Make it possible to enable blob files starting from a certain LSM tree level (#10077)

Summary:
Currently, if blob files are enabled (i.e. `enable_blob_files` is true), large values are extracted both during flush/recovery (when SST files are written into level 0 of the LSM tree) and during compaction into any LSM tree level. For certain use cases that have a mix of short-lived and long-lived values, it might make sense to support extracting large values only during compactions whose output level is greater than or equal to a specified LSM tree level (e.g. compactions into L1/L2/... or above). This could reduce the space amplification caused by large values that are turned into garbage shortly after being written at the price of some write amplification incurred by long-lived values whose extraction to blob files is delayed.

In order to achieve this, we would like to do the following:
- Add a new configuration option `blob_file_starting_level` (default: 0) to `AdvancedColumnFamilyOptions` (and `MutableCFOptions` and extend the related logic)
- Instantiate `BlobFileBuilder` in `BuildTable` (used during flush and recovery, where the LSM tree level is L0) and `CompactionJob` iff `enable_blob_files` is set and the LSM tree level is `>= blob_file_starting_level`
- Add unit tests for the new functionality, and add the new option to our stress tests (`db_stress` and `db_crashtest.py` )
- Add the new option to our benchmarking tool `db_bench` and the BlobDB benchmark script `run_blob_bench.sh`
- Add the new option to the `ldb` tool (see https://github.com/facebook/rocksdb/wiki/Administration-and-Data-Access-Tool)
- Ideally extend the C and Java bindings with the new option
- Update the BlobDB wiki to document the new option.

Pull Request resolved: https://github.com/facebook/rocksdb/pull/10077

Reviewed By: ltamasi

Differential Revision: D36884156

Pulled By: gangliao

fbshipit-source-id: 942bab025f04633edca8564ed64791cb5e31627d

HISTORY.md

@@ -10,6 +10,8 @@
 * Add transaction `get_pinned` and `multi_get` to C API.
 * Add two-phase commit support to C API.
 * Add `rocksdb_transaction_get_writebatch_wi` and `rocksdb_transaction_rebuild_from_writebatch` to C API.
+* Add `rocksdb_options_get_blob_file_starting_level` and `rocksdb_options_set_blob_file_starting_level` to C API.
+* Add `blobFileStartingLevel` and `setBlobFileStartingLevel` to Java API.
 * Add SingleDelete for DB in C API
 * Add User Defined Timestamp in C API.
   * `rocksdb_comparator_with_ts_create` to create timestamp aware comparator
@@ -20,6 +22,7 @@
 * Add FileSystem::ReadAsync API in io_tracing
 * Add blob garbage collection parameters `blob_garbage_collection_policy` and `blob_garbage_collection_age_cutoff` to both force-enable and force-disable GC, as well as selectively override age cutoff when using CompactRange. 
 * Add an extra sanity check in `GetSortedWalFiles()` (also used by `GetLiveFilesStorageInfo()`, `BackupEngine`, and `Checkpoint`) to reduce risk of successfully created backup or checkpoint failing to open because of missing WAL file.
+* Add a new column family option `blob_file_starting_level` to enable writing blob files during flushes and compactions starting from the specified LSM tree level.
 
 ### Behavior changes
 * DB::Open(), DB::OpenAsSecondary() will fail if a Logger cannot be created (#9984)

db/blob/db_blob_compaction_test.cc

@@ -243,6 +243,82 @@ TEST_F(DBBlobCompactionTest, FilterByKeyLength) {
   Close();
 }
 
+TEST_F(DBBlobCompactionTest, BlobCompactWithStartingLevel) {
+  Options options = GetDefaultOptions();
+
+  options.enable_blob_files = true;
+  options.min_blob_size = 1000;
+  options.blob_file_starting_level = 5;
+  options.create_if_missing = true;
+
+  // Open DB with fixed-prefix sst-partitioner so that compaction will cut
+  // new table file when encountering a new key whose 1-byte prefix changes.
+  constexpr size_t key_len = 1;
+  options.sst_partitioner_factory =
+      NewSstPartitionerFixedPrefixFactory(key_len);
+
+  ASSERT_OK(TryReopen(options));
+
+  constexpr size_t blob_size = 3000;
+
+  constexpr char first_key[] = "a";
+  const std::string first_blob(blob_size, 'a');
+  ASSERT_OK(Put(first_key, first_blob));
+
+  constexpr char second_key[] = "b";
+  const std::string second_blob(2 * blob_size, 'b');
+  ASSERT_OK(Put(second_key, second_blob));
+
+  constexpr char third_key[] = "d";
+  const std::string third_blob(blob_size, 'd');
+  ASSERT_OK(Put(third_key, third_blob));
+
+  ASSERT_OK(Flush());
+
+  constexpr char fourth_key[] = "c";
+  const std::string fourth_blob(blob_size, 'c');
+  ASSERT_OK(Put(fourth_key, fourth_blob));
+
+  ASSERT_OK(Flush());
+
+  ASSERT_EQ(0, GetBlobFileNumbers().size());
+  ASSERT_EQ(2, NumTableFilesAtLevel(/*level=*/0));
+  ASSERT_EQ(0, NumTableFilesAtLevel(/*level=*/1));
+
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), /*begin=*/nullptr,
+                              /*end=*/nullptr));
+
+  // No blob file should be created since blob_file_starting_level is 5.
+  ASSERT_EQ(0, GetBlobFileNumbers().size());
+  ASSERT_EQ(0, NumTableFilesAtLevel(/*level=*/0));
+  ASSERT_EQ(4, NumTableFilesAtLevel(/*level=*/1));
+
+  {
+    options.blob_file_starting_level = 1;
+    DestroyAndReopen(options);
+
+    ASSERT_OK(Put(first_key, first_blob));
+    ASSERT_OK(Put(second_key, second_blob));
+    ASSERT_OK(Put(third_key, third_blob));
+    ASSERT_OK(Flush());
+    ASSERT_OK(Put(fourth_key, fourth_blob));
+    ASSERT_OK(Flush());
+
+    ASSERT_EQ(0, GetBlobFileNumbers().size());
+    ASSERT_EQ(2, NumTableFilesAtLevel(/*level=*/0));
+    ASSERT_EQ(0, NumTableFilesAtLevel(/*level=*/1));
+
+    ASSERT_OK(db_->CompactRange(CompactRangeOptions(), /*begin=*/nullptr,
+                                /*end=*/nullptr));
+    // The compaction's output level equals to blob_file_starting_level.
+    ASSERT_EQ(1, GetBlobFileNumbers().size());
+    ASSERT_EQ(0, NumTableFilesAtLevel(/*level=*/0));
+    ASSERT_EQ(4, NumTableFilesAtLevel(/*level=*/1));
+  }
+
+  Close();
+}
+
 TEST_F(DBBlobCompactionTest, BlindWriteFilter) {
   Options options = GetDefaultOptions();
   options.enable_blob_files = true;

