@nxtedition/rocksdb 5.2.21 → 5.2.26

package/deps/rocksdb/rocksdb/utilities/transactions/write_prepared_txn.cc

@@ -154,11 +154,17 @@ Status WritePreparedTxn::CommitInternal() {
   assert(s.ok());
 
   const bool for_recovery = use_only_the_last_commit_time_batch_for_recovery_;
-  if (!empty && for_recovery) {
+  if (!empty) {
     // When not writing to memtable, we can still cache the latest write batch.
     // The cached batch will be written to memtable in WriteRecoverableState
     // during FlushMemTable
-    WriteBatchInternal::SetAsLastestPersistentState(working_batch);
+    if (for_recovery) {
+      WriteBatchInternal::SetAsLatestPersistentState(working_batch);
+    } else {
+      return Status::InvalidArgument(
+          "Commit-time-batch can only be used if "
+          "use_only_the_last_commit_time_batch_for_recovery is true");
+    }
   }
 
   auto prepare_seq = GetId();
@@ -195,6 +201,21 @@ Status WritePreparedTxn::CommitInternal() {
   // redundantly reference the log that contains the prepared data.
   const uint64_t zero_log_number = 0ull;
   size_t batch_cnt = UNLIKELY(commit_batch_cnt) ? commit_batch_cnt : 1;
+  // If `two_write_queues && includes_data`, then `do_one_write` is false. The
+  // following `WriteImpl` will insert the data of the commit-time-batch into
+  // the database before updating the commit cache. Therefore, the data of the
+  // commmit-time-batch is considered uncommitted. Furthermore, since data of
+  // the commit-time-batch are not locked, it is possible for two uncommitted
+  // versions of the same key to co-exist for a (short) period of time until
+  // the commit cache is updated by the second write. If the two uncommitted
+  // keys are compacted to the bottommost level in the meantime, it is possible
+  // that compaction iterator will zero out the sequence numbers of both, thus
+  // violating the invariant that an SST does not have two identical internal
+  // keys. To prevent this situation, we should allow the usage of
+  // commit-time-batch only if the user sets
+  // TransactionOptions::use_only_the_last_commit_time_batch_for_recovery to
+  // true. See the comments about GetCommitTimeWriteBatch() in
+  // include/rocksdb/utilities/transaction.h.
   s = db_impl_->WriteImpl(write_options_, working_batch, nullptr, nullptr,
                           zero_log_number, disable_memtable, &seq_used,
                           batch_cnt, pre_release_callback);
@@ -236,14 +257,6 @@ Status WritePreparedTxn::CommitInternal() {
                           NO_REF_LOG, DISABLE_MEMTABLE, &seq_used, ONE_BATCH,
                           &update_commit_map_with_aux_batch);
   assert(!s.ok() || seq_used != kMaxSequenceNumber);
-  if (UNLIKELY(!db_impl_->immutable_db_options().two_write_queues)) {
-    if (s.ok()) {
-      // Note: RemovePrepared should be called after WriteImpl that publishsed
-      // the seq. Otherwise SmallestUnCommittedSeq optimization breaks.
-      wpt_db_->RemovePrepared(prepare_seq, prepare_batch_cnt_);
-    }
-    wpt_db_->RemovePrepared(commit_batch_seq, commit_batch_cnt);
-  }  // else RemovePrepared is called from within PreReleaseCallback
   return s;
 }
 
@@ -461,9 +474,10 @@ Status WritePreparedTxn::ValidateSnapshot(ColumnFamilyHandle* column_family,
 
