@nxtedition/rocksdb 13.5.13 → 15.0.1

package/deps/rocksdb/rocksdb/db/c.cc

@@ -82,6 +82,7 @@ using ROCKSDB_NAMESPACE::DbPath;
 using ROCKSDB_NAMESPACE::Env;
 using ROCKSDB_NAMESPACE::EnvOptions;
 using ROCKSDB_NAMESPACE::EventListener;
+using ROCKSDB_NAMESPACE::ExportImportFilesMetaData;
 using ROCKSDB_NAMESPACE::ExternalFileIngestionInfo;
 using ROCKSDB_NAMESPACE::FileLock;
 using ROCKSDB_NAMESPACE::FilterPolicy;
@@ -89,6 +90,7 @@ using ROCKSDB_NAMESPACE::FlushJobInfo;
 using ROCKSDB_NAMESPACE::FlushOptions;
 using ROCKSDB_NAMESPACE::HistogramData;
 using ROCKSDB_NAMESPACE::HyperClockCacheOptions;
+using ROCKSDB_NAMESPACE::ImportColumnFamilyOptions;
 using ROCKSDB_NAMESPACE::InfoLogLevel;
 using ROCKSDB_NAMESPACE::IngestExternalFileOptions;
 using ROCKSDB_NAMESPACE::Iterator;
@@ -245,6 +247,9 @@ struct rocksdb_write_buffer_manager_t {
 struct rocksdb_sst_file_manager_t {
   std::shared_ptr<SstFileManager> rep;
 };
+struct rocksdb_livefile_t {
+  LiveFileMetaData rep;
+};
 struct rocksdb_livefiles_t {
   std::vector<LiveFileMetaData> rep;
 };
@@ -255,6 +260,12 @@ struct rocksdb_column_family_handle_t {
 struct rocksdb_column_family_metadata_t {
   ColumnFamilyMetaData rep;
 };
+struct rocksdb_export_import_files_metadata_t {
+  ExportImportFilesMetaData* rep;
+};
+struct rocksdb_import_column_family_options_t {
+  ImportColumnFamilyOptions rep;
+};
 struct rocksdb_level_metadata_t {
   const LevelMetaData* rep;
 };
@@ -947,6 +958,22 @@ void rocksdb_checkpoint_create(rocksdb_checkpoint_t* checkpoint,
                         std::string(checkpoint_dir), log_size_for_flush));
 }
 
+rocksdb_export_import_files_metadata_t* rocksdb_checkpoint_export_column_family(
+    rocksdb_checkpoint_t* checkpoint,
+    rocksdb_column_family_handle_t* column_family, const char* export_dir,
+    char** errptr) {
+  ExportImportFilesMetaData* metadata = nullptr;
+  if (SaveError(errptr,
+                checkpoint->rep->ExportColumnFamily(
+                    column_family->rep, std::string(export_dir), &metadata))) {
+    return nullptr;
+  }
+  rocksdb_export_import_files_metadata_t* result =
+      new rocksdb_export_import_files_metadata_t;
+  result->rep = metadata;
+  return result;
+}
+
 void rocksdb_checkpoint_object_destroy(rocksdb_checkpoint_t* checkpoint) {
   delete checkpoint->rep;
   delete checkpoint;
@@ -1190,6 +1217,26 @@ rocksdb_column_family_handle_t** rocksdb_create_column_families(
   return c_handles;
 }
 
+rocksdb_column_family_handle_t* rocksdb_create_column_family_with_import(
+    rocksdb_t* db, rocksdb_options_t* column_family_options,
+    const char* column_family_name,
+    rocksdb_import_column_family_options_t* import_options,
+    rocksdb_export_import_files_metadata_t* export_import_files_metadata,
+    char** errptr) {
+  rocksdb_column_family_handle_t* handle = new rocksdb_column_family_handle_t;
+  handle->rep = nullptr;
+  if (SaveError(errptr,
+                db->rep->CreateColumnFamilyWithImport(
+                    ColumnFamilyOptions(column_family_options->rep),
+                    std::string(column_family_name), import_options->rep,
+                    *(export_import_files_metadata->rep), &(handle->rep)))) {
+    delete handle;
+    return nullptr;
+  }
+  handle->immortal = false;
+  return handle;
+}
+
 void rocksdb_create_column_families_destroy(
     rocksdb_column_family_handle_t** list) {
   free(list);
@@ -3639,6 +3686,11 @@ void rocksdb_options_set_write_buffer_manager(
   opt->rep.write_buffer_manager = wbm->rep;
 }
 