db/builder.cc

@@ -181,7 +181,10 @@ Status BuildTable(
         snapshots.empty() ? 0 : snapshots.back(), snapshot_checker);
 
     std::unique_ptr<BlobFileBuilder> blob_file_builder(
-        (mutable_cf_options.enable_blob_files && blob_file_additions)
+        (mutable_cf_options.enable_blob_files &&
+         tboptions.level_at_creation >=
+             mutable_cf_options.blob_file_starting_level &&
+         blob_file_additions)
             ? new BlobFileBuilder(
                   versions, fs, &ioptions, &mutable_cf_options, &file_options,
                   job_id, tboptions.column_family_id,

db/c.cc

@@ -3039,6 +3039,15 @@ uint64_t rocksdb_options_get_blob_compaction_readahead_size(
   return opt->rep.blob_compaction_readahead_size;
 }
 
+void rocksdb_options_set_blob_file_starting_level(rocksdb_options_t* opt,
+                                                  int val) {
+  opt->rep.blob_file_starting_level = val;
+}
+
+int rocksdb_options_get_blob_file_starting_level(rocksdb_options_t* opt) {
+  return opt->rep.blob_file_starting_level;
+}
+
 void rocksdb_options_set_num_levels(rocksdb_options_t* opt, int n) {
   opt->rep.num_levels = n;
 }

db/c_test.c

@@ -1857,6 +1857,9 @@ int main(int argc, char** argv) {
     CheckCondition(262144 ==
                    rocksdb_options_get_blob_compaction_readahead_size(o));
 
+    rocksdb_options_set_blob_file_starting_level(o, 5);
+    CheckCondition(5 == rocksdb_options_get_blob_file_starting_level(o));
+
     // Create a copy that should be equal to the original.
     rocksdb_options_t* copy;
     copy = rocksdb_options_create_copy(o);

db/compaction/compaction_job.cc

@@ -1454,7 +1454,9 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   std::vector<std::string> blob_file_paths;
 
   std::unique_ptr<BlobFileBuilder> blob_file_builder(
-      mutable_cf_options->enable_blob_files
+      (mutable_cf_options->enable_blob_files &&
+       sub_compact->compaction->output_level() >=
+           mutable_cf_options->blob_file_starting_level)
           ? new BlobFileBuilder(
                 versions_, fs_.get(),
                 sub_compact->compaction->immutable_options(),

db_stress_tool/db_stress_common.h

@@ -266,6 +266,7 @@ DECLARE_bool(enable_blob_garbage_collection);
 DECLARE_double(blob_garbage_collection_age_cutoff);
 DECLARE_double(blob_garbage_collection_force_threshold);
 DECLARE_uint64(blob_compaction_readahead_size);
+DECLARE_int32(blob_file_starting_level);
 
 DECLARE_int32(approximate_size_one_in);
 DECLARE_bool(sync_fault_injection);

db_stress_tool/db_stress_gflags.cc

@@ -39,8 +39,8 @@ DEFINE_string(key_len_percent_dist, "",
               "24 bytes. If not specified, it will be evenly distributed");
 
 DEFINE_int32(key_window_scale_factor, 10,
-              "This value will be multiplied by 100 to come up with a window "
-              "size for varying the key length");
+             "This value will be multiplied by 100 to come up with a window "
+             "size for varying the key length");
 
 DEFINE_int32(column_families, 10, "Number of column families");
 
@@ -439,6 +439,12 @@ DEFINE_uint64(blob_compaction_readahead_size,
                   .blob_compaction_readahead_size,
               "[Integrated BlobDB] Compaction readahead for blob files.");
 
+DEFINE_int32(
+    blob_file_starting_level,
+    ROCKSDB_NAMESPACE::AdvancedColumnFamilyOptions().blob_file_starting_level,
+    "[Integrated BlobDB] Enable writing blob files during flushes and "
+    "compactions starting from the specified level.");
+
 static const bool FLAGS_subcompactions_dummy __attribute__((__unused__)) =
     RegisterFlagValidator(&FLAGS_subcompactions, &ValidateUint32Range);
 
@@ -857,7 +863,7 @@ DEFINE_int32(approximate_size_one_in, 64,
              " random key ranges.");
 
 DEFINE_int32(read_fault_one_in, 1000,
-            "On non-zero, enables fault injection on read");
+             "On non-zero, enables fault injection on read");
 
