Add an option to trigger flush when the number of range deletions rea…

…ch a threshold (facebook#11358)

Summary:
Add a mutable column family option `memtable_max_range_deletions`. When non-zero, RocksDB will try to flush the current memtable after it has at least `memtable_max_range_deletions` range deletions. Java API is added and crash test is updated accordingly to randomly enable this option.

Pull Request resolved: facebook#11358

Test Plan:
* New unit test: `DBRangeDelTest.MemtableMaxRangeDeletions`
* Ran crash test `python3 ./tools/db_crashtest.py whitebox --simple --memtable_max_range_deletions=20` and saw logs showing flushed memtables usually with 20 range deletions.

Reviewed By: ajkr

Differential Revision: D46582680

Pulled By: cbi42

fbshipit-source-id: f23d6fa8d8264ecf0a18d55c113ba03f5e2504da

db/db_range_del_test.cc

@@ -3475,6 +3475,42 @@ TEST_F(DBRangeDelTest, NonBottommostCompactionDropRangetombstone) {
   db_->ReleaseSnapshot(snapshot);
 }
 
+TEST_F(DBRangeDelTest, MemtableMaxRangeDeletions) {
+  // Tests option `memtable_max_range_deletions`.
+  Options options = CurrentOptions();
+  options.level_compaction_dynamic_file_size = false;
+  options.memtable_max_range_deletions = 50;
+  options.level0_file_num_compaction_trigger = 5;
+  DestroyAndReopen(options);
+
+  for (int i = 0; i < 50; ++i) {
+    // Intentionally delete overlapping ranges to see if the option
+    // checks number of range tombstone fragments instead.
+    ASSERT_OK(Put(Key(i), "val1"));
+    ASSERT_OK(Put(Key(i + 1), "val2"));
+    ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                               Key(i), Key(i + 2)));
+    ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable());
+    ASSERT_EQ(0, NumTableFilesAtLevel(0));
+  }
+  // One more write to trigger flush.
+  ASSERT_OK(Put(Key(50), "val"));
+  ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable());
+  ASSERT_EQ(1, NumTableFilesAtLevel(0));
+
+  // This should take effect for the next new memtable.
+  ASSERT_OK(db_->SetOptions({{"memtable_max_range_deletions", "1"}}));
+  ASSERT_OK(Flush());
+  ASSERT_EQ(2, NumTableFilesAtLevel(0));
+  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
+                             Key(50), Key(100)));
+  ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable());
+  ASSERT_EQ(2, NumTableFilesAtLevel(0));
+  // One more write to trigger flush.
+  ASSERT_OK(Put(Key(50), "new val"));
+  ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable());
+  ASSERT_EQ(3, NumTableFilesAtLevel(0));
+}
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {

db/flush_job.cc

@@ -861,6 +861,7 @@ Status FlushJob::WriteLevel0Table() {
     uint64_t total_num_entries = 0, total_num_deletes = 0;
     uint64_t total_data_size = 0;
     size_t total_memory_usage = 0;
+    uint64_t total_num_range_deletes = 0;
     // Used for testing:
     uint64_t mems_size = mems_.size();
     (void)mems_size;  // avoids unused variable error when
@@ -883,15 +884,20 @@ Status FlushJob::WriteLevel0Table() {
       total_num_deletes += m->num_deletes();
       total_data_size += m->get_data_size();
       total_memory_usage += m->ApproximateMemoryUsage();
+      total_num_range_deletes += m->num_range_deletes();
     }
 
+    // TODO(cbi): when memtable is flushed due to number of range deletions
+    //  hitting limit memtable_max_range_deletions, flush_reason_ is still
+    //  "Write Buffer Full", should make update flush_reason_ accordingly.
     event_logger_->Log() << "job" << job_context_->job_id << "event"
                          << "flush_started"
                          << "num_memtables" << mems_.size() << "num_entries"
                          << total_num_entries << "num_deletes"
                          << total_num_deletes << "total_data_size"
                          << total_data_size << "memory_usage"
-                         << total_memory_usage << "flush_reason"
+                         << total_memory_usage << "num_range_deletes"
+                         << total_num_range_deletes << "flush_reason"
                          << GetFlushReasonString(flush_reason_);
 