   WritePreparedTxnReadCallback snap_checker(wpt_db_, snap_seq, min_uncommitted,
                                             kBackedByDBSnapshot);
-  return TransactionUtil::CheckKeyForConflicts(db_impl_, cfh, key.ToString(),
-                                               snap_seq, false /* cache_only */,
-                                               &snap_checker, min_uncommitted);
+  // TODO(yanqin): support user-defined timestamp
+  return TransactionUtil::CheckKeyForConflicts(
+      db_impl_, cfh, key.ToString(), snap_seq, /*ts=*/nullptr,
+      false /* cache_only */, &snap_checker, min_uncommitted);
 }
 
 void WritePreparedTxn::SetSnapshot() {

package/deps/rocksdb/rocksdb/utilities/transactions/write_prepared_txn_db.cc

@@ -15,6 +15,7 @@
 
 #include "db/arena_wrapped_db_iter.h"
 #include "db/db_impl/db_impl.h"
+#include "logging/logging.h"
 #include "rocksdb/db.h"
 #include "rocksdb/options.h"
 #include "rocksdb/utilities/transaction_db.h"
@@ -245,6 +246,7 @@ Status WritePreparedTxnDB::Get(const ReadOptions& options,
                                                   backed_by_snapshot))) {
     return res;
   } else {
+    res.PermitUncheckedError();
     WPRecordTick(TXN_GET_TRY_AGAIN);
     return Status::TryAgain();
   }
@@ -508,24 +510,22 @@ void WritePreparedTxnDB::AddCommitted(uint64_t prepare_seq, uint64_t commit_seq,
                         prev_max, max_evicted_seq);
       AdvanceMaxEvictedSeq(prev_max, max_evicted_seq);
     }
-    // After each eviction from commit cache, check if the commit entry should
-    // be kept around because it overlaps with a live snapshot.
-    CheckAgainstSnapshots(evicted);
     if (UNLIKELY(!delayed_prepared_empty_.load(std::memory_order_acquire))) {
       WriteLock wl(&prepared_mutex_);
-      for (auto dp : delayed_prepared_) {
-        if (dp == evicted.prep_seq) {
-          // This is a rare case that txn is committed but prepared_txns_ is not
-          // cleaned up yet. Refer to delayed_prepared_commits_ definition for
-          // why it should be kept updated.
-          delayed_prepared_commits_[evicted.prep_seq] = evicted.commit_seq;
-          ROCKS_LOG_DEBUG(info_log_,
-                          "delayed_prepared_commits_[%" PRIu64 "]=%" PRIu64,
-                          evicted.prep_seq, evicted.commit_seq);
-          break;
-        }
+      auto dp_iter = delayed_prepared_.find(evicted.prep_seq);
+      if (dp_iter != delayed_prepared_.end()) {
+        // This is a rare case that txn is committed but prepared_txns_ is not
+        // cleaned up yet. Refer to delayed_prepared_commits_ definition for
+        // why it should be kept updated.
+        delayed_prepared_commits_[evicted.prep_seq] = evicted.commit_seq;
+        ROCKS_LOG_DEBUG(info_log_,
+                        "delayed_prepared_commits_[%" PRIu64 "]=%" PRIu64,
+                        evicted.prep_seq, evicted.commit_seq);
       }
     }
+    // After each eviction from commit cache, check if the commit entry should
+    // be kept around because it overlaps with a live snapshot.
+    CheckAgainstSnapshots(evicted);
   }
   bool succ =
       ExchangeCommitEntry(indexed_seq, evicted_64b, {prepare_seq, commit_seq});

package/deps/rocksdb/rocksdb/utilities/transactions/write_prepared_txn_db.h

@@ -18,6 +18,7 @@
 #include "db/pre_release_callback.h"
 #include "db/read_callback.h"
 #include "db/snapshot_checker.h"
+#include "logging/logging.h"
 #include "rocksdb/db.h"
 #include "rocksdb/options.h"
 #include "rocksdb/utilities/transaction_db.h"
@@ -1080,6 +1081,8 @@ SnapshotBackup WritePreparedTxnDB::AssignMinMaxSeqs(const Snapshot* snapshot,
     *min =
         static_cast_with_check<const SnapshotImpl>(snapshot)->min_uncommitted_;
     *max = static_cast_with_check<const SnapshotImpl>(snapshot)->number_;
+    // A duplicate of the check in EnhanceSnapshot().
+    assert(*min <= *max + 1);
     return kBackedByDBSnapshot;
   } else {
     *min = SmallestUnCommittedSeq();

package/deps/rocksdb/rocksdb/utilities/transactions/write_unprepared_transaction_test.cc

