User-defined timestamp support for DeleteRange() (facebook#10661)

Summary:
Add user-defined timestamp support for range deletion. The new API is `DeleteRange(opt, cf, begin_key, end_key, ts)`. Most of the change is to update the comparator to compare without timestamp. Other than that, major changes are
- internal range tombstone data structures (`FragmentedRangeTombstoneList`, `RangeTombstone`, etc.) to store timestamps.
- Garbage collection of range tombstones and range tombstone covered keys during compaction.
- Get()/MultiGet() to return the timestamp of a range tombstone when needed.
- Get/Iterator with range tombstones bounded by readoptions.timestamp.
- timestamp crash test now issues DeleteRange by default.

Pull Request resolved: facebook#10661

Test Plan:
- Added unit test: `make check`
- Stress test: `python3 tools/db_crashtest.py --enable_ts whitebox --readpercent=57 --prefixpercent=4 --writepercent=25 -delpercent=5 --iterpercent=5 --delrangepercent=4`
- Ran `db_bench` to measure regression when timestamp is not enabled. The tests are for write (with some range deletion) and iterate with DB fitting in memory: `./db_bench--benchmarks=fillrandom,seekrandom --writes_per_range_tombstone=200 --max_write_buffer_number=100 --min_write_buffer_number_to_merge=100 --writes=500000 --reads=500000 --seek_nexts=10 --disable_auto_compactions -disable_wal=true --max_num_range_tombstones=1000`.  Did not see consistent regression in no timestamp case.

| micros/op | fillrandom | seekrandom |
| --- | --- | --- |
|main| 2.58 |10.96|
|PR 10661| 2.68 |10.63|

Reviewed By: riversand963

Differential Revision: D39441192

Pulled By: cbi42

fbshipit-source-id: f05aca3c41605caf110daf0ff405919f300ddec2

HISTORY.md

@@ -1,5 +1,8 @@
 # Rocksdb Change Log
 ## Unreleased
+### New Features
+* `DeleteRange()` now supports user-defined timestamp.
+
 ### Bug Fixes
 * Fix a bug in io_uring_prep_cancel in AbortIO API for posix which expects sqe->addr to match with read request submitted and wrong paramter was being passed.
 * Fixed a regression in iterator performance when the entire DB is a single memtable introduced in #10449. The fix is in #10705 and #10716. 

db/builder.cc

@@ -90,7 +90,7 @@ Status BuildTable(
   iter->SeekToFirst();
   std::unique_ptr<CompactionRangeDelAggregator> range_del_agg(
       new CompactionRangeDelAggregator(&tboptions.internal_comparator,
-                                       snapshots));
+                                       snapshots, full_history_ts_low));
   uint64_t num_unfragmented_tombstones = 0;
   uint64_t total_tombstone_payload_bytes = 0;
   for (auto& range_del_iter : range_del_iters) {

db/compaction/compaction_iterator.cc

@@ -914,8 +914,17 @@ void CompactionIterator::NextFromInput() {
     } else {
       // 1. new user key -OR-
       // 2. different snapshot stripe
-      bool should_delete = range_del_agg_->ShouldDelete(
-          key_, RangeDelPositioningMode::kForwardTraversal);
+      // If user-defined timestamp is enabled, we consider keys for GC if they
+      // are below history_ts_low_. CompactionRangeDelAggregator::ShouldDelete()
+      // only considers range deletions that are at or below history_ts_low_ and
+      // trim_ts_. We drop keys here that are below history_ts_low_ and are
+      // covered by a range tombstone that is at or below history_ts_low_ and
+      // trim_ts.
+      bool should_delete = false;
+      if (!timestamp_size_ || cmp_with_history_ts_low_ < 0) {
+        should_delete = range_del_agg_->ShouldDelete(
+            key_, RangeDelPositioningMode::kForwardTraversal);
+      }
       if (should_delete) {
         ++iter_stats_.num_record_drop_hidden;
         ++iter_stats_.num_record_drop_range_del;

db/compaction/compaction_job.cc

@@ -1035,7 +1035,8 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   NotifyOnSubcompactionBegin(sub_compact);
 
   auto range_del_agg = std::make_unique<CompactionRangeDelAggregator>(
-      &cfd->internal_comparator(), existing_snapshots_);
+      &cfd->internal_comparator(), existing_snapshots_, &full_history_ts_low_,
+      &trim_ts_);
 