 DEFINE_int32(get_property_one_in, 1000,
              "If non-zero, then DB::GetProperty() will be called to get various"

db_stress_tool/db_stress_test_base.cc

@@ -280,6 +280,8 @@ bool StressTest::BuildOptionsTable() {
                         std::vector<std::string>{"0.5", "0.75", "1.0"});
     options_tbl.emplace("blob_compaction_readahead_size",
                         std::vector<std::string>{"0", "1M", "4M"});
+    options_tbl.emplace("blob_file_starting_level",
+                        std::vector<std::string>{"0", "1", "2"});
   }
 
   options_table_ = std::move(options_tbl);
@@ -2384,14 +2386,16 @@ void StressTest::Open(SharedState* shared) {
           "Integrated BlobDB: blob files enabled %d, min blob size %" PRIu64
           ", blob file size %" PRIu64
           ", blob compression type %s, blob GC enabled %d, cutoff %f, force "
-          "threshold %f, blob compaction readahead size %" PRIu64 "\n",
+          "threshold %f, blob compaction readahead size %" PRIu64
+          ", blob file starting level %d\n",
           options_.enable_blob_files, options_.min_blob_size,
           options_.blob_file_size,
           CompressionTypeToString(options_.blob_compression_type).c_str(),
           options_.enable_blob_garbage_collection,
           options_.blob_garbage_collection_age_cutoff,
           options_.blob_garbage_collection_force_threshold,
-          options_.blob_compaction_readahead_size);
+          options_.blob_compaction_readahead_size,
+          options_.blob_file_starting_level);
 
   fprintf(stdout, "DB path: [%s]\n", FLAGS_db.c_str());
 
@@ -2987,6 +2991,7 @@ void InitializeOptionsFromFlags(
   options.blob_garbage_collection_force_threshold =
       FLAGS_blob_garbage_collection_force_threshold;
   options.blob_compaction_readahead_size = FLAGS_blob_compaction_readahead_size;
+  options.blob_file_starting_level = FLAGS_blob_file_starting_level;
 
   options.wal_compression =
       StringToCompressionType(FLAGS_wal_compression.c_str());

include/rocksdb/advanced_options.h

@@ -938,6 +938,21 @@ struct AdvancedColumnFamilyOptions {
   // Dynamically changeable through the SetOptions() API
   uint64_t blob_compaction_readahead_size = 0;
 
+  // Enable blob files starting from a certain LSM tree level.
+  //
+  // For certain use cases that have a mix of short-lived and long-lived values,
+  // it might make sense to support extracting large values only during
+  // compactions whose output level is greater than or equal to a specified LSM
+  // tree level (e.g. compactions into L1/L2/... or above). This could reduce
+  // the space amplification caused by large values that are turned into garbage
+  // shortly after being written at the price of some write amplification
+  // incurred by long-lived values whose extraction to blob files is delayed.
+  //
+  // Default: 0
+  //
+  // Dynamically changeable through the SetOptions() API
+  int blob_file_starting_level = 0;
+
   // Create ColumnFamilyOptions with default values for all fields
   AdvancedColumnFamilyOptions();
   // Create ColumnFamilyOptions from Options

include/rocksdb/c.h

@@ -1259,6 +1259,11 @@ rocksdb_options_set_blob_compaction_readahead_size(rocksdb_options_t* opt,
 extern ROCKSDB_LIBRARY_API uint64_t
 rocksdb_options_get_blob_compaction_readahead_size(rocksdb_options_t* opt);
 
+extern ROCKSDB_LIBRARY_API void rocksdb_options_set_blob_file_starting_level(
+    rocksdb_options_t* opt, int val);
+extern ROCKSDB_LIBRARY_API int rocksdb_options_get_blob_file_starting_level(
+    rocksdb_options_t* opt);
+
 /* returns a pointer to a malloc()-ed, null terminated string */
 extern ROCKSDB_LIBRARY_API char* rocksdb_options_statistics_get_string(
     rocksdb_options_t* opt);

include/rocksdb/utilities/ldb_cmd.h

@@ -69,6 +69,7 @@ class LDBCommand {
   static const std::string ARG_BLOB_GARBAGE_COLLECTION_AGE_CUTOFF;
   static const std::string ARG_BLOB_GARBAGE_COLLECTION_FORCE_THRESHOLD;
   static const std::string ARG_BLOB_COMPACTION_READAHEAD_SIZE;
+  static const std::string ARG_BLOB_FILE_STARTING_LEVEL;
   static const std::string ARG_DECODE_BLOB_INDEX;
   static const std::string ARG_DUMP_UNCOMPRESSED_BLOBS;
 

java/rocksjni/options.cc

@@ -3840,6 +3840,28 @@ jlong Java_org_rocksdb_Options_blobCompactionReadaheadSize(JNIEnv*, jobject,
   return static_cast<jlong>(opts->blob_compaction_readahead_size);
 }
 
+/*
+ * Class:     org_rocksdb_Options
+ * Method:    setBlobFileStartingLevel
+ * Signature: (JJ)V
+ */
+void Java_org_rocksdb_Options_setBlobFileStartingLevel(
+    JNIEnv*, jobject, jlong jhandle, jint jblob_file_starting_level) {
+  auto* opts = reinterpret_cast<ROCKSDB_NAMESPACE::Options*>(jhandle);
+  opts->blob_file_starting_level = jblob_file_starting_level;
+}
+
+/*
+ * Class:     org_rocksdb_Options
+ * Method:    blobFileStartingLevel
+ * Signature: (J)J
+ */
+jint Java_org_rocksdb_Options_blobFileStartingLevel(JNIEnv*, jobject,
+                                                    jlong jhandle) {
+  auto* opts = reinterpret_cast<ROCKSDB_NAMESPACE::Options*>(jhandle);
+  return static_cast<jint>(opts->blob_file_starting_level);
+}
+
 //////////////////////////////////////////////////////////////////////////////
 // ROCKSDB_NAMESPACE::ColumnFamilyOptions
 

java/src/main/java/org/rocksdb/AdvancedMutableColumnFamilyOptionsInterface.java

@@ -754,6 +754,29 @@ public interface AdvancedMutableColumnFamilyOptionsInterface<
    */
   long blobCompactionReadaheadSize();
 