+void rocksdb_options_set_sst_file_manager(rocksdb_options_t* opt,
+                                          rocksdb_sst_file_manager_t* sfm) {
+  opt->rep.sst_file_manager = sfm->rep;
+}
+
 size_t rocksdb_options_get_write_buffer_size(rocksdb_options_t* opt) {
   return opt->rep.write_buffer_size;
 }
@@ -4746,6 +4798,15 @@ void rocksdb_options_add_compact_on_deletion_collector_factory_del_ratio(
   opt->rep.table_properties_collector_factories.emplace_back(compact_on_del);
 }
 
+void rocksdb_options_add_compact_on_deletion_collector_factory_min_file_size(
+    rocksdb_options_t* opt, size_t window_size, size_t num_dels_trigger,
+    double deletion_ratio, uint64_t min_file_size) {
+  std::shared_ptr<ROCKSDB_NAMESPACE::TablePropertiesCollectorFactory>
+      compact_on_del = NewCompactOnDeletionCollectorFactory(
+          window_size, num_dels_trigger, deletion_ratio, min_file_size);
+  opt->rep.table_properties_collector_factories.emplace_back(compact_on_del);
+}
+
 void rocksdb_set_perf_level(int v) {
   PerfLevel level = static_cast<PerfLevel>(v);
   SetPerfLevel(level);
@@ -6195,6 +6256,10 @@ void rocksdb_options_set_min_level_to_compress(rocksdb_options_t* opt,
   }
 }
 
+rocksdb_livefiles_t* rocksdb_livefiles_create() {
+  return new rocksdb_livefiles_t;
+}
+
 int rocksdb_livefiles_count(const rocksdb_livefiles_t* lf) {
   return static_cast<int>(lf->rep.size());
 }
@@ -6208,6 +6273,16 @@ const char* rocksdb_livefiles_name(const rocksdb_livefiles_t* lf, int index) {
   return lf->rep[index].name.c_str();
 }
 
+const char* rocksdb_livefiles_directory(const rocksdb_livefiles_t* lf,
+                                        int index) {
+  if (lf->rep[index].directory.empty()) {
+    // db_path is deprecated but still returned by some code paths
+    return lf->rep[index].db_path.c_str();
+  } else {
+    return lf->rep[index].directory.c_str();
+  }
+}
+
 int rocksdb_livefiles_level(const rocksdb_livefiles_t* lf, int index) {
   return lf->rep[index].level;
 }
@@ -6228,6 +6303,16 @@ const char* rocksdb_livefiles_largestkey(const rocksdb_livefiles_t* lf,
   return lf->rep[index].largestkey.data();
 }
 
+uint64_t rocksdb_livefiles_smallest_seqno(const rocksdb_livefiles_t* lf,
+                                          int index) {
+  return lf->rep[index].smallest_seqno;
+}
+
+uint64_t rocksdb_livefiles_largest_seqno(const rocksdb_livefiles_t* lf,
+                                         int index) {
+  return lf->rep[index].largest_seqno;
+}
+
 uint64_t rocksdb_livefiles_entries(const rocksdb_livefiles_t* lf, int index) {
   return lf->rep[index].num_entries;
 }
@@ -6238,6 +6323,71 @@ uint64_t rocksdb_livefiles_deletions(const rocksdb_livefiles_t* lf, int index) {
 
