@nxtedition/rocksdb 15.4.1 → 15.5.0

package/deps/rocksdb/rocksdb/db/table_cache.cc

@@ -16,6 +16,7 @@
 #include "file/file_util.h"
 #include "file/filename.h"
 #include "file/random_access_file_reader.h"
+#include "monitoring/file_read_sample.h"
 #include "monitoring/perf_context_imp.h"
 #include "rocksdb/advanced_options.h"
 #include "rocksdb/statistics.h"
@@ -72,6 +73,8 @@ TableCache::TableCache(const ImmutableOptions& ioptions,
       file_options_(*file_options),
       cache_(cache),
       immortal_tables_(false),
+      should_pin_table_handles_(cache_.get()->GetCapacity() >=
+                                kInfiniteCapacity),
       block_cache_tracer_(block_cache_tracer),
       loader_mutex_(kLoadConcurency),
       io_tracer_(io_tracer),
@@ -98,6 +101,8 @@ Status TableCache::GetTableReader(
   std::unique_ptr<FSRandomAccessFile> file;
   FileOptions fopts = file_options;
   fopts.temperature = file_temperature;
+  fopts.file_checksum = file_meta.file_checksum;
+  fopts.file_checksum_func_name = file_meta.file_checksum_func_name;
   Status s = PrepareIOFromReadOptions(ro, ioptions_.clock, fopts.io_options);
   TEST_SYNC_POINT_CALLBACK("TableCache::GetTableReader:BeforeOpenFile",
                            const_cast<Status*>(&s));
@@ -113,8 +118,7 @@ Status TableCache::GetTableReader(
     Status temp_s =
         PrepareIOFromReadOptions(ro, ioptions_.clock, fopts.io_options);
     if (temp_s.ok()) {
-      temp_s = ioptions_.fs->NewRandomAccessFile(fname, file_options, &file,
-                                                 nullptr);
+      temp_s = ioptions_.fs->NewRandomAccessFile(fname, fopts, &file, nullptr);
     }
     if (temp_s.ok()) {
       RecordTick(ioptions_.stats, NO_FILE_OPENS);
@@ -168,15 +172,29 @@ Cache::Handle* TableCache::Lookup(Cache* cache, uint64_t file_number) {
   return cache->Lookup(key);
 }
 
+// TODO: consider making handle RAII.
 Status TableCache::FindTable(
     const ReadOptions& ro, const FileOptions& file_options,
     const InternalKeyComparator& internal_comparator,
     const FileMetaData& file_meta, TypedHandle** handle,
-    const MutableCFOptions& mutable_cf_options, const bool no_io,
-    HistogramImpl* file_read_hist, bool skip_filters, int level,
-    bool prefetch_index_and_filter_in_cache,
-    size_t max_file_size_for_l0_meta_pin, Temperature file_temperature) {
+    const MutableCFOptions& mutable_cf_options, TableReader** out_table_reader,
+    const bool no_io, HistogramImpl* file_read_hist, bool skip_filters,
+    int level, bool prefetch_index_and_filter_in_cache,
+    size_t max_file_size_for_l0_meta_pin, Temperature file_temperature,
+    bool pin_table_handle) {
+  assert(out_table_reader != nullptr && *out_table_reader == nullptr);
+  assert(handle != nullptr && *handle == nullptr);
   PERF_TIMER_GUARD_WITH_CLOCK(find_table_nanos, ioptions_.clock);
+
+  // Fast path: if table reader is already pinned, return it directly without a
+  // cache lookup.
+  auto pinned_reader = file_meta.fd.pinned_reader.Get();
+  if (pinned_reader != nullptr) {
+    *handle = nullptr;
+    *out_table_reader = pinned_reader;
+    return Status::OK();
+  }
+
   uint64_t number = file_meta.fd.GetNumber();
   // NOTE: sharing same Cache with BlobFileCache
   Slice key = GetSliceForFileNumber(&number);
@@ -184,39 +202,71 @@ Status TableCache::FindTable(
   TEST_SYNC_POINT_CALLBACK("TableCache::FindTable:0",
                            const_cast<bool*>(&no_io));
 