+  /**
+   * Set a certain LSM tree level to enable blob files.
+   *
+   * Default: 0
+   *
+   * Dynamically changeable through
+   * {@link RocksDB#setOptions(ColumnFamilyHandle, MutableColumnFamilyOptions)}.
+   *
+   * @param setBlobFileStartingLevel the starting level to enable blob files
+   *
+   * @return the reference to the current options.
+   */
+  T setBlobFileStartingLevel(final int blobFileStartingLevel);
+
+  /**
+   * Get the starting LSM tree level to enable blob files.
+   *
+   * Default: 0
+   *
+   * @return the current LSM tree level to enable blob files.
+   */
+  int blobFileStartingLevel();
+
   //
   // END options for blobs (integrated BlobDB)
   //

java/src/main/java/org/rocksdb/ColumnFamilyOptions.java

@@ -1238,6 +1238,36 @@ public class ColumnFamilyOptions extends RocksObject
     return blobCompactionReadaheadSize(nativeHandle_);
   }
 
+  /**
+   * Set a certain LSM tree level to enable blob files.
+   *
+   * Default: 0
+   *
+   * Dynamically changeable through
+   * {@link RocksDB#setOptions(ColumnFamilyHandle, MutableColumnFamilyOptions)}.
+   *
+   * @param setBlobFileStartingLevel the starting level to enable blob files
+   *
+   * @return the reference to the current options.
+   */
+  @Override
+  public ColumnFamilyOptions setBlobFileStartingLevel(final int blobFileStartingLevel) {
+    setBlobFileStartingLevel(nativeHandle_, blobFileStartingLevel);
+    return this;
+  }
+
+  /**
+   * Get the starting LSM tree level to enable blob files.
+   *
+   * Default: 0
+   *
+   * @return the current LSM tree level to enable blob files.
+   */
+  @Override
+  public int blobFileStartingLevel() {
+    return blobFileStartingLevel(nativeHandle_);
+  }
+
   //
   // END options for blobs (integrated BlobDB)
   //
@@ -1440,6 +1470,9 @@ public class ColumnFamilyOptions extends RocksObject
   private native void setBlobCompactionReadaheadSize(
       final long nativeHandle_, final long blobCompactionReadaheadSize);
   private native long blobCompactionReadaheadSize(final long nativeHandle_);
+  private native void setBlobFileStartingLevel(
+      final long nativeHandle_, final int blobFileStartingLevel);
+  private native int blobFileStartingLevel(final long nativeHandle_);
 
   // instance variables
   // NOTE: If you add new member variables, please update the copy constructor above!

java/src/main/java/org/rocksdb/MutableColumnFamilyOptions.java

@@ -121,7 +121,8 @@ public class MutableColumnFamilyOptions
     enable_blob_garbage_collection(ValueType.BOOLEAN),
     blob_garbage_collection_age_cutoff(ValueType.DOUBLE),
     blob_garbage_collection_force_threshold(ValueType.DOUBLE),
-    blob_compaction_readahead_size(ValueType.LONG);
+    blob_compaction_readahead_size(ValueType.LONG),
+    blob_file_starting_level(ValueType.INT);
 
     private final ValueType valueType;
     BlobOption(final ValueType valueType) {
@@ -582,5 +583,16 @@ public class MutableColumnFamilyOptions
     public long blobCompactionReadaheadSize() {
       return getLong(BlobOption.blob_compaction_readahead_size);
     }
+
+    @Override
+    public MutableColumnFamilyOptionsBuilder setBlobFileStartingLevel(
+        final int blobFileStartingLevel) {
+      return setInt(BlobOption.blob_file_starting_level, blobFileStartingLevel);
+    }
+
+    @Override
+    public int blobFileStartingLevel() {
+      return getInt(BlobOption.blob_file_starting_level);
+    }
   }
 }

java/src/main/java/org/rocksdb/Options.java

@@ -2082,6 +2082,17 @@ public class Options extends RocksObject
     return blobCompactionReadaheadSize(nativeHandle_);
   }
 
+  @Override
+  public Options setBlobFileStartingLevel(final int blobFileStartingLevel) {
+    setBlobFileStartingLevel(nativeHandle_, blobFileStartingLevel);
+    return this;
+  }
+
+  @Override
+  public int blobFileStartingLevel() {
+    return blobFileStartingLevel(nativeHandle_);
+  }
+
   //
   // END options for blobs (integrated BlobDB)
   //
@@ -2513,6 +2524,9 @@ public class Options extends RocksObject
   private native void setBlobCompactionReadaheadSize(
       final long nativeHandle_, final long blobCompactionReadaheadSize);
   private native long blobCompactionReadaheadSize(final long nativeHandle_);
+  private native void setBlobFileStartingLevel(
+      final long nativeHandle_, final int blobFileStartingLevel);
+  private native int blobFileStartingLevel(final long nativeHandle_);
 
   // instance variables
   // NOTE: If you add new member variables, please update the copy constructor above!

java/src/test/java/org/rocksdb/BlobOptionsTest.java

@@ -89,6 +89,7 @@ public class BlobOptionsTest {
       assertThat(options.setBlobGarbageCollectionAgeCutoff(0.89)).isEqualTo(options);
       assertThat(options.setBlobGarbageCollectionForceThreshold(0.80)).isEqualTo(options);
       assertThat(options.setBlobCompactionReadaheadSize(262144L)).isEqualTo(options);
+      assertThat(options.setBlobFileStartingLevel(0)).isEqualTo(options);
 
       assertThat(options.enableBlobFiles()).isEqualTo(true);
       assertThat(options.minBlobSize()).isEqualTo(132768L);
@@ -98,6 +99,7 @@ public class BlobOptionsTest {
       assertThat(options.blobGarbageCollectionAgeCutoff()).isEqualTo(0.89);
       assertThat(options.blobGarbageCollectionForceThreshold()).isEqualTo(0.80);
       assertThat(options.blobCompactionReadaheadSize()).isEqualTo(262144L);
+      assertThat(options.blobFileStartingLevel()).isEqualTo(0);
     }
   }
 