 void rocksdb_livefiles_destroy(const rocksdb_livefiles_t* lf) { delete lf; }
 
+rocksdb_livefile_t* rocksdb_livefile_create() { return new rocksdb_livefile_t; }
+
+void rocksdb_livefile_set_column_family_name(rocksdb_livefile_t* lf,
+                                             const char* column_family_name) {
+  lf->rep.column_family_name = std::string(column_family_name);
+}
+
+void rocksdb_livefile_set_level(rocksdb_livefile_t* lf, int level) {
+  lf->rep.level = level;
+}
+
+void rocksdb_livefile_set_name(rocksdb_livefile_t* lf, const char* name) {
+  lf->rep.name = std::string(name);
+}
+
+void rocksdb_livefile_set_directory(rocksdb_livefile_t* lf,
+                                    const char* directory) {
+  lf->rep.directory = std::string(directory);
+  lf->rep.db_path = std::string(directory);  // deprecated but still needed
+}
+
+void rocksdb_livefile_set_size(rocksdb_livefile_t* lf, size_t size) {
+  lf->rep.size = size;
+}
+
+void rocksdb_livefile_set_smallest_key(rocksdb_livefile_t* lf,
+                                       const char* smallest_key,
+                                       size_t smallest_key_len) {
+  lf->rep.smallestkey = std::string(smallest_key, smallest_key_len);
+}
+
+void rocksdb_livefile_set_largest_key(rocksdb_livefile_t* lf,
+                                      const char* largest_key,
+                                      size_t largest_key_len) {
+  lf->rep.largestkey = std::string(largest_key, largest_key_len);
+}
+
+void rocksdb_livefile_set_smallest_seqno(rocksdb_livefile_t* lf,
+                                         uint64_t smallest_seqno) {
+  lf->rep.smallest_seqno = smallest_seqno;
+}
+
+void rocksdb_livefile_set_largest_seqno(rocksdb_livefile_t* lf,
+                                        uint64_t largest_seqno) {
+  lf->rep.largest_seqno = largest_seqno;
+}
+
+void rocksdb_livefile_set_num_entries(rocksdb_livefile_t* lf,
+                                      uint64_t num_entries) {
+  lf->rep.num_entries = num_entries;
+}
+
+void rocksdb_livefile_set_num_deletions(rocksdb_livefile_t* lf,
+                                        uint64_t num_deletions) {
+  lf->rep.num_deletions = num_deletions;
+}
+
+void rocksdb_livefile_destroy(rocksdb_livefile_t* lf) { delete lf; }
+
+void rocksdb_livefiles_add(rocksdb_livefiles_t* lf,
+                           rocksdb_livefile_t* livefile) {
+  lf->rep.push_back(std::move(livefile->rep));
+  delete livefile;
+}
+
 void rocksdb_get_options_from_string(const rocksdb_options_t* base_options,
                                      const char* opts_str,
                                      rocksdb_options_t* new_options,
@@ -6388,6 +6538,58 @@ char* rocksdb_sst_file_metadata_get_largestkey(
   return CopyString(file_meta->rep->largestkey);
 }
 
+rocksdb_import_column_family_options_t*
+rocksdb_import_column_family_options_create() {
+  return new rocksdb_import_column_family_options_t;
+}
+
+void rocksdb_import_column_family_options_set_move_files(
+    rocksdb_import_column_family_options_t* opt, unsigned char v) {
+  opt->rep.move_files = v;
+}
+
+void rocksdb_import_column_family_options_destroy(
+    rocksdb_import_column_family_options_t* metadata) {
+  delete metadata;
+}
+
+rocksdb_export_import_files_metadata_t*
+rocksdb_export_import_files_metadata_create() {
+  auto metadata = new rocksdb_export_import_files_metadata_t;
+  metadata->rep = new ExportImportFilesMetaData;
+  return metadata;
+}
+
+char* rocksdb_export_import_files_metadata_get_db_comparator_name(
+    rocksdb_export_import_files_metadata_t* metadata) {
+  return strdup(metadata->rep->db_comparator_name.c_str());
+}
+
+void rocksdb_export_import_files_metadata_set_db_comparator_name(
+    rocksdb_export_import_files_metadata_t* metadata, const char* name) {
+  metadata->rep->db_comparator_name = std::string(name);
+}
+
+rocksdb_livefiles_t* rocksdb_export_import_files_metadata_get_files(
+    rocksdb_export_import_files_metadata_t* export_import_metadata) {
+  auto files = new rocksdb_livefiles_t;
+  files->rep = std::vector(export_import_metadata->rep->files);
+  return files;
+}
+
+void rocksdb_export_import_files_metadata_set_files(
+    rocksdb_export_import_files_metadata_t* metadata,
+    rocksdb_livefiles_t* files) {
+  metadata->rep->files = std::move(files->rep);
+  delete files;
+}
+
+void rocksdb_export_import_files_metadata_destroy(
+    rocksdb_export_import_files_metadata_t* metadata) {
+  delete metadata->rep;
+  delete metadata;
+}
+
 /* Transactions */
 