+  Status s = Status::OK();
   if (*handle == nullptr) {
     if (no_io) {
-      return Status::Incomplete("Table not found in table_cache, no_io is set");
+      s = Status::Incomplete("Table not found in table_cache, no_io is set");
+      return s;
     }
     MutexLock load_lock(&loader_mutex_.Get(key));
+
+    // Check if another thread has already pinned the table reader
+    pinned_reader = file_meta.fd.pinned_reader.Get();
+    if (pinned_reader != nullptr) {
+      *handle = nullptr;
+      *out_table_reader = pinned_reader;
+      return s;
+    }
+
     // We check the cache again under loading mutex
     *handle = cache_.Lookup(key);
-    if (*handle != nullptr) {
-      return Status::OK();
+    if (*handle == nullptr) {
+      std::unique_ptr<TableReader> table_reader;
+      s = GetTableReader(ro, file_options, internal_comparator, file_meta,
+                         false /* sequential mode */, file_read_hist,
+                         &table_reader, mutable_cf_options, skip_filters, level,
+                         prefetch_index_and_filter_in_cache,
+                         max_file_size_for_l0_meta_pin, file_temperature);
+      if (!s.ok()) {
+        assert(table_reader == nullptr);
+        RecordTick(ioptions_.stats, NO_FILE_ERRORS);
+        // We do not cache error results so that if the error is transient,
+        // or somebody repairs the file, we recover automatically.
+        IGNORE_STATUS_IF_ERROR(s);
+      } else {
+        s = cache_.Insert(key, table_reader.get(), 1, handle);
+        if (s.ok()) {
+          // Release ownership of table reader.
+          (void)table_reader.release();
+        }
+      }
     }
 
-    std::unique_ptr<TableReader> table_reader;
-    Status s = GetTableReader(ro, file_options, internal_comparator, file_meta,
-                              false /* sequential mode */, file_read_hist,
-                              &table_reader, mutable_cf_options, skip_filters,
-                              level, prefetch_index_and_filter_in_cache,
-                              max_file_size_for_l0_meta_pin, file_temperature);
-    if (!s.ok()) {
-      assert(table_reader == nullptr);
-      RecordTick(ioptions_.stats, NO_FILE_ERRORS);
-      // We do not cache error results so that if the error is transient,
-      // or somebody repairs the file, we recover automatically.
-      IGNORE_STATUS_IF_ERROR(s);
-    } else {
-      s = cache_.Insert(key, table_reader.get(), 1, handle);
-      if (s.ok()) {
-        // Release ownership of table reader.
-        table_reader.release();
+    if (s.ok()) {
+      *out_table_reader = cache_.Value(*handle);
+      if (pin_table_handle) {
+        file_meta.fd.pinned_reader.Pin(*handle, *out_table_reader);
+        *handle = nullptr;
       }
     }
-    return s;
+  } else {
+    *out_table_reader = cache_.Value(*handle);
+    if (pin_table_handle) {
+      // handle is in cache but not pinned. This should happen fairly rarely,
+      // and once the reader is pinned, we will no longer need to go through
+      // these mutexes again.
+      MutexLock load_lock(&loader_mutex_.Get(key));
+      if (file_meta.fd.pinned_reader.Get() != nullptr) {
+        // Another thread has pinned the handle; release our lookup ref.
+        cache_.Release(*handle);
+      } else {
+        file_meta.fd.pinned_reader.Pin(*handle, *out_table_reader);
+      }
+      *handle = nullptr;
+    }
   }
-  return Status::OK();
+
+  return s;
 }
 
