crimson/os/seastore: move overwritten delta extents to delta_overwrite_list if needed...

In RBM, Data extents like OBJECT_DATA_BLOCK can be overwritten.
So, this commit overwrites the (delta) extents in Journal to the corresponding  blocks in RBM
if they are Data extents.

Signed-off-by: Myoungwon Oh <myoungwon.oh@samsung.com>

diff --git a/src/crimson/os/seastore/cached_extent.h b/src/crimson/os/seastore/cached_extent.h
index 5fcb8142527d9c872e9e5fbe57ce7b093f350e6f..8bd4a269385118d6410b9c15481cf96aafd5ea9e 100644 (file)
--- a/src/crimson/os/seastore/cached_extent.h
+++ b/src/crimson/os/seastore/cached_extent.h
@@ -587,6 +587,11 @@ public:
     rewrite_generation = gen;
   }
 
+  void set_inplace_rewrite_generation() {
+    user_hint = placement_hint_t::REWRITE;
+    rewrite_generation = OOL_GENERATION;
+  }
+
   bool is_inline() const {
     return poffset.is_relative();
   }

diff --git a/src/crimson/os/seastore/extent_placement_manager.h b/src/crimson/os/seastore/extent_placement_manager.h
index b94c03ec34adabdaf026ed739697fc9fe05adcf5..371b49e07414766b96fb706b036d8362c5eefeae 100644 (file)
--- a/src/crimson/os/seastore/extent_placement_manager.h
+++ b/src/crimson/os/seastore/extent_placement_manager.h
@@ -43,6 +43,9 @@ public:
 
   using close_ertr = base_ertr;
   virtual close_ertr::future<> close() = 0;
+
+  virtual bool can_inplace_rewrite(Transaction& t,
+    CachedExtentRef extent) = 0;
 };
 using ExtentOolWriterRef = std::unique_ptr<ExtentOolWriter>;
 
@@ -79,6 +82,11 @@ public:
     return make_delayed_temp_paddr(0);
   }
 
+  bool can_inplace_rewrite(Transaction& t,
+    CachedExtentRef extent) final {
+    return false;
+  }
+
 private:
   alloc_write_iertr::future<> do_write(
     Transaction& t,
@@ -122,6 +130,17 @@ public:
     return rb_cleaner->alloc_paddr(length);
   }
 
+  bool can_inplace_rewrite(Transaction& t,
+    CachedExtentRef extent) final {
+    if (!extent->is_dirty()) {
+      return false;
+    }
+    assert(t.get_src() == transaction_type_t::TRIM_DIRTY);
+    ceph_assert_always(extent->get_type() == extent_types_t::ROOT ||
+       extent->get_paddr().is_absolute());
+    return crimson::os::seastore::can_inplace_rewrite(extent->get_type());
+  }
+
 private:
   alloc_write_iertr::future<> do_write(
     Transaction& t,
@@ -199,6 +218,14 @@ public:
     background_process.set_extent_callback(cb);
   }
 
+  bool can_inplace_rewrite(Transaction& t, CachedExtentRef extent) {
+    auto writer = get_writer(placement_hint_t::REWRITE,
+      get_extent_category(extent->get_type()),
+      OOL_GENERATION);
+    ceph_assert(writer);
+    return writer->can_inplace_rewrite(t, extent);
+  }
+
   journal_type_t get_journal_type() const {
     return background_process.get_journal_type();
   }

diff --git a/src/crimson/os/seastore/seastore_types.cc b/src/crimson/os/seastore/seastore_types.cc
index 0acfdb74ebb54d21d3c06abf217cea510f8c6210..93818ba02586459d8558b7cec5fc9ac0ee0fcf66 100644 (file)
--- a/src/crimson/os/seastore/seastore_types.cc
+++ b/src/crimson/os/seastore/seastore_types.cc
@@ -286,6 +286,10 @@ std::ostream &operator<<(std::ostream &out, data_category_t c)
   }
 }
 
+bool can_inplace_rewrite(extent_types_t type) {
+  return get_extent_category(type) == data_category_t::DATA;
+}
+
 std::ostream &operator<<(std::ostream &out, sea_time_point_printer_t tp)
 {
   if (tp.tp == NULL_TIME) {

diff --git a/src/crimson/os/seastore/seastore_types.h b/src/crimson/os/seastore/seastore_types.h
index 0b4ad853687fe650de0b1186fdc65efae331c6a9..f0e40fed082e21ac14e6c11d197c38fe06035ed0 100644 (file)
--- a/src/crimson/os/seastore/seastore_types.h
+++ b/src/crimson/os/seastore/seastore_types.h
@@ -1307,6 +1307,8 @@ constexpr data_category_t get_extent_category(extent_types_t type) {
   }
 }
 