     {

db/memtable.cc

@@ -95,6 +95,7 @@ MemTable::MemTable(const InternalKeyComparator& cmp,
       data_size_(0),
       num_entries_(0),
       num_deletes_(0),
+      num_range_deletes_(0),
       write_buffer_size_(mutable_cf_options.write_buffer_size),
       flush_in_progress_(false),
       flush_completed_(false),
@@ -114,7 +115,9 @@ MemTable::MemTable(const InternalKeyComparator& cmp,
           ioptions.memtable_insert_with_hint_prefix_extractor.get()),
       oldest_key_time_(std::numeric_limits<uint64_t>::max()),
       atomic_flush_seqno_(kMaxSequenceNumber),
-      approximate_memory_usage_(0) {
+      approximate_memory_usage_(0),
+      memtable_max_range_deletions_(
+          mutable_cf_options.memtable_max_range_deletions) {
   UpdateFlushState();
   // something went wrong if we need to flush before inserting anything
   assert(!ShouldScheduleFlush());
@@ -174,6 +177,14 @@ size_t MemTable::ApproximateMemoryUsage() {
 }
 
 bool MemTable::ShouldFlushNow() {
+  // This is set if memtable_max_range_deletions is > 0,
+  // and that many range deletions are done
+  if (memtable_max_range_deletions_ > 0 &&
+      num_range_deletes_.load(std::memory_order_relaxed) >=
+          static_cast<uint64_t>(memtable_max_range_deletions_)) {
+    return true;
+  }
+
   size_t write_buffer_size = write_buffer_size_.load(std::memory_order_relaxed);
   // In a lot of times, we cannot allocate arena blocks that exactly matches the
   // buffer size. Thus we have to decide if we should over-allocate or
@@ -756,6 +767,9 @@ Status MemTable::Add(SequenceNumber s, ValueType type,
         type == kTypeDeletionWithTimestamp) {
       num_deletes_.store(num_deletes_.load(std::memory_order_relaxed) + 1,
                          std::memory_order_relaxed);
+    } else if (type == kTypeRangeDeletion) {
+      uint64_t val = num_range_deletes_.load(std::memory_order_relaxed) + 1;
+      num_range_deletes_.store(val, std::memory_order_relaxed);
     }
 
     if (bloom_filter_ && prefix_extractor_ &&
@@ -822,6 +836,7 @@ Status MemTable::Add(SequenceNumber s, ValueType type,
     auto new_cache = std::make_shared<FragmentedRangeTombstoneListCache>();
     size_t size = cached_range_tombstone_.Size();
     if (allow_concurrent) {
+      post_process_info->num_range_deletes++;
       range_del_mutex_.lock();
     }
     for (size_t i = 0; i < size; ++i) {
@@ -840,6 +855,7 @@ Status MemTable::Add(SequenceNumber s, ValueType type,
               new_local_cache_ref, new_cache.get()),
           std::memory_order_relaxed);
     }
+
     if (allow_concurrent) {
       range_del_mutex_.unlock();
     }
@@ -1268,6 +1284,7 @@ bool MemTable::Get(const LookupKey& key, std::string* value,
     // Avoiding recording stats for speed.
     return false;
   }
+
   PERF_TIMER_GUARD(get_from_memtable_time);
 
   std::unique_ptr<FragmentedRangeTombstoneIterator> range_del_iter(

db/memtable.h

@@ -68,6 +68,7 @@ struct MemTablePostProcessInfo {
   uint64_t data_size = 0;
   uint64_t num_entries = 0;
   uint64_t num_deletes = 0;
+  uint64_t num_range_deletes = 0;
 };
 
 using MultiGetRange = MultiGetContext::Range;
@@ -332,6 +333,10 @@ class MemTable {
       num_deletes_.fetch_add(update_counters.num_deletes,
                              std::memory_order_relaxed);
     }
+    if (update_counters.num_range_deletes > 0) {
+      num_range_deletes_.fetch_add(update_counters.num_range_deletes,
+                                   std::memory_order_relaxed);
+    }
     UpdateFlushState();
   }
 
@@ -349,6 +354,13 @@ class MemTable {
     return num_deletes_.load(std::memory_order_relaxed);
   }
 
+  // Get total number of range deletions in the mem table.
+  // REQUIRES: external synchronization to prevent simultaneous
+  // operations on the same MemTable (unless this Memtable is immutable).
+  uint64_t num_range_deletes() const {
+    return num_range_deletes_.load(std::memory_order_relaxed);
+  }
+
   uint64_t get_data_size() const {
     return data_size_.load(std::memory_order_relaxed);
   }
@@ -565,6 +577,7 @@ class MemTable {
   std::atomic<uint64_t> data_size_;
   std::atomic<uint64_t> num_entries_;
   std::atomic<uint64_t> num_deletes_;
+  std::atomic<uint64_t> num_range_deletes_;
 
   // Dynamically changeable memtable option
   std::atomic<size_t> write_buffer_size_;
@@ -626,6 +639,10 @@ class MemTable {
   // Gets refreshed inside `ApproximateMemoryUsage()` or `ShouldFlushNow`
   std::atomic<uint64_t> approximate_memory_usage_;
 