 InternalIterator* TableCache::NewIterator(
@@ -229,7 +279,8 @@ InternalIterator* TableCache::NewIterator(
     const InternalKey* smallest_compaction_key,
     const InternalKey* largest_compaction_key, bool allow_unprepared_value,
     const SequenceNumber* read_seqno,
-    std::unique_ptr<TruncatedRangeDelIterator>* range_del_iter) {
+    std::unique_ptr<TruncatedRangeDelIterator>* range_del_iter,
+    bool maybe_pin_table_handle) {
   PERF_TIMER_GUARD(new_table_iterator_nanos);
 
   Status s;
@@ -239,19 +290,15 @@ InternalIterator* TableCache::NewIterator(
     *table_reader_ptr = nullptr;
   }
   bool for_compaction = caller == TableReaderCaller::kCompaction;
-  auto& fd = file_meta.fd;
-  table_reader = fd.table_reader;
-  if (table_reader == nullptr) {
-    s = FindTable(options, file_options, icomparator, file_meta, &handle,
-                  mutable_cf_options,
-                  options.read_tier == kBlockCacheTier /* no_io */,
-                  file_read_hist, skip_filters, level,
-                  true /* prefetch_index_and_filter_in_cache */,
-                  max_file_size_for_l0_meta_pin, file_meta.temperature);
-    if (s.ok()) {
-      table_reader = cache_.Value(handle);
-    }
-  }
+  TEST_SYNC_POINT_CALLBACK("TableCache::NewIterator::BeforeFindTable",
+                           const_cast<FileDescriptor*>(&file_meta.fd));
+  s = FindTable(options, file_options, icomparator, file_meta, &handle,
+                mutable_cf_options, &table_reader,
+                options.read_tier == kBlockCacheTier /* no_io */,
+                file_read_hist, skip_filters, level,
+                true /* prefetch_index_and_filter_in_cache */,
+                max_file_size_for_l0_meta_pin, file_meta.temperature,
+                maybe_pin_table_handle && should_pin_table_handles_);
   InternalIterator* result = nullptr;
   if (s.ok()) {
     if (options.table_filter &&
@@ -292,7 +339,7 @@ InternalIterator* TableCache::NewIterator(
       }
     }
     if (range_del_agg != nullptr) {
-      if (range_del_agg->AddFile(fd.GetNumber())) {
+      if (range_del_agg->AddFile(file_meta.fd.GetNumber())) {
         std::unique_ptr<FragmentedRangeTombstoneIterator> new_range_del_iter(
             static_cast<FragmentedRangeTombstoneIterator*>(
                 table_reader->NewRangeTombstoneIterator(options)));
@@ -331,17 +378,11 @@ Status TableCache::GetRangeTombstoneIterator(
     const FileMetaData& file_meta, const MutableCFOptions& mutable_cf_options,
     std::unique_ptr<FragmentedRangeTombstoneIterator>* out_iter) {
   assert(out_iter);
-  const FileDescriptor& fd = file_meta.fd;
   Status s;
-  TableReader* t = fd.table_reader;
+  TableReader* t = nullptr;
   TypedHandle* handle = nullptr;
-  if (t == nullptr) {
-    s = FindTable(options, file_options_, internal_comparator, file_meta,
-                  &handle, mutable_cf_options);
-    if (s.ok()) {
-      t = cache_.Value(handle);
-    }
-  }
+  s = FindTable(options, file_options_, internal_comparator, file_meta, &handle,
+                mutable_cf_options, &t);
   if (s.ok()) {
     // Note: NewRangeTombstoneIterator could return nullptr
     out_iter->reset(t->NewRangeTombstoneIterator(options));
@@ -455,20 +496,18 @@ Status TableCache::Get(const ReadOptions& options,
       row_cache_entry = &row_cache_entry_buffer;
     }
   }
-  TableReader* t = fd.table_reader;
+  TEST_SYNC_POINT_CALLBACK("TableCache::Get::BeforeFindTable",
+                           const_cast<FileDescriptor*>(&fd));
+  TableReader* t = nullptr;
   TypedHandle* handle = nullptr;
   if (s.ok() && !done) {
-    if (t == nullptr) {
-      s = FindTable(options, file_options_, internal_comparator, file_meta,
-                    &handle, mutable_cf_options,
-                    options.read_tier == kBlockCacheTier /* no_io */,
-                    file_read_hist, skip_filters, level,
-                    true /* prefetch_index_and_filter_in_cache */,
-                    max_file_size_for_l0_meta_pin, file_meta.temperature);
-      if (s.ok()) {
-        t = cache_.Value(handle);
-      }
-    }
+    s = FindTable(options, file_options_, internal_comparator, file_meta,
+                  &handle, mutable_cf_options, &t,
+                  options.read_tier == kBlockCacheTier /* no_io */,
+                  file_read_hist, skip_filters, level,
+                  true /* prefetch_index_and_filter_in_cache */,
+                  max_file_size_for_l0_meta_pin, file_meta.temperature,
+                  should_pin_table_handles_);
     SequenceNumber* max_covering_tombstone_seq =
         get_context->max_covering_tombstone_seq();
     if (s.ok() && max_covering_tombstone_seq != nullptr &&
@@ -545,8 +584,8 @@ Status TableCache::MultiGetFilter(
     const InternalKeyComparator& internal_comparator,
     const FileMetaData& file_meta, const MutableCFOptions& mutable_cf_options,
     HistogramImpl* file_read_hist, int level,
-    MultiGetContext::Range* mget_range, TypedHandle** table_handle) {
-  auto& fd = file_meta.fd;
+    MultiGetContext::Range* mget_range, TypedHandle** handle) {
+  assert(*handle == nullptr);
   IterKey row_cache_key;
   std::string row_cache_entry_buffer;
 