@@ -136,7 +136,7 @@ TEST_P(WriteUnpreparedTransactionTest, ReadYourOwnWrite) {
   }
 }
 
-#ifndef ROCKSDB_VALGRIND_RUN
+#if !defined(ROCKSDB_VALGRIND_RUN) || defined(ROCKSDB_FULL_VALGRIND_RUN)
 TEST_P(WriteUnpreparedStressTest, ReadYourOwnWriteStress) {
   // This is a stress test where different threads are writing random keys, and
   // then before committing or aborting the transaction, it validates to see
@@ -314,7 +314,7 @@ TEST_P(WriteUnpreparedStressTest, ReadYourOwnWriteStress) {
     t.join();
   }
 }
-#endif  // ROCKSDB_VALGRIND_RUN
+#endif  // !defined(ROCKSDB_VALGRIND_RUN) || defined(ROCKSDB_FULL_VALGRIND_RUN)
 
 // This tests how write unprepared behaves during recovery when the DB crashes
 // after a transaction has either been unprepared or prepared, and tests if

package/deps/rocksdb/rocksdb/utilities/transactions/write_unprepared_txn.cc

@@ -6,6 +6,7 @@
 #ifndef ROCKSDB_LITE
 
 #include "utilities/transactions/write_unprepared_txn.h"
+
 #include "db/db_impl/db_impl.h"
 #include "util/cast_util.h"
 #include "utilities/transactions/write_unprepared_txn_db.h"
@@ -549,11 +550,17 @@ Status WriteUnpreparedTxn::CommitInternal() {
   assert(s.ok());
 
   const bool for_recovery = use_only_the_last_commit_time_batch_for_recovery_;
-  if (!empty && for_recovery) {
+  if (!empty) {
     // When not writing to memtable, we can still cache the latest write batch.
     // The cached batch will be written to memtable in WriteRecoverableState
     // during FlushMemTable
-    WriteBatchInternal::SetAsLastestPersistentState(working_batch);
+    if (for_recovery) {
+      WriteBatchInternal::SetAsLatestPersistentState(working_batch);
+    } else {
+      return Status::InvalidArgument(
+          "Commit-time-batch can only be used if "
+          "use_only_the_last_commit_time_batch_for_recovery is true");
+    }
   }
 
   const bool includes_data = !empty && !for_recovery;
@@ -968,6 +975,7 @@ Status WriteUnpreparedTxn::Get(const ReadOptions& options,
              wupt_db_->ValidateSnapshot(snap_seq, backed_by_snapshot))) {
     return res;
   } else {
+    res.PermitUncheckedError();
     wupt_db_->WPRecordTick(TXN_GET_TRY_AGAIN);
     return Status::TryAgain();
   }
@@ -1024,9 +1032,10 @@ Status WriteUnpreparedTxn::ValidateSnapshot(ColumnFamilyHandle* column_family,
 
   WriteUnpreparedTxnReadCallback snap_checker(
       wupt_db_, snap_seq, min_uncommitted, unprep_seqs_, kBackedByDBSnapshot);
-  return TransactionUtil::CheckKeyForConflicts(db_impl_, cfh, key.ToString(),
-                                               snap_seq, false /* cache_only */,
-                                               &snap_checker, min_uncommitted);
+  // TODO(yanqin): Support user-defined timestamp.
+  return TransactionUtil::CheckKeyForConflicts(
+      db_impl_, cfh, key.ToString(), snap_seq, /*ts=*/nullptr,
+      false /* cache_only */, &snap_checker, min_uncommitted);
 }
 