+  // max range deletions in a memtable,  before automatic flushing, 0 for
+  // unlimited.
+  uint32_t memtable_max_range_deletions_ = 0;
+
   // Flush job info of the current memtable.
   std::unique_ptr<FlushJobInfo> flush_job_info_;
 

db_stress_tool/db_stress_common.h

@@ -327,6 +327,8 @@ DECLARE_bool(allow_data_in_errors);
 
 DECLARE_bool(enable_thread_tracking);
 
+DECLARE_uint32(memtable_max_range_deletions);
+
 // Tiered storage
 DECLARE_bool(enable_tiered_storage);  // set last_level_temperature
 DECLARE_int64(preclude_last_level_data_seconds);

db_stress_tool/db_stress_gflags.cc

@@ -1102,4 +1102,8 @@ DEFINE_uint64(stats_dump_period_sec,
 DEFINE_bool(use_io_uring, false, "Enable the use of IO uring on Posix");
 extern "C" bool RocksDbIOUringEnable() { return FLAGS_use_io_uring; }
 
+DEFINE_uint32(memtable_max_range_deletions, 0,
+              "If nonzero, RocksDB will try to flush the current memtable"
+              "after the number of range deletions is >= this limit");
+
 #endif  // GFLAGS

db_stress_tool/db_stress_test_base.cc

@@ -3278,6 +3278,8 @@ void InitializeOptionsFromFlags(
   options.allow_data_in_errors = FLAGS_allow_data_in_errors;
 
   options.enable_thread_tracking = FLAGS_enable_thread_tracking;
+
+  options.memtable_max_range_deletions = FLAGS_memtable_max_range_deletions;
 }
 
 void InitializeOptionsGeneral(

include/rocksdb/options.h

@@ -331,6 +331,17 @@ struct ColumnFamilyOptions : public AdvancedColumnFamilyOptions {
   // Default: nullptr
   std::shared_ptr<SstPartitionerFactory> sst_partitioner_factory = nullptr;
 
+  // RocksDB will try to flush the current memtable after the number of range
+  // deletions is >= this limit. For workloads with many range
+  // deletions, limiting the number of range deletions in memtable can help
+  // prevent performance degradation and/or OOM caused by too many range
+  // tombstones in a single memtable.
+  //
+  // Default: 0 (disabled)
+  //
+  // Dynamically changeable through SetOptions() API
+  uint32_t memtable_max_range_deletions = 0;
+
   // Create ColumnFamilyOptions with default values for all fields
   ColumnFamilyOptions();
   // Create ColumnFamilyOptions from Options

java/rocksjni/options.cc

@@ -3904,6 +3904,29 @@ jbyte Java_org_rocksdb_Options_prepopulateBlobCache(JNIEnv*, jobject,
       opts->prepopulate_blob_cache);
 }
 
+/*
+ * Class:     org_rocksdb_Options
+ * Method:    setMemtableMaxRangeDeletions
+ * Signature: (JI)V
+ */
+void Java_org_rocksdb_Options_setMemtableMaxRangeDeletions(
+    JNIEnv*, jobject, jlong jhandle, jint jmemtable_max_range_deletions) {
+  auto* opts = reinterpret_cast<ROCKSDB_NAMESPACE::Options*>(jhandle);
+  opts->memtable_max_range_deletions =
+      static_cast<int32_t>(jmemtable_max_range_deletions);
+}
+
+/*
+ * Class:     org_rocksdb_Options
+ * Method:    memtableMaxRangeDeletions
+ * Signature: (J)I
+ */
+jint Java_org_rocksdb_Options_memtableMaxRangeDeletions(JNIEnv*, jobject,
+                                                        jlong jhandle) {
+  auto* opts = reinterpret_cast<ROCKSDB_NAMESPACE::Options*>(jhandle);
+  return static_cast<jint>(opts->memtable_max_range_deletions);
+}
+
 //////////////////////////////////////////////////////////////////////////////
 // ROCKSDB_NAMESPACE::ColumnFamilyOptions
 
@@ -5770,6 +5793,30 @@ jbyte Java_org_rocksdb_ColumnFamilyOptions_prepopulateBlobCache(JNIEnv*,
       opts->prepopulate_blob_cache);
 }
 
+/*
+ * Class:     org_rocksdb_ColumnFamilyOptions
+ * Method:    setMemtableMaxRangeDeletions
+ * Signature: (JI)V
+ */
+void Java_org_rocksdb_ColumnFamilyOptions_setMemtableMaxRangeDeletions(
+    JNIEnv*, jobject, jlong jhandle, jint jmemtable_max_range_deletions) {
+  auto* opts =
+      reinterpret_cast<ROCKSDB_NAMESPACE::ColumnFamilyOptions*>(jhandle);
+  opts->memtable_max_range_deletions = jmemtable_max_range_deletions;
+}
+
+/*
+ * Class:     org_rocksdb_ColumnFamilyOptions
+ * Method:    memtableMaxRangeDeletions
+ * Signature: (J)I
+ */
+jint Java_org_rocksdb_ColumnFamilyOptions_memtableMaxRangeDeletions(
+    JNIEnv*, jobject, jlong jhandle) {
+  auto* opts =
+      reinterpret_cast<ROCKSDB_NAMESPACE::ColumnFamilyOptions*>(jhandle);
+  return static_cast<jint>(opts->memtable_max_range_deletions);
+}
+
 /////////////////////////////////////////////////////////////////////
 // ROCKSDB_NAMESPACE::DBOptions
 