@@ -558,23 +597,17 @@ Status TableCache::MultiGetFilter(
     return Status::NotSupported();
   }
   Status s;
-  TableReader* t = fd.table_reader;
-  TypedHandle* handle = nullptr;
+  TableReader* t = nullptr;
   MultiGetContext::Range tombstone_range(*mget_range, mget_range->begin(),
                                          mget_range->end());
-  if (t == nullptr) {
-    s = FindTable(options, file_options_, internal_comparator, file_meta,
-                  &handle, mutable_cf_options,
-                  options.read_tier == kBlockCacheTier /* no_io */,
-                  file_read_hist,
-                  /*skip_filters=*/false, level,
-                  true /* prefetch_index_and_filter_in_cache */,
-                  /*max_file_size_for_l0_meta_pin=*/0, file_meta.temperature);
-    if (s.ok()) {
-      t = cache_.Value(handle);
-    }
-    *table_handle = handle;
-  }
+  s = FindTable(options, file_options_, internal_comparator, file_meta, handle,
+                mutable_cf_options, &t,
+                options.read_tier == kBlockCacheTier /* no_io */,
+                file_read_hist,
+                /*skip_filters=*/false, level,
+                true /* prefetch_index_and_filter_in_cache */,
+                /*max_file_size_for_l0_meta_pin=*/0, file_meta.temperature,
+                should_pin_table_handles_);
   if (s.ok()) {
     s = t->MultiGetFilter(options, mutable_cf_options.prefix_extractor.get(),
                           mget_range);
@@ -585,9 +618,9 @@ Status TableCache::MultiGetFilter(
     // is, it may be called with fewer keys in the rangedue to filtering.
     UpdateRangeTombstoneSeqnums(options, t, tombstone_range);
   }
-  if (mget_range->empty() && handle) {
-    cache_.Release(handle);
-    *table_handle = nullptr;
+  if (mget_range->empty() && *handle) {
+    cache_.Release(*handle);
+    *handle = nullptr;
   }
 