 const std::map<SequenceNumber, size_t>&

package/deps/rocksdb/rocksdb/utilities/ttl/db_ttl_impl.cc

@@ -8,31 +8,303 @@
 
 #include "db/write_batch_internal.h"
 #include "file/filename.h"
+#include "logging/logging.h"
 #include "rocksdb/convenience.h"
 #include "rocksdb/env.h"
 #include "rocksdb/iterator.h"
+#include "rocksdb/system_clock.h"
 #include "rocksdb/utilities/db_ttl.h"
+#include "rocksdb/utilities/object_registry.h"
+#include "rocksdb/utilities/options_type.h"
 #include "util/coding.h"
 
 namespace ROCKSDB_NAMESPACE {
+static std::unordered_map<std::string, OptionTypeInfo> ttl_merge_op_type_info =
+    {{"user_operator",
+      OptionTypeInfo::AsCustomSharedPtr<MergeOperator>(
+          0, OptionVerificationType::kByName, OptionTypeFlags::kNone)}};
+
+TtlMergeOperator::TtlMergeOperator(
+    const std::shared_ptr<MergeOperator>& merge_op, SystemClock* clock)
+    : user_merge_op_(merge_op), clock_(clock) {
+  RegisterOptions("TtlMergeOptions", &user_merge_op_, &ttl_merge_op_type_info);
+}
+
+bool TtlMergeOperator::FullMergeV2(const MergeOperationInput& merge_in,
+                                   MergeOperationOutput* merge_out) const {
+  const uint32_t ts_len = DBWithTTLImpl::kTSLength;
+  if (merge_in.existing_value && merge_in.existing_value->size() < ts_len) {
+    ROCKS_LOG_ERROR(merge_in.logger,
+                    "Error: Could not remove timestamp from existing value.");
+    return false;
+  }
+
+  // Extract time-stamp from each operand to be passed to user_merge_op_
+  std::vector<Slice> operands_without_ts;
+  for (const auto& operand : merge_in.operand_list) {
+    if (operand.size() < ts_len) {
+      ROCKS_LOG_ERROR(merge_in.logger,
+                      "Error: Could not remove timestamp from operand value.");
+      return false;
+    }
+    operands_without_ts.push_back(operand);
+    operands_without_ts.back().remove_suffix(ts_len);
+  }
+
+  // Apply the user merge operator (store result in *new_value)
+  bool good = true;
+  MergeOperationOutput user_merge_out(merge_out->new_value,
+                                      merge_out->existing_operand);
+  if (merge_in.existing_value) {
+    Slice existing_value_without_ts(merge_in.existing_value->data(),
+                                    merge_in.existing_value->size() - ts_len);
+    good = user_merge_op_->FullMergeV2(
+        MergeOperationInput(merge_in.key, &existing_value_without_ts,
+                            operands_without_ts, merge_in.logger),
+        &user_merge_out);
+  } else {
+    good = user_merge_op_->FullMergeV2(
+        MergeOperationInput(merge_in.key, nullptr, operands_without_ts,
+                            merge_in.logger),
+        &user_merge_out);
+  }
+
+  // Return false if the user merge operator returned false