@@ -127,6 +129,7 @@ public class BlobOptionsTest {
           .isEqualTo(columnFamilyOptions);
       assertThat(columnFamilyOptions.setBlobCompactionReadaheadSize(262144L))
           .isEqualTo(columnFamilyOptions);
+      assertThat(columnFamilyOptions.setBlobFileStartingLevel(0)).isEqualTo(columnFamilyOptions);
 
       assertThat(columnFamilyOptions.enableBlobFiles()).isEqualTo(true);
       assertThat(columnFamilyOptions.minBlobSize()).isEqualTo(132768L);
@@ -137,6 +140,7 @@ public class BlobOptionsTest {
       assertThat(columnFamilyOptions.blobGarbageCollectionAgeCutoff()).isEqualTo(0.89);
       assertThat(columnFamilyOptions.blobGarbageCollectionForceThreshold()).isEqualTo(0.80);
       assertThat(columnFamilyOptions.blobCompactionReadaheadSize()).isEqualTo(262144L);
+      assertThat(columnFamilyOptions.blobFileStartingLevel()).isEqualTo(0);
     }
   }
 
@@ -151,7 +155,8 @@ public class BlobOptionsTest {
         .setEnableBlobGarbageCollection(true)
         .setBlobGarbageCollectionAgeCutoff(0.89)
         .setBlobGarbageCollectionForceThreshold(0.80)
-        .setBlobCompactionReadaheadSize(262144);
+        .setBlobCompactionReadaheadSize(262144)
+        .setBlobFileStartingLevel(1);
 
     assertThat(builder.enableBlobFiles()).isEqualTo(true);
     assertThat(builder.minBlobSize()).isEqualTo(1024);
@@ -161,6 +166,7 @@ public class BlobOptionsTest {
     assertThat(builder.blobGarbageCollectionAgeCutoff()).isEqualTo(0.89);
     assertThat(builder.blobGarbageCollectionForceThreshold()).isEqualTo(0.80);
     assertThat(builder.blobCompactionReadaheadSize()).isEqualTo(262144);
+    assertThat(builder.blobFileStartingLevel()).isEqualTo(1);
 
     builder.setEnableBlobFiles(false)
         .setMinBlobSize(4096)
@@ -169,7 +175,8 @@ public class BlobOptionsTest {
         .setEnableBlobGarbageCollection(false)
         .setBlobGarbageCollectionAgeCutoff(0.91)
         .setBlobGarbageCollectionForceThreshold(0.96)
-        .setBlobCompactionReadaheadSize(1024);
+        .setBlobCompactionReadaheadSize(1024)
+        .setBlobFileStartingLevel(0);
 
     assertThat(builder.enableBlobFiles()).isEqualTo(false);
     assertThat(builder.minBlobSize()).isEqualTo(4096);
@@ -179,13 +186,14 @@ public class BlobOptionsTest {
     assertThat(builder.blobGarbageCollectionAgeCutoff()).isEqualTo(0.91);
     assertThat(builder.blobGarbageCollectionForceThreshold()).isEqualTo(0.96);
     assertThat(builder.blobCompactionReadaheadSize()).isEqualTo(1024);
+    assertThat(builder.blobFileStartingLevel()).isEqualTo(0);
 
     final MutableColumnFamilyOptions options = builder.build();
     assertThat(options.getKeys())
         .isEqualTo(new String[] {"enable_blob_files", "min_blob_size", "blob_file_size",
             "blob_compression_type", "enable_blob_garbage_collection",
             "blob_garbage_collection_age_cutoff", "blob_garbage_collection_force_threshold",
-            "blob_compaction_readahead_size"});
+            "blob_compaction_readahead_size", "blob_file_starting_level"});
     assertThat(options.getValues())
         .isEqualTo(new String[] {
             "false", "4096", "2048", "LZ4_COMPRESSION", "false", "0.91", "0.96", "1024"});

java/src/test/java/org/rocksdb/MutableColumnFamilyOptionsTest.java

@@ -97,7 +97,7 @@ public class MutableColumnFamilyOptionsTest {
     final String optionsString =
         "bottommost_compression=kDisableCompressionOption;  sample_for_compression=0;  "
         + "blob_garbage_collection_age_cutoff=0.250000;  blob_garbage_collection_force_threshold=0.800000;  arena_block_size=1048576;  enable_blob_garbage_collection=false;  "
-        + "level0_stop_writes_trigger=36;  min_blob_size=65536;  blob_compaction_readahead_size=262144;  "
+        + "level0_stop_writes_trigger=36;  min_blob_size=65536;  blob_compaction_readahead_size=262144;  blob_file_starting_level=5;  "
         + "compaction_options_universal={allow_trivial_move=false;stop_style=kCompactionStopStyleTotalSize;min_merge_width=2;"
         + "compression_size_percent=-1;max_size_amplification_percent=200;max_merge_width=4294967295;size_ratio=1;};  "
         + "target_file_size_base=67108864;  max_bytes_for_level_base=268435456;  memtable_whole_key_filtering=false;  "
@@ -132,6 +132,7 @@ public class MutableColumnFamilyOptionsTest {
     assertThat(cf.level0StopWritesTrigger()).isEqualTo(36);
     assertThat(cf.minBlobSize()).isEqualTo(65536);
     assertThat(cf.blobCompactionReadaheadSize()).isEqualTo(262144);
+    assertThat(cf.blobFileStartingLevel()).isEqualTo(5);
     assertThat(cf.targetFileSizeBase()).isEqualTo(67108864);
     assertThat(cf.maxBytesForLevelBase()).isEqualTo(268435456);
     assertThat(cf.softPendingCompactionBytesLimit()).isEqualTo(68719476736L);