   return s;
@@ -599,24 +632,18 @@ Status TableCache::GetTableProperties(
     const FileMetaData& file_meta,
     std::shared_ptr<const TableProperties>* properties,
     const MutableCFOptions& mutable_cf_options, bool no_io) {
-  auto table_reader = file_meta.fd.table_reader;
-  // table already been pre-loaded?
-  if (table_reader) {
-    *properties = table_reader->GetTableProperties();
-
-    return Status::OK();
-  }
-
   TypedHandle* table_handle = nullptr;
-  Status s = FindTable(read_options, file_options, internal_comparator,
-                       file_meta, &table_handle, mutable_cf_options, no_io);
+  TableReader* table = nullptr;
+  Status s =
+      FindTable(read_options, file_options, internal_comparator, file_meta,
+                &table_handle, mutable_cf_options, &table, no_io);
   if (!s.ok()) {
     return s;
   }
-  assert(table_handle);
-  auto table = cache_.Value(table_handle);
   *properties = table->GetTableProperties();
-  cache_.Release(table_handle);
+  if (table_handle) {
+    cache_.Release(table_handle);
+  }
   return s;
 }
 
@@ -626,15 +653,10 @@ Status TableCache::ApproximateKeyAnchors(
 
     std::vector<TableReader::Anchor>& anchors) {
   Status s;
-  TableReader* t = file_meta.fd.table_reader;
+  TableReader* t = nullptr;
   TypedHandle* handle = nullptr;
-  if (t == nullptr) {
-    s = FindTable(ro, file_options_, internal_comparator, file_meta, &handle,
-                  mutable_cf_options);
-    if (s.ok()) {
-      t = cache_.Value(handle);
-    }
-  }
+  s = FindTable(ro, file_options_, internal_comparator, file_meta, &handle,
+                mutable_cf_options, &t);
   if (s.ok() && t != nullptr) {
     s = t->ApproximateKeyAnchors(ro, anchors);
   }
@@ -648,23 +670,18 @@ size_t TableCache::GetMemoryUsageByTableReader(
     const FileOptions& file_options, const ReadOptions& read_options,
     const InternalKeyComparator& internal_comparator,
     const FileMetaData& file_meta, const MutableCFOptions& mutable_cf_options) {
-  auto table_reader = file_meta.fd.table_reader;
-  // table already been pre-loaded?
-  if (table_reader) {
-    return table_reader->ApproximateMemoryUsage();
-  }
-
   TypedHandle* table_handle = nullptr;
+  TableReader* table = nullptr;
   Status s =
       FindTable(read_options, file_options, internal_comparator, file_meta,
-                &table_handle, mutable_cf_options, true /* no_io */);
+                &table_handle, mutable_cf_options, &table, true /* no_io */);
   if (!s.ok()) {
     return 0;
   }
-  assert(table_handle);
-  auto table = cache_.Value(table_handle);
   auto ret = table->ApproximateMemoryUsage();
-  cache_.Release(table_handle);
+  if (table_handle) {
+    cache_.Release(table_handle);
+  }
   return ret;
 }
 
@@ -678,18 +695,13 @@ uint64_t TableCache::ApproximateOffsetOf(
     const InternalKeyComparator& internal_comparator,
     const MutableCFOptions& mutable_cf_options) {
   uint64_t result = 0;
-  TableReader* table_reader = file_meta.fd.table_reader;
+  TableReader* table_reader = nullptr;
   TypedHandle* table_handle = nullptr;
-  if (table_reader == nullptr) {
-    Status s =
-        FindTable(read_options, file_options_, internal_comparator, file_meta,
-                  &table_handle, mutable_cf_options, false /* no_io */);
-    if (s.ok()) {
-      table_reader = cache_.Value(table_handle);
-    }
-  }
+  Status s =
+      FindTable(read_options, file_options_, internal_comparator, file_meta,
+                &table_handle, mutable_cf_options, &table_reader);
 