 rocksdb_transactiondb_options_t* rocksdb_transactiondb_options_create() {
@@ -6419,6 +6621,11 @@ void rocksdb_transactiondb_options_set_default_lock_timeout(
   opt->rep.default_lock_timeout = default_lock_timeout;
 }
 
+void rocksdb_transactiondb_options_set_use_per_key_point_lock_mgr(
+    rocksdb_transactiondb_options_t* opt, int use_per_key_point_lock_mgr) {
+  opt->rep.use_per_key_point_lock_mgr = use_per_key_point_lock_mgr;
+}
+
 rocksdb_transaction_options_t* rocksdb_transaction_options_create() {
   return new rocksdb_transaction_options_t;
 }

package/deps/rocksdb/rocksdb/db/c_test.c

@@ -1036,6 +1036,78 @@ int main(int argc, char** argv) {
     rocksdb_options_set_error_if_exists(options, 1);
   }
 
+  StartPhase("checkpoint_export_column_family");
+  {
+    static char cf_export_path[200];
+    static char db_import_path[200];
+    snprintf(cf_export_path, sizeof(cf_export_path),
+             "%s/rocksdb_c_test-%d-cf_export", GetTempDir(), ((int)geteuid()));
+    snprintf(db_import_path, sizeof(db_import_path),
+             "%s/rocksdb_c_test-%d-db_import", GetTempDir(), ((int)geteuid()));
+
+    rocksdb_options_t* db_options = rocksdb_options_create();
+    rocksdb_column_family_handle_t* cf_export =
+        rocksdb_create_column_family(db, db_options, "cf_export", &err);
+    CheckNoError(err);
+
+    rocksdb_put_cf(db, woptions, cf_export, "k1", 2, "v1", 2, &err);
+    CheckNoError(err);
+    rocksdb_put_cf(db, woptions, cf_export, "k2", 2, "v2", 2, &err);
+    CheckNoError(err);
+
+    rocksdb_checkpoint_t* checkpoint =
+        rocksdb_checkpoint_object_create(db, &err);
+    CheckNoError(err);
+
+    rocksdb_export_import_files_metadata_t* export_metadata =
+        rocksdb_checkpoint_export_column_family(checkpoint, cf_export,
+                                                cf_export_path, &err);
+    CheckNoError(err);
+    const char* comparator_name =
+        rocksdb_export_import_files_metadata_get_db_comparator_name(
+            export_metadata);
+    CheckEqual("leveldb.BytewiseComparator", comparator_name, 26);
+    rocksdb_free((void*)comparator_name);
+    rocksdb_checkpoint_object_destroy(checkpoint);
+    checkpoint = NULL;
+    rocksdb_drop_column_family(db, cf_export, &err);
+    CheckNoError(err);
+    rocksdb_column_family_handle_destroy(cf_export);
+    rocksdb_options_set_create_if_missing(db_options, 1);
+    rocksdb_options_set_error_if_exists(db_options, 1);
+    rocksdb_t* db_import = rocksdb_open(db_options, db_import_path, &err);
+    CheckNoError(err);
+    rocksdb_import_column_family_options_t* import_options =
+        rocksdb_import_column_family_options_create();
+    rocksdb_column_family_handle_t* cf_import =
+        rocksdb_create_column_family_with_import(db_import, db_options,
+                                                 "cf_import", import_options,
+                                                 export_metadata, &err);
+    CheckNoError(err);
+    rocksdb_import_column_family_options_destroy(import_options);
+    rocksdb_export_import_files_metadata_destroy(export_metadata);
+    size_t val_len;
+    char* val =
+        rocksdb_get_cf(db_import, roptions, cf_import, "k1", 2, &val_len, &err);
+    CheckNoError(err);
+    CheckEqual("v1", val, val_len);
+    free(val);
+
+    val =
+        rocksdb_get_cf(db_import, roptions, cf_import, "k2", 2, &val_len, &err);
+    CheckNoError(err);
+    CheckEqual("v2", val, val_len);
+    free(val);
+
+    rocksdb_column_family_handle_destroy(cf_import);
+    cf_import = NULL;
+    rocksdb_close(db_import);
+    rocksdb_destroy_db(db_options, db_import_path, &err);
+    CheckNoError(err);
+    rocksdb_options_destroy(db_options);
+    db_options = NULL;
+  }
+
   StartPhase("compactall");
   rocksdb_compact_range(db, NULL, 0, NULL, 0);
   CheckGet(db, roptions, "foo", "hello");

