Add unsafe_add option to SstFileWriter

Summary: We observed that key compare in SstFileWriter::Add() adds significant overhead to our ingest. This diff adds an option to avoid this check.

Test Plan: make check

Reviewers: hieu, dhruba

Differential Revision: https://rockset.phacility.com/D8914

Author: igorcanadi

include/rocksdb/sst_file_writer.h

@@ -80,21 +80,23 @@ class SstFileWriter {
   // hint that this file pages is not needed every time we write 1MB to the
   // file. To use the rate limiter an io_priority smaller than IO_TOTAL can be
   // passed.
+  // If unsafe_add is true, SstFileWriter doesn't check that keys are written in
+  // ascending order.
   SstFileWriter(const EnvOptions& env_options, const Options& options,
                 ColumnFamilyHandle* column_family = nullptr,
                 bool invalidate_page_cache = true,
                 Env::IOPriority io_priority = Env::IOPriority::IO_TOTAL,
-                bool skip_filters = false)
+                bool skip_filters = false, bool unsafe_add = false)
       : SstFileWriter(env_options, options, options.comparator, column_family,
-                      invalidate_page_cache, io_priority, skip_filters) {}
+                      invalidate_page_cache, io_priority, skip_filters, unsafe_add) {}
 
   // Deprecated API
   SstFileWriter(const EnvOptions& env_options, const Options& options,
                 const Comparator* user_comparator,
                 ColumnFamilyHandle* column_family = nullptr,
                 bool invalidate_page_cache = true,
                 Env::IOPriority io_priority = Env::IOPriority::IO_TOTAL,
-                bool skip_filters = false);
+                bool skip_filters = false, bool unsafe_add = false);
 
   ~SstFileWriter();
 

table/sst_file_writer.cc

@@ -29,7 +29,8 @@ const size_t kFadviseTrigger = 1024 * 1024; // 1MB
 struct SstFileWriter::Rep {
   Rep(const EnvOptions& _env_options, const Options& options,
       Env::IOPriority _io_priority, const Comparator* _user_comparator,
-      ColumnFamilyHandle* _cfh, bool _invalidate_page_cache, bool _skip_filters)
+      ColumnFamilyHandle* _cfh, bool _invalidate_page_cache, bool _skip_filters,
+      bool _unsafe_add)
       : env_options(_env_options),
         ioptions(options),
         mutable_cf_options(options),
@@ -38,7 +39,8 @@ struct SstFileWriter::Rep {
         cfh(_cfh),
         invalidate_page_cache(_invalidate_page_cache),
         last_fadvise_size(0),
-        skip_filters(_skip_filters) {}
+        skip_filters(_skip_filters),
+        unsafe_add(_unsafe_add) {}
 
   std::unique_ptr<WritableFileWriter> file_writer;
   std::unique_ptr<TableBuilder> builder;
@@ -58,6 +60,7 @@ struct SstFileWriter::Rep {
   // cached pages from page cache.
   uint64_t last_fadvise_size;
   bool skip_filters;
+  bool unsafe_add;
   Status Add(const Slice& user_key, const Slice& value,
              const ValueType value_type) {
     if (!builder) {
@@ -66,7 +69,7 @@ struct SstFileWriter::Rep {
 
     if (file_info.num_entries == 0) {
       file_info.smallest_key.assign(user_key.data(), user_key.size());
-    } else {
+    } else if (!unsafe_add) {
       if (internal_comparator.user_comparator()->Compare(
               user_key, file_info.largest_key) <= 0) {
         // Make sure that keys are added in order
@@ -163,9 +166,11 @@ SstFileWriter::SstFileWriter(const EnvOptions& env_options,
                              const Comparator* user_comparator,
                              ColumnFamilyHandle* column_family,
                              bool invalidate_page_cache,
-                             Env::IOPriority io_priority, bool skip_filters)
+                             Env::IOPriority io_priority, bool skip_filters,
+                             bool unsafe_add)
     : rep_(new Rep(env_options, options, io_priority, user_comparator,
-                   column_family, invalidate_page_cache, skip_filters)) {
+                   column_family, invalidate_page_cache, skip_filters,
+                   unsafe_add)) {
   rep_->file_info.file_size = 0;
 }