+  if (!good) {
+    return false;
+  }
+
+  if (merge_out->existing_operand.data()) {
+    merge_out->new_value.assign(merge_out->existing_operand.data(),
+                                merge_out->existing_operand.size());
+    merge_out->existing_operand = Slice(nullptr, 0);
+  }
+
+  // Augment the *new_value with the ttl time-stamp
+  int64_t curtime;
+  if (!clock_->GetCurrentTime(&curtime).ok()) {
+    ROCKS_LOG_ERROR(
+        merge_in.logger,
+        "Error: Could not get current time to be attached internally "
+        "to the new value.");
+    return false;
+  } else {
+    char ts_string[ts_len];
+    EncodeFixed32(ts_string, (int32_t)curtime);
+    merge_out->new_value.append(ts_string, ts_len);
+    return true;
+  }
+}
+
+bool TtlMergeOperator::PartialMergeMulti(const Slice& key,
+                                         const std::deque<Slice>& operand_list,
+                                         std::string* new_value,
+                                         Logger* logger) const {
+  const uint32_t ts_len = DBWithTTLImpl::kTSLength;
+  std::deque<Slice> operands_without_ts;
+
+  for (const auto& operand : operand_list) {
+    if (operand.size() < ts_len) {
+      ROCKS_LOG_ERROR(logger, "Error: Could not remove timestamp from value.");
+      return false;
+    }
+
+    operands_without_ts.push_back(
+        Slice(operand.data(), operand.size() - ts_len));
+  }
+
+  // Apply the user partial-merge operator (store result in *new_value)
+  assert(new_value);
+  if (!user_merge_op_->PartialMergeMulti(key, operands_without_ts, new_value,
+                                         logger)) {
+    return false;
+  }
+
+  // Augment the *new_value with the ttl time-stamp
+  int64_t curtime;
+  if (!clock_->GetCurrentTime(&curtime).ok()) {
+    ROCKS_LOG_ERROR(
+        logger,
+        "Error: Could not get current time to be attached internally "
+        "to the new value.");
+    return false;
+  } else {
+    char ts_string[ts_len];
+    EncodeFixed32(ts_string, (int32_t)curtime);
+    new_value->append(ts_string, ts_len);
+    return true;
+  }
+}
+
+Status TtlMergeOperator::PrepareOptions(const ConfigOptions& config_options) {
+  if (clock_ == nullptr) {
+    clock_ = config_options.env->GetSystemClock().get();
+  }
+  return MergeOperator::PrepareOptions(config_options);
+}
+
+Status TtlMergeOperator::ValidateOptions(
+    const DBOptions& db_opts, const ColumnFamilyOptions& cf_opts) const {
+  if (user_merge_op_ == nullptr) {
+    return Status::InvalidArgument(
+        "UserMergeOperator required by TtlMergeOperator");
+  } else if (clock_ == nullptr) {
+    return Status::InvalidArgument("SystemClock required by TtlMergeOperator");
+  } else {
+    return MergeOperator::ValidateOptions(db_opts, cf_opts);
+  }
+}
 
