readahead backwards from sst end

Summary:
prefetch some data from the end of the file for each compaction to reduce IO.
Closes https://github.com/facebook/rocksdb/pull/2149

Differential Revision: D4880576

Pulled By: lightmark

fbshipit-source-id: aa767cd1afc84c541837fbf1ad6c0d45b34d3932

include/rocksdb/env.h

@@ -472,6 +472,7 @@ class SequentialFile {
 // A file abstraction for randomly reading the contents of a file.
 class RandomAccessFile {
  public:
+
   RandomAccessFile() { }
   virtual ~RandomAccessFile();
 
@@ -488,6 +489,11 @@ class RandomAccessFile {
   virtual Status Read(uint64_t offset, size_t n, Slice* result,
                       char* scratch) const = 0;
 
+  // Readahead the file starting from offset by n bytes for caching.
+  virtual Status Prefetch(uint64_t offset, size_t n) {
+    return Status::OK();
+  }
+
   // Used by the file_reader_writer to decide if the ReadAhead wrapper
   // should simply forward the call and do not enact buffering or locking.
   virtual bool ShouldForwardRawRequest() const {

table/block_based_table_reader.cc

@@ -530,7 +530,12 @@ Status BlockBasedTable::Open(const ImmutableCFOptions& ioptions,
   table_reader->reset();
 
   Footer footer;
-  auto s = ReadFooterFromFile(file.get(), file_size, &footer,
+
+  // Before read footer, readahead backwards to prefetch data
+  Status s =
+      file->Prefetch((file_size < 512 * 1024 ? 0 : file_size - 512 * 1024),
+                     512 * 1024 /* 512 KB prefetching */);
+  s = ReadFooterFromFile(file.get(), file_size, &footer,
                               kBlockBasedTableMagicNumber);
   if (!s.ok()) {
     return s;
@@ -541,8 +546,7 @@ Status BlockBasedTable::Open(const ImmutableCFOptions& ioptions,
         "version of RocksDB?");
   }
 
-  // We've successfully read the footer and the index block: we're
-  // ready to serve requests.
+  // We've successfully read the footer. We are ready to serve requests.
   // Better not mutate rep_ after the creation. eg. internal_prefix_transform
   // raw pointer will be used to create HashIndexReader, whose reset may
   // access a dangling pointer.

util/file_reader_writer.cc

@@ -472,7 +472,7 @@ class ReadaheadRandomAccessFile : public RandomAccessFile {
     // complitely or partially in the buffer
     // If it's completely cached, including end of file case when offset + n is
     // greater than EOF, return
-    if (TryReadFromCache_(offset, n, &cached_len, scratch) &&
+    if (TryReadFromCache(offset, n, &cached_len, scratch) &&
         (cached_len == n ||
          // End of file
          buffer_len_ < readahead_size_)) {
@@ -484,35 +484,35 @@ class ReadaheadRandomAccessFile : public RandomAccessFile {
     // chunk_offset equals to advanced_offset
     size_t chunk_offset = TruncateToPageBoundary(alignment_, advanced_offset);
     Slice readahead_result;
-    Status s = file_->Read(chunk_offset, readahead_size_, &readahead_result,
-                           buffer_.BufferStart());
-    if (!s.ok()) {
-      return s;
-    }
+
+    Status s = ReadIntoBuffer(chunk_offset, readahead_size_);
+    if (s.ok()) {
       // In the case of cache miss, i.e. when cached_len equals 0, an offset can
       // exceed the file end position, so the following check is required
-    if (advanced_offset < chunk_offset + readahead_result.size()) {
-      // In the case of cache miss, the first chunk_padding bytes in buffer_ are
+      if (advanced_offset < chunk_offset + buffer_len_) {
+        // In the case of cache miss, the first chunk_padding bytes in buffer_
+        // are
         // stored for alignment only and must be skipped
         size_t chunk_padding = advanced_offset - chunk_offset;
         auto remaining_len =
-          std::min(readahead_result.size() - chunk_padding, n - cached_len);
-      memcpy(scratch + cached_len, readahead_result.data() + chunk_padding,
+            std::min(buffer_len_ - chunk_padding, n - cached_len);
+        memcpy(scratch + cached_len, buffer_.BufferStart() + chunk_padding,
                remaining_len);
         *result = Slice(scratch, cached_len + remaining_len);
       } else {
         *result = Slice(scratch, cached_len);
       }
-
-    if (readahead_result.data() == buffer_.BufferStart()) {
-      buffer_offset_ = chunk_offset;
-      buffer_len_ = readahead_result.size();
-    } else {
-      buffer_len_ = 0;
+    }
+    return s;
   }
 
+  virtual Status Prefetch(uint64_t offset, size_t n) override {
+    size_t prefetch_offset = TruncateToPageBoundary(alignment_, offset);
+    if (prefetch_offset == buffer_offset_) {
       return Status::OK();
     }
+    return ReadIntoBuffer(prefetch_offset, offset - prefetch_offset + n);
+  }
 
   virtual size_t GetUniqueId(char* id, size_t max_size) const override {
     return file_->GetUniqueId(id, max_size);
@@ -529,7 +529,7 @@ class ReadaheadRandomAccessFile : public RandomAccessFile {
   }
 
  private:
-  bool TryReadFromCache_(uint64_t offset, size_t n, size_t* cached_len,
+  bool TryReadFromCache(uint64_t offset, size_t n, size_t* cached_len,
                          char* scratch) const {
     if (offset < buffer_offset_ || offset >= buffer_offset_ + buffer_len_) {
       *cached_len = 0;
@@ -542,6 +542,19 @@ class ReadaheadRandomAccessFile : public RandomAccessFile {
     return true;
   }
 
+  Status ReadIntoBuffer(uint64_t offset, size_t n) const {
+    if (n > buffer_.Capacity()) {
+      n = buffer_.Capacity();
+    }
+    Slice result;
+    Status s = file_->Read(offset, n, &result, buffer_.BufferStart());
+    if (s.ok()) {
+      buffer_offset_ = offset;
+      buffer_len_ = result.size();
+    }
+    return s;
+  }
+
   std::unique_ptr<RandomAccessFile> file_;
   const size_t alignment_;
   size_t               readahead_size_;

util/file_reader_writer.h

@@ -92,6 +92,10 @@ class RandomAccessFileReader {
 
   Status Read(uint64_t offset, size_t n, Slice* result, char* scratch) const;
 
+  Status Prefetch(uint64_t offset, size_t n) const {
+    return file_->Prefetch(offset, n);
+  }
+
   RandomAccessFile* file() { return file_.get(); }
 
   bool use_direct_io() const { return file_->use_direct_io(); }