better tuning of arena block size

Summary: Currently, if users didn't set options.arena_block_size, we set "result.arena_block_size = result.write_buffer_size / 10". It makes result.arena_block_size not a multiplier of 4KB, even if options.write_buffer_size is a multiplier of MBs. When calling malloc to arena_block_size, we may waste a small amount of memory for it. We now make the default to be /8 or /16 and align it to 4KB.

Test Plan: unit tests

Reviewers: sdong

Reviewed By: sdong

Subscribers: dhruba, leveldb

Differential Revision: https://reviews.facebook.net/D46467

db/column_family.cc

@@ -125,7 +125,12 @@ ColumnFamilyOptions SanitizeOptions(const DBOptions& db_options,
   // if user sets arena_block_size, we trust user to use this value. Otherwise,
   // calculate a proper value from writer_buffer_size;
   if (result.arena_block_size <= 0) {
-    result.arena_block_size = result.write_buffer_size / 10;
+    result.arena_block_size = result.write_buffer_size / 8;
+
+    // Align up to 4k
+    const size_t align = 4 * 1024;
+    result.arena_block_size =
+        ((result.arena_block_size + align - 1) / align) * align;
   }
   result.min_write_buffer_number_to_merge =
       std::min(result.min_write_buffer_number_to_merge,

db/column_family_test.cc

@@ -1102,6 +1102,9 @@ TEST_F(ColumnFamilyTest, SanitizeOptions) {
             original.level0_stop_writes_trigger = i;
             original.level0_slowdown_writes_trigger = j;
             original.level0_file_num_compaction_trigger = k;
+            original.write_buffer_size =
+                l * 4 * 1024 * 1024 + i * 1024 * 1024 + j * 1024 + k;
+
             ColumnFamilyOptions result =
                 SanitizeOptions(db_options, nullptr, original);
             ASSERT_TRUE(result.level0_stop_writes_trigger >=
@@ -1118,6 +1121,16 @@ TEST_F(ColumnFamilyTest, SanitizeOptions) {
                 ASSERT_EQ(result.num_levels, original.num_levels);
               }
             }
+
+            // Make sure Sanitize options sets arena_block_size to 1/8 of
+            // the write_buffer_size, rounded up to a multiple of 4k.
+            size_t expected_arena_block_size =
+                l * 4 * 1024 * 1024 / 8 + i * 1024 * 1024 / 8;
+            if (j + k != 0) {
+              // not a multiple of 4k, round up 4k
+              expected_arena_block_size += 4 * 1024;
+            }
+            ASSERT_EQ(expected_arena_block_size, result.arena_block_size);
           }
         }
       }

db/db_compaction_test.cc

@@ -234,7 +234,7 @@ TEST_P(DBCompactionTestWithParam, CompactionDeletionTrigger) {
     DestroyAndReopen(options);
     Random rnd(301);
 
-    const int kTestSize = kCDTKeysPerBuffer * 512;
+    const int kTestSize = kCDTKeysPerBuffer * 1024;
     std::vector<std::string> values;
     for (int k = 0; k < kTestSize; ++k) {
       values.push_back(RandomString(&rnd, kCDTValueSize));

db/db_test.cc

@@ -2692,7 +2692,7 @@ TEST_F(DBTest, FlushSchedule) {
     Random rnd(a);
     WriteOptions wo;
     // this should fill up 2 memtables
-    for (int k = 0; k < 6144; ++k) {
+    for (int k = 0; k < 5000; ++k) {
       ASSERT_OK(db_->Put(wo, handles_[a & 1], RandomString(&rnd, 13), ""));
     }
   };

include/rocksdb/options.h

@@ -515,8 +515,8 @@ struct ColumnFamilyOptions {
   unsigned int rate_limit_delay_max_milliseconds;
 
   // size of one block in arena memory allocation.
-  // If <= 0, a proper value is automatically calculated (usually 1/10 of
+  // If <= 0, a proper value is automatically calculated (usually 1/8 of
-  // writer_buffer_size).
+  // writer_buffer_size, rounded up to a multiple of 4KB).
   //
   // There are two additonal restriction of the The specified size:
   // (1) size should be in the range of [4096, 2 << 30] and