package/deps/rocksdb/rocksdb/db/column_family.cc

@@ -280,7 +280,8 @@ ColumnFamilyOptions SanitizeCfOptions(const ImmutableDBOptions& db_options,
   }
 
   if (result.compaction_style == kCompactionStyleUniversal &&
-      db_options.allow_ingest_behind && result.num_levels < 3) {
+      (db_options.allow_ingest_behind || result.cf_allow_ingest_behind) &&
+      result.num_levels < 3) {
     result.num_levels = 3;
   }
 
@@ -1331,7 +1332,7 @@ Compaction* ColumnFamilyData::CompactRange(
     const InternalKey* begin, const InternalKey* end,
     InternalKey** compaction_end, bool* conflict,
     uint64_t max_file_num_to_ignore, const std::string& trim_ts) {
-  auto* result = compaction_picker_->CompactRange(
+  auto* result = compaction_picker_->PickCompactionForCompactRange(
       GetName(), mutable_cf_options, mutable_db_options,
       current_->storage_info(), input_level, output_level,
       compact_range_options, begin, end, compaction_end, conflict,

package/deps/rocksdb/rocksdb/db/column_family.h

@@ -600,6 +600,11 @@ class ColumnFamilyData {
     return (mem_->IsEmpty() ? 0 : 1) + imm_.NumNotFlushed();
   }
 
+  // thread-safe, DB mutex not needed.
+  bool AllowIngestBehind() const {
+    return ioptions_.cf_allow_ingest_behind || ioptions_.allow_ingest_behind;
+  }
+
  private:
   friend class ColumnFamilySet;
   ColumnFamilyData(

package/deps/rocksdb/rocksdb/db/compact_files_test.cc

@@ -441,6 +441,10 @@ TEST_F(CompactFilesTest, SentinelCompressionType) {
 }
 
 TEST_F(CompactFilesTest, CompressionWithBlockAlign) {
+  if (!Snappy_Supported()) {
+    ROCKSDB_GTEST_SKIP("Test requires Snappy support");
+    return;
+  }
   Options options;
   options.compression = CompressionType::kNoCompression;
   options.create_if_missing = true;

package/deps/rocksdb/rocksdb/db/compaction/compaction.cc

@@ -647,6 +647,8 @@ bool Compaction::KeyNotExistsBeyondOutputLevel(
     return true;
   } else if (output_level_ != 0 &&
              cfd_->ioptions().compaction_style == kCompactionStyleLevel) {
+    // TODO: apply the optimization here to other compaction styles and
+    // compaction/flush to L0.
     // Maybe use binary search to find right entry instead of linear search?
     const Comparator* user_cmp = cfd_->user_comparator();
     for (int lvl = output_level_ + 1; lvl < number_levels_; lvl++) {