   // TODO: since we already use C++17, should use
   // std::optional<const Slice> instead.
@@ -1455,7 +1456,7 @@ Status CompactionJob::FinishCompactionOutputFile(
                                          : nullptr,
           sub_compact->end.has_value() ? &(sub_compact->end.value()) : nullptr,
           range_del_out_stats, bottommost_level_, cfd->internal_comparator(),
-          earliest_snapshot, next_table_min_key);
+          earliest_snapshot, next_table_min_key, full_history_ts_low_);
     }
     RecordDroppedKeys(range_del_out_stats, &sub_compact->compaction_job_stats);
     TEST_SYNC_POINT("CompactionJob::FinishCompactionOutputFile1");

db/compaction/compaction_outputs.cc

@@ -346,10 +346,10 @@ Status CompactionOutputs::AddToOutput(
 }
 
 Status CompactionOutputs::AddRangeDels(
-    const Slice* comp_start, const Slice* comp_end,
+    const Slice* comp_start_user_key, const Slice* comp_end_user_key,
     CompactionIterationStats& range_del_out_stats, bool bottommost_level,
     const InternalKeyComparator& icmp, SequenceNumber earliest_snapshot,
-    const Slice& next_table_min_key) {
+    const Slice& next_table_min_key, const std::string& full_history_ts_low) {
   assert(HasRangeDel());
   FileMetaData& meta = current_output().meta;
   const Comparator* ucmp = icmp.user_comparator();
@@ -363,7 +363,7 @@ Status CompactionOutputs::AddRangeDels(
   if (output_size == 1) {
     // For the first output table, include range tombstones before the min
     // key but after the subcompaction boundary.
-    lower_bound = comp_start;
+    lower_bound = comp_start_user_key;
     lower_bound_from_sub_compact = true;
   } else if (meta.smallest.size() > 0) {
     // For subsequent output tables, only include range tombstones from min
@@ -383,21 +383,22 @@ Status CompactionOutputs::AddRangeDels(
     // use the smaller key as the upper bound of the output file, to ensure
     // that there is no overlapping between different output files.
     upper_bound_guard = ExtractUserKey(next_table_min_key);
-    if (comp_end != nullptr &&
-        ucmp->Compare(upper_bound_guard, *comp_end) >= 0) {
-      upper_bound = comp_end;
+    if (comp_end_user_key != nullptr &&
+        ucmp->CompareWithoutTimestamp(upper_bound_guard, *comp_end_user_key) >=
+            0) {
+      upper_bound = comp_end_user_key;
     } else {
       upper_bound = &upper_bound_guard;
     }
   } else {
     // This is the last file in the subcompaction, so extend until the
     // subcompaction ends.
-    upper_bound = comp_end;
+    upper_bound = comp_end_user_key;
   }
   bool has_overlapping_endpoints;
   if (upper_bound != nullptr && meta.largest.size() > 0) {
-    has_overlapping_endpoints =
-        ucmp->Compare(meta.largest.user_key(), *upper_bound) == 0;
+    has_overlapping_endpoints = ucmp->CompareWithoutTimestamp(
+                                    meta.largest.user_key(), *upper_bound) == 0;
   } else {
     has_overlapping_endpoints = false;
   }
@@ -406,8 +407,8 @@ Status CompactionOutputs::AddRangeDels(
   // bound. If the end of subcompaction is null or the upper bound is null,
   // it means that this file is the last file in the compaction. So there
   // will be no overlapping between this file and others.
-  assert(comp_end == nullptr || upper_bound == nullptr ||
-         ucmp->Compare(*upper_bound, *comp_end) <= 0);
+  assert(comp_end_user_key == nullptr || upper_bound == nullptr ||
+         ucmp->CompareWithoutTimestamp(*upper_bound, *comp_end_user_key) <= 0);
   auto it = range_del_agg_->NewIterator(lower_bound, upper_bound,
                                         has_overlapping_endpoints);
   // Position the range tombstone output iterator. There may be tombstone
@@ -421,7 +422,8 @@ Status CompactionOutputs::AddRangeDels(
   for (; it->Valid(); it->Next()) {
     auto tombstone = it->Tombstone();
     if (upper_bound != nullptr) {
-      int cmp = ucmp->Compare(*upper_bound, tombstone.start_key_);
+      int cmp =
+          ucmp->CompareWithoutTimestamp(*upper_bound, tombstone.start_key_);
       if ((has_overlapping_endpoints && cmp < 0) ||
           (!has_overlapping_endpoints && cmp <= 0)) {
         // Tombstones starting after upper_bound only need to be included in
@@ -434,7 +436,17 @@ Status CompactionOutputs::AddRangeDels(
       }
     }
 