java/src/test/java/org/rocksdb/MutableOptionsGetSetTest.java

@@ -47,6 +47,7 @@ public class MutableOptionsGetSetTest {
                    .setBlobGarbageCollectionAgeCutoff(0.25)
                    .setBlobGarbageCollectionForceThreshold(0.80)
                    .setBlobCompactionReadaheadSize(262144)
+                   .setBlobFileStartingLevel(2)
                    .setArenaBlockSize(42)
                    .setMemtablePrefixBloomSizeRatio(0.17)
                    .setMemtableWholeKeyFiltering(false)
@@ -106,6 +107,7 @@ public class MutableOptionsGetSetTest {
         assertThat(builder1.blobGarbageCollectionAgeCutoff()).isEqualTo(0.25);
         assertThat(builder1.blobGarbageCollectionForceThreshold()).isEqualTo(0.80);
         assertThat(builder1.blobCompactionReadaheadSize()).isEqualTo(262144);
+        assertThat(builder1.blobFileStartingLevel()).isEqualTo(2);
         assertThat(builder1.minBlobSize()).isEqualTo(minBlobSize);
         assertThat(builder1.arenaBlockSize()).isEqualTo(42);
         assertThat(builder1.memtablePrefixBloomSizeRatio()).isEqualTo(0.17);
@@ -197,6 +199,7 @@ public class MutableOptionsGetSetTest {
                 .setBlobGarbageCollectionAgeCutoff(0.25)
                 .setBlobGarbageCollectionForceThreshold(0.80)
                 .setBlobCompactionReadaheadSize(262144)
+                .setBlobFileStartingLevel(3)
                 .setArenaBlockSize(42)
                 .setMemtablePrefixBloomSizeRatio(0.17)
                 .setMemtableWholeKeyFiltering(false)
@@ -222,6 +225,7 @@ public class MutableOptionsGetSetTest {
         assertThat(builder1.blobGarbageCollectionAgeCutoff()).isEqualTo(0.25);
         assertThat(builder1.blobGarbageCollectionForceThreshold()).isEqualTo(0.80);
         assertThat(builder1.blobCompactionReadaheadSize()).isEqualTo(262144);
+        assertThat(builder1.blobFileStartingLevel()).isEqualTo(3);
         assertThat(builder1.minBlobSize()).isEqualTo(minBlobSize);
         assertThat(builder1.arenaBlockSize()).isEqualTo(42);
         assertThat(builder1.memtablePrefixBloomSizeRatio()).isEqualTo(0.17);
@@ -317,6 +321,7 @@ public class MutableOptionsGetSetTest {
               .setBlobGarbageCollectionAgeCutoff(0.25)
               .setBlobGarbageCollectionForceThreshold(0.80)
               .setBlobCompactionReadaheadSize(131072)
+              .setBlobFileStartingLevel(4)
               .setArenaBlockSize(42)
               .setMemtablePrefixBloomSizeRatio(0.17)
               .setMemtableWholeKeyFiltering(false)
@@ -341,6 +346,7 @@ public class MutableOptionsGetSetTest {
       assertThat(builder1.blobGarbageCollectionAgeCutoff()).isEqualTo(0.25);
       assertThat(builder1.blobGarbageCollectionForceThreshold()).isEqualTo(0.80);
       assertThat(builder1.blobCompactionReadaheadSize()).isEqualTo(131072);
+      assertThat(builder1.blobFileStartingLevel()).isEqualTo(4);
       assertThat(builder1.minBlobSize()).isEqualTo(minBlobSize);
       assertThat(builder1.arenaBlockSize()).isEqualTo(42);
       assertThat(builder1.memtablePrefixBloomSizeRatio()).isEqualTo(0.17);

options/cf_options.cc

@@ -447,6 +447,10 @@ static std::unordered_map<std::string, OptionTypeInfo>
          {offsetof(struct MutableCFOptions, blob_compaction_readahead_size),
           OptionType::kUInt64T, OptionVerificationType::kNormal,
           OptionTypeFlags::kMutable}},
+        {"blob_file_starting_level",
+         {offsetof(struct MutableCFOptions, blob_file_starting_level),
+          OptionType::kInt, OptionVerificationType::kNormal,
+          OptionTypeFlags::kMutable}},
         {"sample_for_compression",
          {offsetof(struct MutableCFOptions, sample_for_compression),
           OptionType::kUInt64T, OptionVerificationType::kNormal,
@@ -1067,6 +1071,8 @@ void MutableCFOptions::Dump(Logger* log) const {
                  blob_garbage_collection_force_threshold);
   ROCKS_LOG_INFO(log, "           blob_compaction_readahead_size: %" PRIu64,
                  blob_compaction_readahead_size);
+  ROCKS_LOG_INFO(log, "                 blob_file_starting_level: %d",
+                 blob_file_starting_level);
 
   ROCKS_LOG_INFO(log, "                   bottommost_temperature: %d",
                  static_cast<int>(bottommost_temperature));

options/cf_options.h

@@ -140,6 +140,7 @@ struct MutableCFOptions {
         blob_garbage_collection_force_threshold(
             options.blob_garbage_collection_force_threshold),
         blob_compaction_readahead_size(options.blob_compaction_readahead_size),
+        blob_file_starting_level(options.blob_file_starting_level),
         max_sequential_skip_in_iterations(
             options.max_sequential_skip_in_iterations),
         check_flush_compaction_key_order(
@@ -189,6 +190,7 @@ struct MutableCFOptions {
         blob_garbage_collection_age_cutoff(0.0),
         blob_garbage_collection_force_threshold(0.0),
         blob_compaction_readahead_size(0),
+        blob_file_starting_level(0),
         max_sequential_skip_in_iterations(0),
         check_flush_compaction_key_order(true),
         paranoid_file_checks(false),
@@ -255,6 +257,7 @@ struct MutableCFOptions {
   double blob_garbage_collection_age_cutoff;
   double blob_garbage_collection_force_threshold;
   uint64_t blob_compaction_readahead_size;
+  int blob_file_starting_level;
 