-  if (table_reader != nullptr) {
+  if (s.ok() && table_reader != nullptr) {
     result = table_reader->ApproximateOffsetOf(read_options, key, caller);
   }
   if (table_handle != nullptr) {
@@ -705,18 +717,13 @@ uint64_t TableCache::ApproximateSize(
     const InternalKeyComparator& internal_comparator,
     const MutableCFOptions& mutable_cf_options) {
   uint64_t result = 0;
-  TableReader* table_reader = file_meta.fd.table_reader;
+  TableReader* table_reader = nullptr;
   TypedHandle* table_handle = nullptr;
-  if (table_reader == nullptr) {
-    Status s =
-        FindTable(read_options, file_options_, internal_comparator, file_meta,
-                  &table_handle, mutable_cf_options, false /* no_io */);
-    if (s.ok()) {
-      table_reader = cache_.Value(table_handle);
-    }
-  }
+  Status s =
+      FindTable(read_options, file_options_, internal_comparator, file_meta,
+                &table_handle, mutable_cf_options, &table_reader);
 
-  if (table_reader != nullptr) {
+  if (s.ok() && table_reader != nullptr) {
     result = table_reader->ApproximateSize(read_options, start, end, caller);
   }
   if (table_handle != nullptr) {

package/deps/rocksdb/rocksdb/db/table_cache.h

@@ -99,7 +99,8 @@ class TableCache {
       const InternalKey* smallest_compaction_key,
       const InternalKey* largest_compaction_key, bool allow_unprepared_value,
       const SequenceNumber* range_del_read_seqno = nullptr,
-      std::unique_ptr<TruncatedRangeDelIterator>* range_del_iter = nullptr);
+      std::unique_ptr<TruncatedRangeDelIterator>* range_del_iter = nullptr,
+      bool maybe_pin_table_handle = false);
 
   // If a seek to internal key "k" in specified file finds an entry,
   // call get_context->SaveValue() repeatedly until
@@ -172,19 +173,31 @@ class TableCache {
   // Return handle to an existing cache entry if there is one
   static Cache::Handle* Lookup(Cache* cache, uint64_t file_number);
 
-  // Find table reader
-  // @param skip_filters Disables loading/accessing the filter block
-  // @param level == -1 means not specified
+  // Look up the TableReader for the given file in the cache, or open the file
+  // and create a new TableReader if not cached. On success, sets *table_reader
+  // to point to the TableReader (owned by the cache) and *handle to the cache
+  // handle (caller must release via cache_.Release() unless pin_table_handle is
+  // true). If the table reader is already pinned on file_meta, returns it
+  // directly without a cache lookup.
+  //
+  // @param no_io If true, returns Status::Incomplete() when the table is not
+  //              already in cache rather than reading from disk.
+  // @param skip_filters Disables loading/accessing the filter block.
+  // @param level The LSM level of this table, -1 if not specified.
+  // @param pin_table_handle If true, pins the table reader on file_meta so
+  //              future lookups bypass the cache. *handle is set to nullptr
+  //              on return in this case.
   Status FindTable(const ReadOptions& ro, const FileOptions& toptions,
                    const InternalKeyComparator& internal_comparator,
                    const FileMetaData& file_meta, TypedHandle**,
                    const MutableCFOptions& mutable_cf_options,
-                   const bool no_io = false,
+                   TableReader** table_reader, const bool no_io = false,
                    HistogramImpl* file_read_hist = nullptr,
                    bool skip_filters = false, int level = -1,
                    bool prefetch_index_and_filter_in_cache = true,
                    size_t max_file_size_for_l0_meta_pin = 0,
-                   Temperature file_temperature = Temperature::kUnknown);
+                   Temperature file_temperature = Temperature::kUnknown,
+                   bool pin_table_handle = false);
 
   // Get the table properties of a given table.
   // @no_io: indicates if we should load table to the cache if it is not present
@@ -231,6 +244,8 @@ class TableCache {
 