-    if (bottommost_level && tombstone.seq_ <= earliest_snapshot) {
+    const size_t ts_sz = ucmp->timestamp_size();
+    // Garbage collection for range tombstones.
+    // If user-defined timestamp is enabled, range tombstones are dropped if
+    // they are at bottommost_level, below full_history_ts_low and not visible
+    // in any snapshot. trim_ts_ is passed to the constructor for
+    // range_del_agg_, and range_del_agg_ internally drops tombstones above
+    // trim_ts_.
+    if (bottommost_level && tombstone.seq_ <= earliest_snapshot &&
+        (ts_sz == 0 ||
+         (!full_history_ts_low.empty() &&
+          ucmp->CompareTimestamp(tombstone.ts_, full_history_ts_low) < 0))) {
       // TODO(andrewkr): tombstones that span multiple output files are
       // counted for each compaction output file, so lots of double
       // counting.
@@ -445,12 +457,13 @@ Status CompactionOutputs::AddRangeDels(
 
     auto kv = tombstone.Serialize();
     assert(lower_bound == nullptr ||
-           ucmp->Compare(*lower_bound, kv.second) < 0);
+           ucmp->CompareWithoutTimestamp(*lower_bound, kv.second) < 0);
     // Range tombstone is not supported by output validator yet.
     builder_->Add(kv.first.Encode(), kv.second);
     InternalKey smallest_candidate = std::move(kv.first);
     if (lower_bound != nullptr &&
-        ucmp->Compare(smallest_candidate.user_key(), *lower_bound) <= 0) {
+        ucmp->CompareWithoutTimestamp(smallest_candidate.user_key(),
+                                      *lower_bound) <= 0) {
       // Pretend the smallest key has the same user key as lower_bound
       // (the max key in the previous table or subcompaction) in order for
       // files to appear key-space partitioned.
@@ -470,13 +483,23 @@ Status CompactionOutputs::AddRangeDels(
       // choose lowest seqnum so this file's smallest internal key comes
       // after the previous file's largest. The fake seqnum is OK because
       // the read path's file-picking code only considers user key.
-      smallest_candidate = InternalKey(
-          *lower_bound, lower_bound_from_sub_compact ? tombstone.seq_ : 0,
-          kTypeRangeDeletion);
+      if (lower_bound_from_sub_compact) {
+        if (ts_sz) {
+          assert(tombstone.ts_.size() == ts_sz);
+          smallest_candidate = InternalKey(*lower_bound, tombstone.seq_,
+                                           kTypeRangeDeletion, tombstone.ts_);
+        } else {
+          smallest_candidate =
+              InternalKey(*lower_bound, tombstone.seq_, kTypeRangeDeletion);
+        }
+      } else {
+        smallest_candidate = InternalKey(*lower_bound, 0, kTypeRangeDeletion);
+      }
     }
     InternalKey largest_candidate = tombstone.SerializeEndKey();
     if (upper_bound != nullptr &&
-        ucmp->Compare(*upper_bound, largest_candidate.user_key()) <= 0) {
+        ucmp->CompareWithoutTimestamp(*upper_bound,
+                                      largest_candidate.user_key()) <= 0) {
       // Pretend the largest key has the same user key as upper_bound (the
       // min key in the following table or subcompaction) in order for files
       // to appear key-space partitioned.
@@ -490,9 +513,22 @@ Status CompactionOutputs::AddRangeDels(
       // kMaxSequenceNumber), but with kTypeDeletion (0x7) instead of
       // kTypeRangeDeletion (0xF), so the range tombstone comes before the
       // Seek() key in InternalKey's ordering. So Seek() will look in the
-      // next file for the user key.
-      largest_candidate =
-          InternalKey(*upper_bound, kMaxSequenceNumber, kTypeRangeDeletion);
+      // next file for the user key
+      if (ts_sz) {
+        static constexpr char kTsMax[] = "\xff\xff\xff\xff\xff\xff\xff\xff\xff";
+        if (ts_sz <= strlen(kTsMax)) {
+          largest_candidate =
+              InternalKey(*upper_bound, kMaxSequenceNumber, kTypeRangeDeletion,
+                          Slice(kTsMax, ts_sz));
+        } else {
+          largest_candidate =
+              InternalKey(*upper_bound, kMaxSequenceNumber, kTypeRangeDeletion,
+                          std::string(ts_sz, '\xff'));
+        }
+      } else {
+        largest_candidate =
+            InternalKey(*upper_bound, kMaxSequenceNumber, kTypeRangeDeletion);
+      }
     }
 #ifndef NDEBUG
     SequenceNumber smallest_ikey_seqnum = kMaxSequenceNumber;