package/deps/rocksdb/rocksdb/db/compaction/compaction_iterator.cc

@@ -28,7 +28,7 @@ CompactionIterator::CompactionIterator(
     SequenceNumber earliest_snapshot,
     SequenceNumber earliest_write_conflict_snapshot,
     SequenceNumber job_snapshot, const SnapshotChecker* snapshot_checker,
-    Env* env, bool report_detailed_time, bool expect_valid_internal_key,
+    Env* env, bool report_detailed_time,
     CompactionRangeDelAggregator* range_del_agg,
     BlobFileBuilder* blob_file_builder, bool allow_data_in_errors,
     bool enforce_single_del_contracts,
@@ -42,8 +42,8 @@ CompactionIterator::CompactionIterator(
     : CompactionIterator(
           input, cmp, merge_helper, last_sequence, snapshots, earliest_snapshot,
           earliest_write_conflict_snapshot, job_snapshot, snapshot_checker, env,
-          report_detailed_time, expect_valid_internal_key, range_del_agg,
-          blob_file_builder, allow_data_in_errors, enforce_single_del_contracts,
+          report_detailed_time, range_del_agg, blob_file_builder,
+          allow_data_in_errors, enforce_single_del_contracts,
           manual_compaction_canceled,
           compaction ? std::make_unique<RealCompaction>(compaction) : nullptr,
           must_count_input_entries, compaction_filter, shutting_down, info_log,
@@ -55,7 +55,7 @@ CompactionIterator::CompactionIterator(
     SequenceNumber earliest_snapshot,
     SequenceNumber earliest_write_conflict_snapshot,
     SequenceNumber job_snapshot, const SnapshotChecker* snapshot_checker,
-    Env* env, bool report_detailed_time, bool expect_valid_internal_key,
+    Env* env, bool report_detailed_time,
     CompactionRangeDelAggregator* range_del_agg,
     BlobFileBuilder* blob_file_builder, bool allow_data_in_errors,
     bool enforce_single_del_contracts,
@@ -76,16 +76,14 @@ CompactionIterator::CompactionIterator(
       env_(env),
       clock_(env_->GetSystemClock().get()),
       report_detailed_time_(report_detailed_time),
-      expect_valid_internal_key_(expect_valid_internal_key),
       range_del_agg_(range_del_agg),
       blob_file_builder_(blob_file_builder),
       compaction_(std::move(compaction)),
       compaction_filter_(compaction_filter),
       shutting_down_(shutting_down),
       manual_compaction_canceled_(manual_compaction_canceled),
-      bottommost_level_(!compaction_ ? false
-                                     : compaction_->bottommost_level() &&
-                                           !compaction_->allow_ingest_behind()),
+      bottommost_level_(compaction_ && compaction_->bottommost_level() &&
+                        !compaction_->allow_ingest_behind()),
       // snapshots_ cannot be nullptr, but we will assert later in the body of
       // the constructor.
       visible_at_tip_(snapshots_ ? snapshots_->empty() : false),
@@ -161,6 +159,7 @@ void CompactionIterator::Next() {
       // MergeUntil stops when it encounters a corrupt key and does not
       // include them in the result, so we expect the keys here to be valid.
       if (!s.ok()) {
+        // FIXME: should fail compaction after this fatal logging.
         ROCKS_LOG_FATAL(
             info_log_, "Invalid ikey %s in compaction. %s",
             allow_data_in_errors_ ? key_.ToString(true).c_str() : "hidden",
@@ -464,18 +463,9 @@ void CompactionIterator::NextFromInput() {
     if (!pik_status.ok()) {
       iter_stats_.num_input_corrupt_records++;
 