 void DBWithTTLImpl::SanitizeOptions(int32_t ttl, ColumnFamilyOptions* options,
-                                    Env* env) {
+                                    SystemClock* clock) {
   if (options->compaction_filter) {
     options->compaction_filter =
-        new TtlCompactionFilter(ttl, env, options->compaction_filter);
+        new TtlCompactionFilter(ttl, clock, options->compaction_filter);
   } else {
     options->compaction_filter_factory =
         std::shared_ptr<CompactionFilterFactory>(new TtlCompactionFilterFactory(
-            ttl, env, options->compaction_filter_factory));
+            ttl, clock, options->compaction_filter_factory));
   }
 
   if (options->merge_operator) {
     options->merge_operator.reset(
-        new TtlMergeOperator(options->merge_operator, env));
+        new TtlMergeOperator(options->merge_operator, clock));
+  }
+}
+
+static std::unordered_map<std::string, OptionTypeInfo> ttl_type_info = {
+    {"ttl", {0, OptionType::kInt32T}},
+};
+
+static std::unordered_map<std::string, OptionTypeInfo> ttl_cff_type_info = {
+    {"user_filter_factory",
+     OptionTypeInfo::AsCustomSharedPtr<CompactionFilterFactory>(
+         0, OptionVerificationType::kByNameAllowFromNull,
+         OptionTypeFlags::kNone)}};
+static std::unordered_map<std::string, OptionTypeInfo> user_cf_type_info = {
+    {"user_filter",
+     OptionTypeInfo::AsCustomRawPtr<const CompactionFilter>(
+         0, OptionVerificationType::kByName, OptionTypeFlags::kAllowNull)}};
+
+TtlCompactionFilter::TtlCompactionFilter(
+    int32_t ttl, SystemClock* clock, const CompactionFilter* _user_comp_filter,
+    std::unique_ptr<const CompactionFilter> _user_comp_filter_from_factory)
+    : LayeredCompactionFilterBase(_user_comp_filter,
+                                  std::move(_user_comp_filter_from_factory)),
+      ttl_(ttl),
+      clock_(clock) {
+  RegisterOptions("TTL", &ttl_, &ttl_type_info);
+  RegisterOptions("UserFilter", &user_comp_filter_, &user_cf_type_info);
+}
+
+bool TtlCompactionFilter::Filter(int level, const Slice& key,
+                                 const Slice& old_val, std::string* new_val,
+                                 bool* value_changed) const {
+  if (DBWithTTLImpl::IsStale(old_val, ttl_, clock_)) {
+    return true;
+  }
+  if (user_comp_filter() == nullptr) {
+    return false;
+  }
+  assert(old_val.size() >= DBWithTTLImpl::kTSLength);
+  Slice old_val_without_ts(old_val.data(),
+                           old_val.size() - DBWithTTLImpl::kTSLength);
+  if (user_comp_filter()->Filter(level, key, old_val_without_ts, new_val,
+                                 value_changed)) {
+    return true;
+  }
+  if (*value_changed) {
+    new_val->append(old_val.data() + old_val.size() - DBWithTTLImpl::kTSLength,
+                    DBWithTTLImpl::kTSLength);
+  }
+  return false;
+}
+
+Status TtlCompactionFilter::PrepareOptions(
+    const ConfigOptions& config_options) {
+  if (clock_ == nullptr) {
+    clock_ = config_options.env->GetSystemClock().get();
+  }
+  return LayeredCompactionFilterBase::PrepareOptions(config_options);
+}
+
+Status TtlCompactionFilter::ValidateOptions(
+    const DBOptions& db_opts, const ColumnFamilyOptions& cf_opts) const {
+  if (clock_ == nullptr) {
+    return Status::InvalidArgument(
+        "SystemClock required by TtlCompactionFilter");
+  } else {
+    return LayeredCompactionFilterBase::ValidateOptions(db_opts, cf_opts);
+  }
+}
+
+TtlCompactionFilterFactory::TtlCompactionFilterFactory(
+    int32_t ttl, SystemClock* clock,
+    std::shared_ptr<CompactionFilterFactory> comp_filter_factory)
+    : ttl_(ttl), clock_(clock), user_comp_filter_factory_(comp_filter_factory) {
+  RegisterOptions("UserOptions", &user_comp_filter_factory_,
+                  &ttl_cff_type_info);
+  RegisterOptions("TTL", &ttl_, &ttl_type_info);
+}
+
+std::unique_ptr<CompactionFilter>
+TtlCompactionFilterFactory::CreateCompactionFilter(
+    const CompactionFilter::Context& context) {
+  std::unique_ptr<const CompactionFilter> user_comp_filter_from_factory =
+      nullptr;
+  if (user_comp_filter_factory_) {
+    user_comp_filter_from_factory =
+        user_comp_filter_factory_->CreateCompactionFilter(context);
   }
+
+  return std::unique_ptr<TtlCompactionFilter>(new TtlCompactionFilter(
+      ttl_, clock_, nullptr, std::move(user_comp_filter_from_factory)));
 }
 