   // Misc options
   uint64_t max_sequential_skip_in_iterations;

options/options.cc

@@ -100,7 +100,8 @@ AdvancedColumnFamilyOptions::AdvancedColumnFamilyOptions(const Options& options)
           options.blob_garbage_collection_age_cutoff),
       blob_garbage_collection_force_threshold(
           options.blob_garbage_collection_force_threshold),
-      blob_compaction_readahead_size(options.blob_compaction_readahead_size) {
+      blob_compaction_readahead_size(options.blob_compaction_readahead_size),
+      blob_file_starting_level(options.blob_file_starting_level) {
   assert(memtable_factory.get() != nullptr);
   if (max_bytes_for_level_multiplier_additional.size() <
       static_cast<unsigned int>(num_levels)) {
@@ -414,6 +415,8 @@ void ColumnFamilyOptions::Dump(Logger* log) const {
     ROCKS_LOG_HEADER(
         log, "         Options.blob_compaction_readahead_size: %" PRIu64,
         blob_compaction_readahead_size);
+    ROCKS_LOG_HEADER(log, "               Options.blob_file_starting_level: %d",
+                     blob_file_starting_level);
 }  // ColumnFamilyOptions::Dump
 
 void Options::Dump(Logger* log) const {

options/options_helper.cc

@@ -254,6 +254,7 @@ void UpdateColumnFamilyOptions(const MutableCFOptions& moptions,
       moptions.blob_garbage_collection_force_threshold;
   cf_opts->blob_compaction_readahead_size =
       moptions.blob_compaction_readahead_size;
+  cf_opts->blob_file_starting_level = moptions.blob_file_starting_level;
 
   // Misc options
   cf_opts->max_sequential_skip_in_iterations =

options/options_settable_test.cc

@@ -520,6 +520,7 @@ TEST_F(OptionsSettableTest, ColumnFamilyOptionsAllFieldsSettable) {
       "blob_garbage_collection_age_cutoff=0.5;"
       "blob_garbage_collection_force_threshold=0.75;"
       "blob_compaction_readahead_size=262144;"
+      "blob_file_starting_level=1;"
       "bottommost_temperature=kWarm;"
       "compaction_options_fifo={max_table_files_size=3;allow_"
       "compaction=false;age_for_warm=1;};",

options/options_test.cc

@@ -109,6 +109,7 @@ TEST_F(OptionsTest, GetOptionsFromMapTest) {
       {"blob_garbage_collection_age_cutoff", "0.5"},
       {"blob_garbage_collection_force_threshold", "0.75"},
       {"blob_compaction_readahead_size", "256K"},
+      {"blob_file_starting_level", "1"},
       {"bottommost_temperature", "kWarm"},
   };
 
@@ -247,6 +248,7 @@ TEST_F(OptionsTest, GetOptionsFromMapTest) {
   ASSERT_EQ(new_cf_opt.blob_garbage_collection_age_cutoff, 0.5);
   ASSERT_EQ(new_cf_opt.blob_garbage_collection_force_threshold, 0.75);
   ASSERT_EQ(new_cf_opt.blob_compaction_readahead_size, 262144);
+  ASSERT_EQ(new_cf_opt.blob_file_starting_level, 1);
   ASSERT_EQ(new_cf_opt.bottommost_temperature, Temperature::kWarm);
 
   cf_options_map["write_buffer_size"] = "hello";
@@ -2330,6 +2332,7 @@ TEST_F(OptionsOldApiTest, GetOptionsFromMapTest) {
       {"blob_garbage_collection_age_cutoff", "0.5"},
       {"blob_garbage_collection_force_threshold", "0.75"},
       {"blob_compaction_readahead_size", "256K"},
+      {"blob_file_starting_level", "1"},
       {"bottommost_temperature", "kWarm"},
   };
 
@@ -2462,6 +2465,7 @@ TEST_F(OptionsOldApiTest, GetOptionsFromMapTest) {
   ASSERT_EQ(new_cf_opt.blob_garbage_collection_age_cutoff, 0.5);
   ASSERT_EQ(new_cf_opt.blob_garbage_collection_force_threshold, 0.75);
   ASSERT_EQ(new_cf_opt.blob_compaction_readahead_size, 262144);
+  ASSERT_EQ(new_cf_opt.blob_file_starting_level, 1);
   ASSERT_EQ(new_cf_opt.bottommost_temperature, Temperature::kWarm);
 
   cf_options_map["write_buffer_size"] = "hello";

tools/db_bench_tool.cc

@@ -1069,6 +1069,11 @@ DEFINE_uint64(blob_compaction_readahead_size,
                   .blob_compaction_readahead_size,
               "[Integrated BlobDB] Compaction readahead for blob files.");
 
+DEFINE_int32(
+    blob_file_starting_level,
+    ROCKSDB_NAMESPACE::AdvancedColumnFamilyOptions().blob_file_starting_level,
+    "[Integrated BlobDB] The starting level for blob files.");
+
 #ifndef ROCKSDB_LITE
 
 // Secondary DB instance Options
@@ -4441,6 +4446,7 @@ class Benchmark {
         FLAGS_blob_garbage_collection_force_threshold;
     options.blob_compaction_readahead_size =
         FLAGS_blob_compaction_readahead_size;
+    options.blob_file_starting_level = FLAGS_blob_file_starting_level;
 
 #ifndef ROCKSDB_LITE
     if (FLAGS_readonly && FLAGS_transaction_db) {

tools/db_bench_tool_test.cc

@@ -275,6 +275,7 @@ const std::string options_file_content = R"OPTIONS_FILE(
   blob_garbage_collection_age_cutoff=0.5
   blob_garbage_collection_force_threshold=0.75
   blob_compaction_readahead_size=262144
+  blob_file_starting_level=0
 
 [TableOptions/BlockBasedTable "default"]
   format_version=0

tools/db_crashtest.py

@@ -339,6 +339,7 @@ blob_params = {
     "blob_garbage_collection_age_cutoff": lambda: random.choice([0.0, 0.25, 0.5, 0.75, 1.0]),
     "blob_garbage_collection_force_threshold": lambda: random.choice([0.5, 0.75, 1.0]),
     "blob_compaction_readahead_size": lambda: random.choice([0, 1048576, 4194304]),
+    "blob_file_starting_level": lambda: random.choice([0] * 4 + [1] * 3 + [2] * 2 + [3]),
 }
 