   CacheInterface& get_cache() { return cache_; }
 
+  const FileOptions& file_options() const { return file_options_; }
+
   // Capacity of the backing Cache that indicates infinite TableCache capacity.
   // For example when max_open_files is -1 we set the backing Cache to this.
   static const int kInfiniteCapacity = 0x400000;
@@ -243,6 +258,14 @@ class TableCache {
     }
   }
 
+  // Re-evaluates should_pin_table_handles_ from the current cache capacity.
+  // Must be called after the underlying cache capacity changes (e.g. via
+  // SetDBOptions changing max_open_files).
+  void UpdateShouldPinTableHandles() {
+    should_pin_table_handles_ =
+        cache_.get()->GetCapacity() >= kInfiniteCapacity;
+  }
+
  private:
   // Build a table reader
   Status GetTableReader(const ReadOptions& ro, const FileOptions& file_options,
@@ -283,6 +306,7 @@ class TableCache {
   CacheInterface cache_;
   std::string row_cache_id_;
   bool immortal_tables_;
+  bool should_pin_table_handles_;
   BlockCacheTracer* const block_cache_tracer_;
   Striped<CacheAlignedWrapper<port::Mutex>> loader_mutex_;
   std::shared_ptr<IOTracer> io_tracer_;

package/deps/rocksdb/rocksdb/db/table_cache_sync_and_async.h

@@ -21,7 +21,9 @@ DEFINE_SYNC_AND_ASYNC(Status, TableCache::MultiGet)
  bool skip_filters, bool skip_range_deletions, int level, TypedHandle* handle) {
   auto& fd = file_meta.fd;
   Status s;
-  TableReader* t = fd.table_reader;
+  TEST_SYNC_POINT_CALLBACK("TableCache::MultiGet::BeforeFindTable",
+                           const_cast<FileDescriptor*>(&fd));
+  TableReader* t = fd.pinned_reader.Get();
   MultiGetRange table_range(*mget_range, mget_range->begin(),
                             mget_range->end());
   if (handle != nullptr && t == nullptr) {
@@ -70,16 +72,14 @@ DEFINE_SYNC_AND_ASYNC(Status, TableCache::MultiGet)
     if (t == nullptr) {
       assert(handle == nullptr);
       s = FindTable(options, file_options_, internal_comparator, file_meta,
-                    &handle, mutable_cf_options,
+                    &handle, mutable_cf_options, &t,
                     options.read_tier == kBlockCacheTier /* no_io */,
                     file_read_hist, skip_filters, level,
                     true /* prefetch_index_and_filter_in_cache */,
-                    0 /*max_file_size_for_l0_meta_pin*/, file_meta.temperature);
+                    0 /*max_file_size_for_l0_meta_pin*/, file_meta.temperature,
+                    should_pin_table_handles_);
       TEST_SYNC_POINT_CALLBACK("TableCache::MultiGet:FindTable", &s);
-      if (s.ok()) {
-        t = cache_.Value(handle);
-        assert(t);
-      }
+      assert(!s.ok() || t);
     }
     if (s.ok() && !options.ignore_range_deletions && !skip_range_deletions) {
       UpdateRangeTombstoneSeqnums(options, t, table_range);

package/deps/rocksdb/rocksdb/db/version_builder.cc

@@ -32,8 +32,10 @@
 #include "db/version_edit.h"
 #include "db/version_edit_handler.h"
 #include "db/version_set.h"
+#include "db/version_util.h"
 #include "port/port.h"
 #include "table/table_reader.h"
+#include "test_util/sync_point.h"
 #include "util/string_util.h"
 
