crimson/os/seastore: distinguish from commit-time and submit-time sequences

Also, the committed journal head should be updated atomically with the
rest of the complete-commit logics.

Signed-off-by: Yingxin Cheng <yingxin.cheng@intel.com>

diff --git a/src/crimson/os/seastore/async_cleaner.cc b/src/crimson/os/seastore/async_cleaner.cc
index 67690e91f80993608aa599ea5d691e5e4519e2f0..970ddc644afc2425df230a084b9fe305caca7833 100644 (file)
--- a/src/crimson/os/seastore/async_cleaner.cc
+++ b/src/crimson/os/seastore/async_cleaner.cc
@@ -168,7 +168,7 @@ void segments_info_t::init_closed(
   ++num_closed;
   if (type == segment_type_t::JOURNAL) {
     // init_closed won't initialize journal_segment_id
-    ceph_assert(get_journal_head() == JOURNAL_SEQ_NULL);
+    ceph_assert(get_submitted_journal_head() == JOURNAL_SEQ_NULL);
     ++num_type_journal;
   } else {
     ++num_type_ool;
@@ -603,11 +603,7 @@ segment_id_t AsyncCleaner::allocate_segment(
     if (segment_info.is_empty()) {
       auto old_usage = calc_utilization(seg_id);
       segments.mark_open(seg_id, seq, type, category, generation);
-      if (type == segment_type_t::JOURNAL) {
-        set_journal_head(segments.get_journal_head());
-      } else {
-        gc_process.maybe_wake_on_space_used();
-      }
+      gc_process.maybe_wake_on_space_used();
       auto new_usage = calc_utilization(seg_id);
       adjust_segment_util(old_usage, new_usage);
       INFO("opened, should_block_on_trim {}, should_block_on_reclaim {}, "

diff --git a/src/crimson/os/seastore/async_cleaner.h b/src/crimson/os/seastore/async_cleaner.h
index 4f6adbb540923f093f9f066404ca8730d2b1998f..440c21a1646db718f2003ee3d07970f0e22dbde8 100644 (file)
--- a/src/crimson/os/seastore/async_cleaner.h
+++ b/src/crimson/os/seastore/async_cleaner.h
@@ -179,7 +179,7 @@ public:
     return (double)get_available_bytes() / (double)total_bytes;
   }
 
-  journal_seq_t get_journal_head() const {
+  journal_seq_t get_submitted_journal_head() const {
     if (unlikely(journal_segment_id == NULL_SEG_ID)) {
       return JOURNAL_SEQ_NULL;
     }
@@ -269,16 +269,21 @@ private:
  */
 class SegmentProvider {
 public:
+  // set the committed journal head
   virtual void set_journal_head(journal_seq_t) = 0;
 
+  // get the committed journal tail
   journal_seq_t get_journal_tail() const {
     return std::min(get_alloc_tail(), get_dirty_tail());
   }
 
+  // get the committed journal dirty tail
   virtual journal_seq_t get_dirty_tail() const = 0;
 
+  // get the committed journal alloc tail
   virtual journal_seq_t get_alloc_tail() const = 0;
 
+  // set the committed journal tails
   virtual void update_journal_tails(
       journal_seq_t dirty_tail, journal_seq_t alloc_tail) = 0;
 
@@ -289,6 +294,7 @@ public:
 
   virtual void close_segment(segment_id_t) = 0;
 
+  // set the submitted segment writes in order
   virtual void update_segment_avail_bytes(segment_type_t, paddr_t) = 0;
 
   virtual void update_modify_time(
@@ -729,7 +735,7 @@ private:
 
   journal_seq_t journal_dirty_tail;
 
-  /// head of journal
+  /// the committed journal head
   journal_seq_t journal_head;
 