-      // If `expect_valid_internal_key_` is false, return the corrupted key
-      // and let the caller decide what to do with it.
-      if (expect_valid_internal_key_) {
-        status_ = pik_status;
-        return;
-      }
-      key_ = current_key_.SetInternalKey(key_);
-      has_current_user_key_ = false;
-      current_user_key_sequence_ = kMaxSequenceNumber;
-      current_user_key_snapshot_ = 0;
-      validity_info_.SetValid(ValidContext::kParseKeyError);
-      break;
+      // Always fail compaction when encountering corrupted internal keys
+      status_ = pik_status;
+      return;
     }
     TEST_SYNC_POINT_CALLBACK("CompactionIterator:ProcessKV", &ikey_);
     if (is_range_del_) {
@@ -642,7 +632,8 @@ void CompactionIterator::NextFromInput() {
     } else if (ikey_.type == kTypeSingleDeletion) {
       // We can compact out a SingleDelete if:
       // 1) We encounter the corresponding PUT -OR- we know that this key
-      //    doesn't appear past this output level
+      //    doesn't appear past this output level and  we are not in
+      //    ingest_behind mode.
       // =AND=
       // 2) We've already returned a record in this snapshot -OR-
       //    there are no earlier earliest_write_conflict_snapshot.
@@ -731,6 +722,8 @@ void CompactionIterator::NextFromInput() {
             "CompactionIterator::NextFromInput:SingleDelete:1",
             const_cast<Compaction*>(c));
         if (last_key_seq_zeroed_) {
+          // Drop SD and the next key since they are both in the last
+          // snapshot (since last key has seqno zeroed).
           ++iter_stats_.num_record_drop_hidden;
           ++iter_stats_.num_record_drop_obsolete;
           assert(bottommost_level_);
@@ -841,7 +834,7 @@ void CompactionIterator::NextFromInput() {
         // iteration. If the next key is corrupt, we return before the
         // comparison, so the value of has_current_user_key does not matter.
         has_current_user_key_ = false;
-        if (compaction_ != nullptr &&
+        if (compaction_ != nullptr && !compaction_->allow_ingest_behind() &&
             DefinitelyInSnapshot(ikey_.sequence, earliest_snapshot_) &&
             compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                        &level_ptrs_) &&
@@ -854,6 +847,9 @@ void CompactionIterator::NextFromInput() {
             ++iter_stats_.num_optimized_del_drop_obsolete;
           }
         } else if (last_key_seq_zeroed_) {
+          // Sequence number zeroing requires bottommost_level_, which is
+          // false with ingest_behind.
+          assert(!compaction_->allow_ingest_behind());
           // Skip.
           ++iter_stats_.num_record_drop_hidden;
           ++iter_stats_.num_record_drop_obsolete;
@@ -870,6 +866,7 @@ void CompactionIterator::NextFromInput() {
     } else if (last_sequence != kMaxSequenceNumber &&
                (last_snapshot == current_user_key_snapshot_ ||
                 last_snapshot < current_user_key_snapshot_)) {
+      // rule (A):
       // If the earliest snapshot is which this key is visible in
       // is the same as the visibility of a previous instance of the
       // same key, then this kv is not visible in any snapshot.
@@ -878,6 +875,15 @@ void CompactionIterator::NextFromInput() {
       // Note: Dropping this key will not affect TransactionDB write-conflict
       // checking since there has already been a record returned for this key
       // in this snapshot.
+      // When ingest_behind is enabled, it's ok that we drop an overwritten
+      // Delete here. The overwritting key still covers whatever that will be
+      // ingested. Note that we will not drop SingleDelete here as SingleDelte
+      // is handled entirely in its own if clause. This is important, see
+      // example: from new to old: SingleDelete_1, PUT_1, SingleDelete_2, PUT_2,
+      // where all operations are on the same key and PUT_2 is ingested with
+      // ingest_behind=true. If SingleDelete_2 is dropped due to being compacted
+      // together with PUT_1, and then PUT_1 is compacted away together with
+      // SingleDelete_1, PUT_2 can incorrectly becomes visible.
       if (last_sequence < current_user_key_sequence_) {
         ROCKS_LOG_FATAL(info_log_,
                         "key %s, last_sequence (%" PRIu64
@@ -887,12 +893,13 @@ void CompactionIterator::NextFromInput() {
         assert(false);
       }
 