 ts_params = {

tools/ldb_cmd.cc

@@ -100,6 +100,8 @@ const std::string LDBCommand::ARG_BLOB_GARBAGE_COLLECTION_FORCE_THRESHOLD =
     "blob_garbage_collection_force_threshold";
 const std::string LDBCommand::ARG_BLOB_COMPACTION_READAHEAD_SIZE =
     "blob_compaction_readahead_size";
+const std::string LDBCommand::ARG_BLOB_FILE_STARTING_LEVEL =
+    "blob_file_starting_level";
 const std::string LDBCommand::ARG_DECODE_BLOB_INDEX = "decode_blob_index";
 const std::string LDBCommand::ARG_DUMP_UNCOMPRESSED_BLOBS =
     "dump_uncompressed_blobs";
@@ -553,6 +555,7 @@ std::vector<std::string> LDBCommand::BuildCmdLineOptions(
                                   ARG_BLOB_GARBAGE_COLLECTION_AGE_CUTOFF,
                                   ARG_BLOB_GARBAGE_COLLECTION_FORCE_THRESHOLD,
                                   ARG_BLOB_COMPACTION_READAHEAD_SIZE,
+                                  ARG_BLOB_FILE_STARTING_LEVEL,
                                   ARG_IGNORE_UNKNOWN_OPTIONS,
                                   ARG_CF_NAME};
   ret.insert(ret.end(), options.begin(), options.end());
@@ -819,6 +822,17 @@ void LDBCommand::OverrideBaseCFOptions(ColumnFamilyOptions* cf_opts) {
     }
   }
 
+  int blob_file_starting_level;
+  if (ParseIntOption(option_map_, ARG_BLOB_FILE_STARTING_LEVEL,
+                     blob_file_starting_level, exec_state_)) {
+    if (blob_file_starting_level >= 0) {
+      cf_opts->blob_file_starting_level = blob_file_starting_level;
+    } else {
+      exec_state_ = LDBCommandExecuteResult::Failed(
+          ARG_BLOB_FILE_STARTING_LEVEL + " must be >= 0.");
+    }
+  }
+
   auto itr = option_map_.find(ARG_AUTO_COMPACTION);
   if (itr != option_map_.end()) {
     cf_opts->disable_auto_compactions = !StringToBool(itr->second);

tools/ldb_test.py

@@ -191,6 +191,24 @@ class LDBTestCase(unittest.TestCase):
         blob_files = self.getBlobFiles(dbPath)
         self.assertTrue(len(blob_files) >= 1)
 
+    def testBlobStartingLevel(self):
+        print("Running testBlobStartingLevel...")
+
+        dbPath = os.path.join(self.TMP_DIR, self.DB_NAME)
+        self.assertRunOK("put --create_if_missing --enable_blob_files --blob_file_starting_level=10 x1 y1", "OK")
+        self.assertRunOK("get x1", "y1")
+
+        blob_files = self.getBlobFiles(dbPath)
+        self.assertTrue(len(blob_files) == 0)
+
+        self.assertRunOK("put --enable_blob_files --blob_file_starting_level=0 x2 y2", "OK")
+        self.assertRunOK("get x1", "y1")
+        self.assertRunOK("get x2", "y2")
+        self.assertRunFAIL("get x3")
+
+        blob_files = self.getBlobFiles(dbPath)
+        self.assertTrue(len(blob_files) >= 1)
+
     def testCountDelimDump(self):
         print("Running testCountDelimDump...")
         self.assertRunOK("batchput x.1 x1 --create_if_missing", "OK")

tools/run_blob_bench.sh

@@ -54,6 +54,7 @@ function display_usage() {
   echo -e "\tBLOB_GC_AGE_CUTOFF\t\tBlob garbage collection age cutoff (default: 0.25)"
   echo -e "\tBLOB_GC_FORCE_THRESHOLD\t\tThreshold for forcing garbage collection of the oldest blob files (default: 1.0)"
   echo -e "\tBLOB_COMPACTION_READAHEAD_SIZE\tBlob compaction readahead size (default: 0)"
+  echo -e "\tBLOB_FILE_STARTING_LEVEL\t\tBlob file starting level (default: 0)"
   echo -e "\tTARGET_FILE_SIZE_BASE\t\tTarget SST file size for compactions (default: write buffer size, scaled down if blob files are enabled)"
   echo -e "\tMAX_BYTES_FOR_LEVEL_BASE\tMaximum size for the base level (default: 8 * target SST file size)"
 }
@@ -113,6 +114,7 @@ enable_blob_garbage_collection=${ENABLE_BLOB_GC:-1}
 blob_garbage_collection_age_cutoff=${BLOB_GC_AGE_CUTOFF:-0.25}
 blob_garbage_collection_force_threshold=${BLOB_GC_FORCE_THRESHOLD:-1.0}
 blob_compaction_readahead_size=${BLOB_COMPACTION_READAHEAD_SIZE:-0}
+blob_file_starting_level=${BLOB_FILE_STARTING_LEVEL:-0}
 
 if [ "$enable_blob_files" == "1" ]; then
   target_file_size_base=${TARGET_FILE_SIZE_BASE:-$((32 * write_buffer_size / value_size))}
@@ -142,6 +144,7 @@ echo -e "Blob GC enabled:\t\t\t$enable_blob_garbage_collection"
 echo -e "Blob GC age cutoff:\t\t\t$blob_garbage_collection_age_cutoff"
 echo -e "Blob GC force threshold:\t\t$blob_garbage_collection_force_threshold"
 echo -e "Blob compaction readahead size:\t\t$blob_compaction_readahead_size"
+echo -e "Blob file starting level:\t\t$blob_file_starting_level"
 echo -e "Target SST file size:\t\t\t$target_file_size_base"
 echo -e "Maximum size of base level:\t\t$max_bytes_for_level_base"
 echo "================================================================="
@@ -167,6 +170,7 @@ PARAMS="\
   --min_blob_size=$min_blob_size \
   --blob_file_size=$blob_file_size \
   --blob_compression_type=$blob_compression_type \
+  --blob_file_starting_level=$blob_file_starting_level \
   --write_buffer_size=$write_buffer_size \
   --target_file_size_base=$target_file_size_base \
   --max_bytes_for_level_base=$max_bytes_for_level_base"