   ExtentCallbackInterface *ecb = nullptr;
@@ -778,6 +784,13 @@ public:
                 head >= journal_alloc_tail);
     ceph_assert(journal_dirty_tail == JOURNAL_SEQ_NULL ||
                 head >= journal_dirty_tail);
+
+    if (head.offset.get_addr_type() == addr_types_t::SEGMENT) {
+      auto submitted_journal_head = segments.get_submitted_journal_head();
+      ceph_assert(submitted_journal_head != JOURNAL_SEQ_NULL &&
+                  head <= submitted_journal_head);
+    }
+
     journal_head = head;
     gc_process.maybe_wake_on_space_used();
   }
@@ -789,11 +802,7 @@ public:
 
   void update_segment_avail_bytes(segment_type_t type, paddr_t offset) final {
     segments.update_written_to(type, offset);
-    if (type == segment_type_t::JOURNAL) {
-      set_journal_head(segments.get_journal_head());
-    } else {
-      gc_process.maybe_wake_on_space_used();
-    }
+    gc_process.maybe_wake_on_space_used();
   }
 
   void update_modify_time(
@@ -1228,12 +1237,16 @@ private:
    */
   bool should_block_on_trim() const {
     if (disable_trim) return false;
+    if (!init_complete) {
+      return false;
+    }
     return get_tail_limit() > get_journal_tail();
   }
 
   bool should_block_on_reclaim() const {
     if (disable_trim) return false;
-    if (get_segments_reclaimable() == 0) {
+    if (!init_complete ||
+        get_segments_reclaimable() == 0) {
       return false;
     }
     auto aratio = get_projected_available_ratio();

diff --git a/src/crimson/os/seastore/journal/segment_allocator.cc b/src/crimson/os/seastore/journal/segment_allocator.cc
index 81c937bfda984ee51a39a6a320301238848e18f2..b96032f9e8935f30b7ad56f0941b43d4199e2ffd 100644 (file)
--- a/src/crimson/os/seastore/journal/segment_allocator.cc
+++ b/src/crimson/os/seastore/journal/segment_allocator.cc
@@ -68,8 +68,6 @@ SegmentAllocator::do_open(bool is_mkfs)
         };
         dirty_tail = mkfs_seq;
         alloc_tail = mkfs_seq;
-        segment_provider.update_journal_tails(
-            dirty_tail, alloc_tail);
       } else {
         ceph_assert(dirty_tail != JOURNAL_SEQ_NULL);
         ceph_assert(alloc_tail != JOURNAL_SEQ_NULL);

diff --git a/src/crimson/os/seastore/transaction_manager.cc b/src/crimson/os/seastore/transaction_manager.cc
index b91acea653832ced4a40079587c262ae45f0e11b..2f62431ddaa9f66b7551719d01f2c236d8c2d205 100644 (file)
--- a/src/crimson/os/seastore/transaction_manager.cc
+++ b/src/crimson/os/seastore/transaction_manager.cc
@@ -48,7 +48,9 @@ TransactionManager::mkfs_ertr::future<> TransactionManager::mkfs()
   return async_cleaner->mount(
   ).safe_then([this] {
     return journal->open_for_mkfs();
-  }).safe_then([this](auto) {
+  }).safe_then([this](auto start_seq) {
+    async_cleaner->update_journal_tails(start_seq, start_seq);
+    async_cleaner->set_journal_head(start_seq);
     return epm->open();
   }).safe_then([this, FNAME]() {
     return with_transaction_intr(
@@ -106,7 +108,8 @@ TransactionManager::mount_ertr::future<> TransactionManager::mount()
       });
   }).safe_then([this] {
     return journal->open_for_mount();
-  }).safe_then([this, FNAME](auto) {
+  }).safe_then([this, FNAME](auto start_seq) {
+    async_cleaner->set_journal_head(start_seq);
     return seastar::do_with(
       create_weak_transaction(
         Transaction::src_t::READ, "mount"),
@@ -372,6 +375,7 @@ TransactionManager::submit_transaction_direct(
     ).safe_then([this, FNAME, &tref](auto submit_result) mutable {
       SUBDEBUGT(seastore_t, "committed with {}", tref, submit_result);
       auto start_seq = submit_result.write_result.start_seq;
+      async_cleaner->set_journal_head(start_seq);
       cache->complete_commit(
           tref,
           submit_result.record_block_base,