-      ++iter_stats_.num_record_drop_hidden;  // rule (A)
+      ++iter_stats_.num_record_drop_hidden;
       AdvanceInputIter();
     } else if (compaction_ != nullptr &&
                (ikey_.type == kTypeDeletion ||
                 (ikey_.type == kTypeDeletionWithTimestamp &&
                  cmp_with_history_ts_low_ < 0)) &&
+               !compaction_->allow_ingest_behind() &&
                DefinitelyInSnapshot(ikey_.sequence, earliest_snapshot_) &&
                compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                           &level_ptrs_)) {
@@ -928,11 +935,13 @@ void CompactionIterator::NextFromInput() {
                 (ikey_.type == kTypeDeletionWithTimestamp &&
                  cmp_with_history_ts_low_ < 0)) &&
                bottommost_level_) {
+      assert(compaction_);
+      assert(!compaction_->allow_ingest_behind());  // bottommost_level_ is true
       // Handle the case where we have a delete key at the bottom most level
       // We can skip outputting the key iff there are no subsequent puts for
       // this key
-      assert(!compaction_ || compaction_->KeyNotExistsBeyondOutputLevel(
-                                 ikey_.user_key, &level_ptrs_));
+      assert(compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
+                                                        &level_ptrs_));
       ParsedInternalKey next_ikey;
       AdvanceInputIter();
 #ifndef NDEBUG
@@ -974,6 +983,12 @@ void CompactionIterator::NextFromInput() {
                 (compaction_ != nullptr &&
                  compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                             &level_ptrs_)))) {
+      // FIXME: it's possible that we are setting sequence number to 0 as
+      // preferred sequence number here. If cf_ingest_behind is enabled, this
+      // may fail ingestions since they expect all keys above the last level
+      // to have non-zero sequence number. We should probably not allow seqno
+      // zeroing here.
+      //
       // This section that attempts to swap preferred sequence number will not
       // be invoked if this is a CompactionIterator created for flush, since
       // `compaction_` will be nullptr and it's not bottommost either.
@@ -1105,17 +1120,15 @@ void CompactionIterator::NextFromInput() {
     }
   }
 
-  if (!Valid() && IsShuttingDown()) {
-    status_ = Status::ShutdownInProgress();
-  }
-
-  if (IsPausingManualCompaction()) {
-    status_ = Status::Incomplete(Status::SubCode::kManualCompactionPaused);
-  }
-
-  // Propagate corruption status from memtable itereator
-  if (!input_.Valid() && input_.status().IsCorruption()) {
-    status_ = input_.status();
+  if (status_.ok()) {
+    if (!Valid() && IsShuttingDown()) {
+      status_ = Status::ShutdownInProgress();
+    } else if (IsPausingManualCompaction()) {
+      status_ = Status::Incomplete(Status::SubCode::kManualCompactionPaused);
+    } else if (!input_.Valid() && input_.status().IsCorruption()) {
+      // Propagate corruption status from memtable iterator
+      status_ = input_.status();
+    }
   }
 }
 
@@ -1274,11 +1287,11 @@ void CompactionIterator::PrepareOutput() {
     //
     // Can we do the same for levels above bottom level as long as
     // KeyNotExistsBeyondOutputLevel() return true?
-    if (Valid() && compaction_ != nullptr &&
-        !compaction_->allow_ingest_behind() && bottommost_level_ &&
+    if (Valid() && bottommost_level_ &&
         DefinitelyInSnapshot(ikey_.sequence, earliest_snapshot_) &&
         ikey_.type != kTypeMerge && current_key_committed_ &&
         ikey_.sequence <= preserve_seqno_after_ && !is_range_del_) {
+      assert(compaction_ != nullptr && !compaction_->allow_ingest_behind());
       if (ikey_.type == kTypeDeletion ||
           (ikey_.type == kTypeSingleDeletion && timestamp_size_ == 0)) {
         ROCKS_LOG_FATAL(