db/compaction/compaction_outputs.h

@@ -168,11 +168,16 @@ class CompactionOutputs {
   }
 
   // Add range-dels from the aggregator to the current output file
-  Status AddRangeDels(const Slice* comp_start, const Slice* comp_end,
+  // @param comp_start_user_key and comp_end_user_key include timestamp if
+  // user-defined timestamp is enabled.
+  // @param full_history_ts_low used for range tombstone garbage collection.
+  Status AddRangeDels(const Slice* comp_start_user_key,
+                      const Slice* comp_end_user_key,
                       CompactionIterationStats& range_del_out_stats,
                       bool bottommost_level, const InternalKeyComparator& icmp,
                       SequenceNumber earliest_snapshot,
-                      const Slice& next_table_min_key);
+                      const Slice& next_table_min_key,
+                      const std::string& full_history_ts_low);
 
   // if the outputs have range delete, range delete is also data
   bool HasRangeDel() const {

db/db_impl/db_impl.h

@@ -222,6 +222,9 @@ class DBImpl : public DB {
   Status DeleteRange(const WriteOptions& options,
                      ColumnFamilyHandle* column_family, const Slice& begin_key,
                      const Slice& end_key) override;
+  Status DeleteRange(const WriteOptions& options,
+                     ColumnFamilyHandle* column_family, const Slice& begin_key,
+                     const Slice& end_key, const Slice& ts) override;
 
   using DB::Write;
   virtual Status Write(const WriteOptions& options,

db/db_impl/db_impl_write.cc

@@ -111,6 +111,17 @@ Status DBImpl::DeleteRange(const WriteOptions& write_options,
   return DB::DeleteRange(write_options, column_family, begin_key, end_key);
 }
 
+Status DBImpl::DeleteRange(const WriteOptions& write_options,
+                           ColumnFamilyHandle* column_family,
+                           const Slice& begin_key, const Slice& end_key,
+                           const Slice& ts) {
+  const Status s = FailIfTsMismatchCf(column_family, ts, /*ts_for_read=*/false);
+  if (!s.ok()) {
+    return s;
+  }
+  return DB::DeleteRange(write_options, column_family, begin_key, end_key, ts);
+}
+
 void DBImpl::SetRecoverableStatePreReleaseCallback(
     PreReleaseCallback* callback) {
   recoverable_state_pre_release_callback_.reset(callback);
@@ -2361,6 +2372,24 @@ Status DB::DeleteRange(const WriteOptions& opt,
   return Write(opt, &batch);
 }
 
+Status DB::DeleteRange(const WriteOptions& opt,
+                       ColumnFamilyHandle* column_family,
+                       const Slice& begin_key, const Slice& end_key,
+                       const Slice& ts) {
+  ColumnFamilyHandle* default_cf = DefaultColumnFamily();
+  assert(default_cf);
+  const Comparator* const default_cf_ucmp = default_cf->GetComparator();
+  assert(default_cf_ucmp);
+  WriteBatch batch(0 /* reserved_bytes */, 0 /* max_bytes */,
+                   opt.protection_bytes_per_key,
+                   default_cf_ucmp->timestamp_size());
+  Status s = batch.DeleteRange(column_family, begin_key, end_key, ts);
+  if (!s.ok()) {
+    return s;
+  }
+  return Write(opt, &batch);
+}
+
 Status DB::Merge(const WriteOptions& opt, ColumnFamilyHandle* column_family,
                  const Slice& key, const Slice& value) {
   WriteBatch batch(0 /* reserved_bytes */, 0 /* max_bytes */,

db/db_range_del_test.cc

@@ -13,6 +13,7 @@
 
 namespace ROCKSDB_NAMESPACE {
 
+// TODO(cbi): parameterize the test to cover user-defined timestamp cases
 class DBRangeDelTest : public DBTestBase {
  public:
   DBRangeDelTest() : DBTestBase("db_range_del_test", /*env_do_fsync=*/false) {}