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

@@ -959,6 +959,17 @@ public SstPartitionerFactory sstPartitionerFactory() {
     return sstPartitionerFactory_;
   }
 
+  @Override
+  public ColumnFamilyOptions setMemtableMaxRangeDeletions(final int count) {
+    setMemtableMaxRangeDeletions(nativeHandle_, count);
+    return this;
+  }
+
+  @Override
+  public int memtableMaxRangeDeletions() {
+    return memtableMaxRangeDeletions(nativeHandle_);
+  }
+
   //
   // BEGIN options for blobs (integrated BlobDB)
   //
@@ -1498,6 +1509,8 @@ private native void setForceConsistencyChecks(final long handle,
   private native void setSstPartitionerFactory(long nativeHandle_, long newFactoryHandle);
   private static native void setCompactionThreadLimiter(
       final long nativeHandle_, final long compactionThreadLimiterHandle);
+  private native void setMemtableMaxRangeDeletions(final long handle, final int count);
+  private native int memtableMaxRangeDeletions(final long handle);
 
   private native void setEnableBlobFiles(final long nativeHandle_, final boolean enableBlobFiles);
   private native boolean enableBlobFiles(final long nativeHandle_);

java/src/main/java/org/rocksdb/ColumnFamilyOptionsInterface.java

@@ -506,6 +506,23 @@ T setCompressionOptions(
   @Experimental("Caution: this option is experimental")
   SstPartitionerFactory sstPartitionerFactory();
 
+  /**
+   * Sets the maximum range delete calls, after which memtable is flushed.
+   * This applies to the mutable memtable.
+   *
+   * @param count a positive integer, 0 (default) to disable the feature.
+   * @return the reference of the current options.
+   */
+  T setMemtableMaxRangeDeletions(final int count);
+
+  /**
+   * Gets the current setting of maximum range deletes allowed
+   * 0(default) indicates that feature is disabled.
+   *
+   * @return current value of memtable_max_range_deletions
+   */
+  int memtableMaxRangeDeletions();
+
   /**
    * Compaction concurrent thread limiter for the column family.
    * If non-nullptr, use given concurrent thread limiter to control

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

@@ -1984,6 +1984,17 @@ public SstPartitionerFactory sstPartitionerFactory() {
     return sstPartitionerFactory_;
   }
 
+  @Override
+  public Options setMemtableMaxRangeDeletions(final int count) {
+    setMemtableMaxRangeDeletions(nativeHandle_, count);
+    return this;
+  }
+
+  @Override
+  public int memtableMaxRangeDeletions() {
+    return memtableMaxRangeDeletions(nativeHandle_);
+  }
+
   @Override
   public Options setCompactionThreadLimiter(final ConcurrentTaskLimiter compactionThreadLimiter) {
     setCompactionThreadLimiter(nativeHandle_, compactionThreadLimiter.nativeHandle_);
@@ -2502,6 +2513,8 @@ private native void setAtomicFlush(final long handle,
       final boolean atomicFlush);
   private native boolean atomicFlush(final long handle);
   private native void setSstPartitionerFactory(long nativeHandle_, long newFactoryHandle);
+  private native void setMemtableMaxRangeDeletions(final long handle, final int count);
+  private native int memtableMaxRangeDeletions(final long handle);
   private static native void setCompactionThreadLimiter(
       final long nativeHandle_, final long newLimiterHandle);
   private static native void setAvoidUnnecessaryBlockingIO(

java/src/test/java/org/rocksdb/ColumnFamilyOptionsTest.java

@@ -709,4 +709,14 @@ public void cfPaths() throws IOException {
       assertThat(options.cfPaths()).isEqualTo(paths);
     }
   }
+
+  @Test
+  public void memtableMaxRangeDeletions() {
+    try (final ColumnFamilyOptions options = new ColumnFamilyOptions()) {
+      assertThat(options.memtableMaxRangeDeletions()).isEqualTo(0);
+      final int val = 32;
+      assertThat(options.setMemtableMaxRangeDeletions(val)).isEqualTo(options);
+      assertThat(options.memtableMaxRangeDeletions()).isEqualTo(val);
+    }
+  }
 }