+Status TtlCompactionFilterFactory::PrepareOptions(
+    const ConfigOptions& config_options) {
+  if (clock_ == nullptr) {
+    clock_ = config_options.env->GetSystemClock().get();
+  }
+  return CompactionFilterFactory::PrepareOptions(config_options);
+}
+
+Status TtlCompactionFilterFactory::ValidateOptions(
+    const DBOptions& db_opts, const ColumnFamilyOptions& cf_opts) const {
+  if (clock_ == nullptr) {
+    return Status::InvalidArgument(
+        "SystemClock required by TtlCompactionFilterFactory");
+  } else {
+    return CompactionFilterFactory::ValidateOptions(db_opts, cf_opts);
+  }
+}
+
+int RegisterTtlObjects(ObjectLibrary& library, const std::string& /*arg*/) {
+  library.AddFactory<MergeOperator>(
+      TtlMergeOperator::kClassName(),
+      [](const std::string& /*uri*/, std::unique_ptr<MergeOperator>* guard,
+         std::string* /* errmsg */) {
+        guard->reset(new TtlMergeOperator(nullptr, nullptr));
+        return guard->get();
+      });
+  library.AddFactory<CompactionFilterFactory>(
+      TtlCompactionFilterFactory::kClassName(),
+      [](const std::string& /*uri*/,
+         std::unique_ptr<CompactionFilterFactory>* guard,
+         std::string* /* errmsg */) {
+        guard->reset(new TtlCompactionFilterFactory(0, nullptr, nullptr));
+        return guard->get();
+      });
+  library.AddFactory<CompactionFilter>(
+      TtlCompactionFilter::kClassName(),
+      [](const std::string& /*uri*/,
+         std::unique_ptr<CompactionFilter>* /*guard*/,
+         std::string* /* errmsg */) {
+        return new TtlCompactionFilter(0, nullptr, nullptr);
+      });
+  size_t num_types;
+  return static_cast<int>(library.GetFactoryCount(&num_types));
+}
 // Open the db inside DBWithTTLImpl because options needs pointer to its ttl
 DBWithTTLImpl::DBWithTTLImpl(DB* db) : DBWithTTL(db), closed_(false) {}
 
@@ -55,21 +327,15 @@ Status DBWithTTLImpl::Close() {
   return ret;
 }
 
-Status UtilityDB::OpenTtlDB(const Options& options, const std::string& dbname,
-                            StackableDB** dbptr, int32_t ttl, bool read_only) {
-  DBWithTTL* db;
-  Status s = DBWithTTL::Open(options, dbname, &db, ttl, read_only);
-  if (s.ok()) {
-    *dbptr = db;
-  } else {
-    *dbptr = nullptr;
-  }
-  return s;
+void DBWithTTLImpl::RegisterTtlClasses() {
+  static std::once_flag once;
+  std::call_once(once, [&]() {
+    ObjectRegistry::Default()->AddLibrary("TTL", RegisterTtlObjects, "");
+  });
 }
 
 Status DBWithTTL::Open(const Options& options, const std::string& dbname,
                        DBWithTTL** dbptr, int32_t ttl, bool read_only) {
-
   DBOptions db_options(options);
   ColumnFamilyOptions cf_options(options);
   std::vector<ColumnFamilyDescriptor> column_families;
@@ -92,17 +358,21 @@ Status DBWithTTL::Open(
     const std::vector<ColumnFamilyDescriptor>& column_families,
     std::vector<ColumnFamilyHandle*>* handles, DBWithTTL** dbptr,
     const std::vector<int32_t>& ttls, bool read_only) {
+  DBWithTTLImpl::RegisterTtlClasses();
   if (ttls.size() != column_families.size()) {
     return Status::InvalidArgument(
         "ttls size has to be the same as number of column families");
   }
 
+  SystemClock* clock = (db_options.env == nullptr)
+                           ? SystemClock::Default().get()
+                           : db_options.env->GetSystemClock().get();
+
   std::vector<ColumnFamilyDescriptor> column_families_sanitized =
       column_families;
   for (size_t i = 0; i < column_families_sanitized.size(); ++i) {
     DBWithTTLImpl::SanitizeOptions(
-        ttls[i], &column_families_sanitized[i].options,
-        db_options.env == nullptr ? Env::Default() : db_options.env);
+        ttls[i], &column_families_sanitized[i].options, clock);
   }
   DB* db;
 