+bool can_inplace_rewrite(extent_types_t type);
+
 // type for extent modification time, milliseconds since the epoch
 using sea_time_point = seastar::lowres_system_clock::time_point;
 using sea_duration = seastar::lowres_system_clock::duration;

diff --git a/src/crimson/os/seastore/transaction.h b/src/crimson/os/seastore/transaction.h
index d423196feba70963ce17ded8947e16e124495b3d..b43fb0d9cd8ee642b9574e66c2f757bd9dca787d 100644 (file)
--- a/src/crimson/os/seastore/transaction.h
+++ b/src/crimson/os/seastore/transaction.h
@@ -217,6 +217,14 @@ public:
     written_ool_block_list.push_back(ref);
   }
 
+  void add_inplace_rewrite_extent(CachedExtentRef ref) {
+   ceph_assert(!is_weak());
+   ceph_assert(ref);
+   ceph_assert(ref->get_paddr().is_absolute());
+   assert(ref->state == CachedExtent::extent_state_t::DIRTY);
+   pre_inplace_rewrite_list.emplace_back(ref->cast<LogicalCachedExtent>());
+  }
+
   void add_mutated_extent(CachedExtentRef ref) {
     ceph_assert(!is_weak());
     assert(ref->is_exist_mutation_pending() ||
@@ -278,6 +286,11 @@ public:
        ++num_allocated_invalid_extents;
       }
     }
+    for (auto& extent : pre_inplace_rewrite_list) {
+      if (extent->is_valid()) {
+       ret.push_back(extent);
+      } 
+    }
     return ret;
   }
 
@@ -388,6 +401,7 @@ public:
     inline_block_list.clear();
     written_ool_block_list.clear();
     pre_alloc_list.clear();
+    pre_inplace_rewrite_list.clear();
     retired_set.clear();
     existing_block_list.clear();
     existing_block_stats = {};
@@ -529,16 +543,18 @@ private:
   io_stat_t fresh_block_stats;
   uint64_t num_delayed_invalid_extents = 0;
   uint64_t num_allocated_invalid_extents = 0;
-  /// blocks that will be committed with journal record inline
-  std::list<CachedExtentRef> inline_block_list;
-  /// blocks that will be committed with out-of-line record
-  std::list<CachedExtentRef> written_ool_block_list;
-  /// blocks with delayed allocation, may become inline or ool above
+  /// fresh blocks with delayed allocation, may become inline or ool below
   std::list<LogicalCachedExtentRef> delayed_alloc_list;
-
-  /// Extents with pre-allocated addresses,
-  /// will be added to written_ool_block_list after write
+  /// fresh blocks with pre-allocated addresses with RBM,
+  /// should be released upon conflicts, will be added to ool below
   std::list<LogicalCachedExtentRef> pre_alloc_list;
+  /// dirty blocks for inplace rewrite with RBM, will be added to inplace ool below
+  std::list<LogicalCachedExtentRef> pre_inplace_rewrite_list;
+
+  /// fresh blocks that will be committed with inline journal record
+  std::list<CachedExtentRef> inline_block_list;
+  /// fresh blocks that will be committed with out-of-line record
+  std::list<CachedExtentRef> written_ool_block_list;
 
   /// list of mutated blocks, holds refcounts, subset of write_set
   std::list<CachedExtentRef> mutated_block_list;

diff --git a/src/crimson/os/seastore/transaction_manager.cc b/src/crimson/os/seastore/transaction_manager.cc
index 7261c84d2df735e976f28d000fc49b96af2cb669..e45224412057395f377ec95f8dfff48a03d24bab 100644 (file)
--- a/src/crimson/os/seastore/transaction_manager.cc
+++ b/src/crimson/os/seastore/transaction_manager.cc
@@ -487,6 +487,12 @@ TransactionManager::rewrite_extent_ret TransactionManager::rewrite_extent(
 
   assert(extent->is_valid() && !extent->is_initial_pending());
   if (extent->is_dirty()) {
+    if (epm->can_inplace_rewrite(t, extent)) {
+      DEBUGT("delta overwriting extent -- {}", t, *extent);
+      t.add_inplace_rewrite_extent(extent);
+      extent->set_inplace_rewrite_generation();
+      return rewrite_extent_iertr::now();
+    }
     extent->set_target_rewrite_generation(INIT_GENERATION);
   } else {
     extent->set_target_rewrite_generation(target_generation);