 namespace ROCKSDB_NAMESPACE {
@@ -430,12 +432,11 @@ class VersionBuilder::Rep {
   void UnrefFile(FileMetaData* f) {
     f->refs--;
     if (f->refs <= 0) {
-      if (f->table_reader_handle) {
+      if (f->fd.pinned_reader.Get() != nullptr) {
         assert(table_cache_ != nullptr);
         // NOTE: have to release in raw cache interface to avoid using a
-        // TypedHandle for FileMetaData::table_reader_handle
-        table_cache_->get_cache().get()->Release(f->table_reader_handle);
-        f->table_reader_handle = nullptr;
+        // TypedHandle for PinnedTableReader's internal handle
+        f->fd.pinned_reader.Release(table_cache_->get_cache().get());
       }
 
       if (file_metadata_cache_res_mgr_) {
@@ -1680,7 +1681,6 @@ class VersionBuilder::Rep {
 
     // <file metadata, level>
     std::vector<std::pair<FileMetaData*, int>> files_meta;
-    std::vector<Status> statuses;
     for (int level = 0; level < num_levels_; level++) {
       for (auto& file_meta_pair : levels_[level].added_files) {
         auto* file_meta = file_meta_pair.second;
@@ -1690,9 +1690,8 @@ class VersionBuilder::Rep {
           continue;
         }
         // If the file has been opened before, just skip it.
-        if (!file_meta->table_reader_handle) {
+        if (file_meta->fd.pinned_reader.Get() == nullptr) {
           files_meta.emplace_back(file_meta, level);
-          statuses.emplace_back(Status::OK());
         }
         if (files_meta.size() >= max_load) {
           break;
@@ -1703,49 +1702,11 @@ class VersionBuilder::Rep {
       }
     }
 
-    std::atomic<size_t> next_file_meta_idx(0);
-    std::function<void()> load_handlers_func([&]() {
-      while (true) {
-        size_t file_idx = next_file_meta_idx.fetch_add(1);
-        if (file_idx >= files_meta.size()) {
-          break;
-        }
-
-        auto* file_meta = files_meta[file_idx].first;
-        int level = files_meta[file_idx].second;
-        TableCache::TypedHandle* handle = nullptr;
-        statuses[file_idx] = table_cache_->FindTable(
-            read_options, file_options_,
-            *(base_vstorage_->InternalComparator()), *file_meta, &handle,
-            mutable_cf_options, false /*no_io */,
-            internal_stats->GetFileReadHist(level), false, level,
-            prefetch_index_and_filter_in_cache, max_file_size_for_l0_meta_pin,
-            file_meta->temperature);
-        if (handle != nullptr) {
-          file_meta->table_reader_handle = handle;
-          // Load table_reader
-          file_meta->fd.table_reader = table_cache_->get_cache().Value(handle);
-        }
-      }
-    });
-
-    std::vector<port::Thread> threads;
-    for (int i = 1; i < max_threads; i++) {
-      threads.emplace_back(load_handlers_func);
-    }
-    load_handlers_func();
-    for (auto& t : threads) {
-      t.join();
-    }
-    Status ret;
-    for (const auto& s : statuses) {
-      if (!s.ok()) {
-        if (ret.ok()) {
-          ret = s;
-        }
-      }
-    }
-    return ret;
+    return LoadTableHandlersHelper(
+        files_meta, table_cache_, file_options_,
+        *base_vstorage_->InternalComparator(), internal_stats, max_threads,
+        prefetch_index_and_filter_in_cache, mutable_cf_options,
+        max_file_size_for_l0_meta_pin, read_options);
   }
 };
 

package/deps/rocksdb/rocksdb/db/version_builder.h

@@ -53,7 +53,8 @@ class VersionBuilder {
   // Save the current Version to the provided `vstorage`.
   Status SaveTo(VersionStorageInfo* vstorage) const;
 
-  // Load all the table handlers for the current Version in the builder.
+  // Load table handlers for newly added files in the builder. This does not
+  // load any files in the base storage.
   Status LoadTableHandlers(InternalStats* internal_stats, int max_threads,
                            bool prefetch_index_and_filter_in_cache,
                            bool is_initial_load,