@@ -124,8 +394,10 @@ Status DBWithTTL::Open(
 Status DBWithTTLImpl::CreateColumnFamilyWithTtl(
     const ColumnFamilyOptions& options, const std::string& column_family_name,
     ColumnFamilyHandle** handle, int ttl) {
+  RegisterTtlClasses();
   ColumnFamilyOptions sanitized_options = options;
-  DBWithTTLImpl::SanitizeOptions(ttl, &sanitized_options, GetEnv());
+  DBWithTTLImpl::SanitizeOptions(ttl, &sanitized_options,
+                                 GetEnv()->GetSystemClock().get());
 
   return DBWithTTL::CreateColumnFamily(sanitized_options, column_family_name,
                                        handle);
@@ -140,11 +412,11 @@ Status DBWithTTLImpl::CreateColumnFamily(const ColumnFamilyOptions& options,
 // Appends the current timestamp to the string.
 // Returns false if could not get the current_time, true if append succeeds
 Status DBWithTTLImpl::AppendTS(const Slice& val, std::string* val_with_ts,
-                               Env* env) {
+                               SystemClock* clock) {
   val_with_ts->reserve(kTSLength + val.size());
   char ts_string[kTSLength];
   int64_t curtime;
-  Status st = env->GetCurrentTime(&curtime);
+  Status st = clock->GetCurrentTime(&curtime);
   if (!st.ok()) {
     return st;
   }
@@ -170,12 +442,13 @@ Status DBWithTTLImpl::SanityCheckTimestamp(const Slice& str) {
 }
 
 // Checks if the string is stale or not according to TTl provided
-bool DBWithTTLImpl::IsStale(const Slice& value, int32_t ttl, Env* env) {
+bool DBWithTTLImpl::IsStale(const Slice& value, int32_t ttl,
+                            SystemClock* clock) {
   if (ttl <= 0) {  // Data is fresh if TTL is non-positive
     return false;
   }
   int64_t curtime;
-  if (!env->GetCurrentTime(&curtime).ok()) {
+  if (!clock->GetCurrentTime(&curtime).ok()) {
     return false;  // Treat the data as fresh if could not get current time
   }
   int32_t timestamp_value =
@@ -273,12 +546,12 @@ Status DBWithTTLImpl::Merge(const WriteOptions& options,
 Status DBWithTTLImpl::Write(const WriteOptions& opts, WriteBatch* updates) {
   class Handler : public WriteBatch::Handler {
    public:
-    explicit Handler(Env* env) : env_(env) {}
+    explicit Handler(SystemClock* clock) : clock_(clock) {}
     WriteBatch updates_ttl;
     Status PutCF(uint32_t column_family_id, const Slice& key,
                  const Slice& value) override {
       std::string value_with_ts;
-      Status st = AppendTS(value, &value_with_ts, env_);
+      Status st = AppendTS(value, &value_with_ts, clock_);
       if (!st.ok()) {
         return st;
       }
@@ -288,7 +561,7 @@ Status DBWithTTLImpl::Write(const WriteOptions& opts, WriteBatch* updates) {
     Status MergeCF(uint32_t column_family_id, const Slice& key,
                    const Slice& value) override {
       std::string value_with_ts;
-      Status st = AppendTS(value, &value_with_ts, env_);
+      Status st = AppendTS(value, &value_with_ts, clock_);
       if (!st.ok()) {
         return st;
       }
@@ -298,12 +571,17 @@ Status DBWithTTLImpl::Write(const WriteOptions& opts, WriteBatch* updates) {
     Status DeleteCF(uint32_t column_family_id, const Slice& key) override {
       return WriteBatchInternal::Delete(&updates_ttl, column_family_id, key);
     }
+    Status DeleteRangeCF(uint32_t column_family_id, const Slice& begin_key,
+                         const Slice& end_key) override {
+      return WriteBatchInternal::DeleteRange(&updates_ttl, column_family_id,
+                                             begin_key, end_key);
+    }
     void LogData(const Slice& blob) override { updates_ttl.PutLogData(blob); }
 
    private:
-    Env* env_;
+    SystemClock* clock_;
   };
-  Handler handler(GetEnv());
+  Handler handler(GetEnv()->GetSystemClock().get());
   Status st = updates->Iterate(&handler);
   if (!st.ok()) {
     return st;