test::GetRandomAccessFileReader(new test::StringSource(
-        GetSink()->contents(), uniq_id_, ioptions.allow_mmap_reads)));
+        TEST_GetSink()->contents(), uniq_id_, ioptions.allow_mmap_reads)));
     return ioptions.table_factory->NewTableReader(
         TableReaderOptions(ioptions, moptions.prefix_extractor.get(), soptions,
                            *last_internal_key_),
-        std::move(file_reader_), GetSink()->contents().size(), &table_reader_);
+        std::move(file_reader_), TEST_GetSink()->contents().size(), &table_reader_);
   }
 
   virtual TableReader* GetTableReader() {
@@ -409,6 +409,10 @@ class TableConstructor: public Constructor {
 
   bool ConvertToInternalKey() { return convert_to_internal_key_; }
 
+  test::StringSink* TEST_GetSink() {
+    return static_cast<test::StringSink*>(file_writer_->writable_file());
+  }
+
  private:
   void Reset() {
     uniq_id_ = 0;
@@ -417,10 +421,6 @@ class TableConstructor: public Constructor {
     file_reader_.reset();
   }
 
-  test::StringSink* GetSink() {
-    return static_cast<test::StringSink*>(file_writer_->writable_file());
-  }
-
   uint64_t uniq_id_;
   unique_ptr<WritableFileWriter> file_writer_;
   unique_ptr<RandomAccessFileReader> file_reader_;
@@ -3494,6 +3494,86 @@ TEST_P(BlockBasedTableTest, PropertiesBlockRestartPointTest) {
   }
 }
 
+TEST_P(BlockBasedTableTest, PropertiesMetaBlockLast) {
+  // The properties meta-block should come at the end since we always need to
+  // read it when opening a file, unlike index/filter/other meta-blocks, which
+  // are sometimes read depending on the user's configuration. This ordering
+  // allows us to do a small readahead on the end of the file to read properties
+  // and meta-index blocks with one I/O.
+  TableConstructor c(BytewiseComparator(), true /* convert_to_internal_key_ */);
+  c.Add("a1", "val1");
+  c.Add("b2", "val2");
+  c.Add("c3", "val3");
+  c.Add("d4", "val4");
+  c.Add("e5", "val5");
+  c.Add("f6", "val6");
+  c.Add("g7", "val7");
+  c.Add("h8", "val8");
+  c.Add("j9", "val9");
+
+  // write an SST file
+  Options options;
+  BlockBasedTableOptions table_options = GetBlockBasedTableOptions();
+  table_options.filter_policy.reset(NewBloomFilterPolicy(
+      8 /* bits_per_key */, false /* use_block_based_filter */));
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  ImmutableCFOptions ioptions(options);
+  MutableCFOptions moptions(options);
+  std::vector<std::string> keys;
+  stl_wrappers::KVMap kvmap;
+  c.Finish(options, ioptions, moptions, table_options,
+           GetPlainInternalComparator(options.comparator), &keys, &kvmap);
+
+  // get file reader
+  test::StringSink* table_sink = c.TEST_GetSink();
+  std::unique_ptr<RandomAccessFileReader> table_reader{
+      test::GetRandomAccessFileReader(
+          new test::StringSource(table_sink->contents(), 0 /* unique_id */,
+                                 false /* allow_mmap_reads */))};
+  size_t table_size = table_sink->contents().size();
+
+  // read footer
+  Footer footer;
+  ASSERT_OK(ReadFooterFromFile(table_reader.get(),
+                               nullptr /* prefetch_buffer */, table_size,
+                               &footer, kBlockBasedTableMagicNumber));
+
+  // read metaindex
+  auto metaindex_handle = footer.metaindex_handle();
+  BlockContents metaindex_contents;
+  Slice compression_dict;
+  PersistentCacheOptions pcache_opts;
+  BlockFetcher block_fetcher(
+      table_reader.get(), nullptr /* prefetch_buffer */, footer, ReadOptions(),
+      metaindex_handle, &metaindex_contents, ioptions, false /* decompress */,
+      compression_dict, pcache_opts);
+  ASSERT_OK(block_fetcher.ReadBlockContents());
+  Block metaindex_block(std::move(metaindex_contents),
+                        kDisableGlobalSequenceNumber);
+
+  // verify properties block comes last
+  std::unique_ptr<InternalIterator> metaindex_iter{
+      metaindex_block.NewIterator<DataBlockIter>(options.comparator,
+                                                 options.comparator)};
+  uint64_t max_offset = 0;
+  std::string key_at_max_offset;
+  for (metaindex_iter->SeekToFirst(); metaindex_iter->Valid();
+       metaindex_iter->Next()) {
+    BlockHandle handle;
+    Slice value = metaindex_iter->value();
+    ASSERT_OK(handle.DecodeFrom(&value));
+    if (handle.offset() > max_offset) {
+      max_offset = handle.offset();
+      key_at_max_offset = metaindex_iter->key().ToString();
+    }
+  }
+  ASSERT_EQ(kPropertiesBlock, key_at_max_offset);
+  // index handle is stored in footer rather than metaindex block, so need
+  // separate logic to verify it comes before properties block.
+  ASSERT_GT(max_offset, footer.index_handle().offset());
+  c.ResetTableReader();
+}
+
 TEST_P(BlockBasedTableTest, BadOptions) {
   rocksdb::Options options;
   options.compression = kNoCompression;