os/bluestore: avoid bluefs log compaction when doing cold_close

Signed-off-by: Igor Fedotov <ifedotov@suse.com>
(cherry picked from commit db6f3f66fcc28c3db2eba01611664676a820cb90)

diff --git a/src/os/bluestore/BlueFS.cc b/src/os/bluestore/BlueFS.cc
index d0e1dd6aaab57d67bfa21b42427f7ad47c43a45a..6f21823ecfa448c4010db14b2c358e5494f0aa55 100644 (file)
--- a/src/os/bluestore/BlueFS.cc
+++ b/src/os/bluestore/BlueFS.cc
@@ -692,11 +692,11 @@ int BlueFS::maybe_verify_layout(const bluefs_layout_t& layout) const
   return 0;
 }
 
-void BlueFS::umount()
+void BlueFS::umount(bool avoid_compact)
 {
   dout(1) << __func__ << dendl;
 
-  sync_metadata();
+  sync_metadata(avoid_compact);
 
   _close_writer(log_writer);
   log_writer = NULL;
@@ -3175,7 +3175,7 @@ int BlueFS::_preallocate(FileRef f, uint64_t off, uint64_t len)
   return 0;
 }
 
-void BlueFS::sync_metadata()
+void BlueFS::sync_metadata(bool avoid_compact)
 {
   std::unique_lock l(lock);
   if (log_t.empty() && dirty_files.empty()) {
@@ -3188,7 +3188,7 @@ void BlueFS::sync_metadata()
     dout(10) << __func__ << " done in " << (ceph_clock_now() - start) << dendl;
   }
 
-  if (_should_compact_log()) {
+  if (!avoid_compact && _should_compact_log()) {
     if (cct->_conf->bluefs_compact_log_sync) {
       _compact_log_sync();
     } else {

diff --git a/src/os/bluestore/BlueFS.h b/src/os/bluestore/BlueFS.h
index 3fb57373b8254ff9aa4a4749039d161c30dd9040..66cd259d45082e568ed8e94005862f2d1204c4da 100644 (file)
--- a/src/os/bluestore/BlueFS.h
+++ b/src/os/bluestore/BlueFS.h
@@ -443,7 +443,7 @@ public:
   int mkfs(uuid_d osd_uuid, const bluefs_layout_t& layout);
   int mount();
   int maybe_verify_layout(const bluefs_layout_t& layout) const;
-  void umount();
+  void umount(bool avoid_compact = false);
   int prepare_new_device(int id, const bluefs_layout_t& layout);
   
   int log_dump();
@@ -514,7 +514,7 @@ public:
   void compact_log();
 
   /// sync any uncommitted state to disk
-  void sync_metadata();
+  void sync_metadata(bool avoid_compact);
 
   void set_slow_device_expander(BlueFSDeviceExpander* a) {
     slow_dev_expander = a;

diff --git a/src/os/bluestore/BlueRocksEnv.cc b/src/os/bluestore/BlueRocksEnv.cc
index 7f2f8991b9e12e141c5c4d5c62f6f66aae99ec56..51614c09d2cd61a736b8e4eb4585e6f44543f86c 100644 (file)
--- a/src/os/bluestore/BlueRocksEnv.cc
+++ b/src/os/bluestore/BlueRocksEnv.cc
@@ -299,7 +299,7 @@ class BlueRocksDirectory : public rocksdb::Directory {
   // Fsync directory. Can be called concurrently from multiple threads.
   rocksdb::Status Fsync() override {
     // it is sufficient to flush the log.
-    fs->sync_metadata();
+    fs->sync_metadata(false);
     return rocksdb::Status::OK();
   }
 };

diff --git a/src/os/bluestore/BlueStore.cc b/src/os/bluestore/BlueStore.cc
index 0e634fb56becc5bf90b4aa956a8c5aaa64454834..cf1524e3f4dcbcb85556eb74a87bcfd8725415b4 100644 (file)
--- a/src/os/bluestore/BlueStore.cc
+++ b/src/os/bluestore/BlueStore.cc
@@ -5456,9 +5456,9 @@ int BlueStore::_open_bluefs(bool create)
   return r;
 }
 
-void BlueStore::_close_bluefs()
+void BlueStore::_close_bluefs(bool cold_close)
 {
-  bluefs->umount();
+  bluefs->umount(cold_close);
   _minimal_close_bluefs();
 }
 
@@ -5523,7 +5523,7 @@ int BlueStore::_open_db_and_around(bool read_only)
 
     // now open in R/W mode
     if (!read_only) {
-      _close_db();
+      _close_db(true);
 
       r = _open_db(false, false, false);
       if (r < 0) {
@@ -5555,18 +5555,18 @@ int BlueStore::_open_db_and_around(bool read_only)
  out_fm:
   _close_fm();
  out_db:
-  _close_db();
+  _close_db(read_only);
   return r;
 }
 
-void BlueStore::_close_db_and_around()
+void BlueStore::_close_db_and_around(bool read_only)
 {
   if (bluefs) {
-    if (out_of_sync_fm.fetch_and(0)) {
+    if (!read_only && out_of_sync_fm.fetch_and(0)) {
       _sync_bluefs_and_fm();
     }
-    _close_db();
-    while(out_of_sync_fm.fetch_and(0)) {
+    _close_db(read_only);
+    while(!read_only && out_of_sync_fm.fetch_and(0)) {
       // if seen some allocations during close - repeat open_db, sync fm, close
       dout(0) << __func__ << " syncing FreelistManager" << dendl;
       int r = _open_db(false, false, false);
@@ -5577,7 +5577,7 @@ void BlueStore::_close_db_and_around()
        break;
       }
       _sync_bluefs_and_fm();
-      _close_db();
+      _close_db(false);
     }
     if (!_kv_only) {
       _close_alloc();
@@ -5586,7 +5586,7 @@ void BlueStore::_close_db_and_around()
   } else {
     _close_alloc();
     _close_fm();
-    _close_db();
+    _close_db(read_only);
   }
 }
 
@@ -5754,7 +5754,7 @@ int BlueStore::_open_db(bool create, bool to_repair_db, bool read_only)
   if (!db) {
     derr << __func__ << " error creating db" << dendl;
     if (bluefs) {
-      _close_bluefs();
+      _close_bluefs(read_only);
     }
     // delete env manually here since we can't depend on db to do this
     // under this case
@@ -5799,7 +5799,7 @@ int BlueStore::_open_db(bool create, bool to_repair_db, bool read_only)
   }
   if (r) {
     derr << __func__ << " erroring opening db: " << err.str() << dendl;
-    _close_db();
+    _close_db(read_only);
     return -EIO;
   }
   dout(1) << __func__ << " opened " << kv_backend
@@ -5807,13 +5807,13 @@ int BlueStore::_open_db(bool create, bool to_repair_db, bool read_only)
   return 0;
 }
 
-void BlueStore::_close_db()
+void BlueStore::_close_db(bool cold_close)
 {
   ceph_assert(db);
   delete db;
   db = NULL;
   if (bluefs) {
-    _close_bluefs();
+    _close_bluefs(cold_close);
   }
 }
 
@@ -6444,7 +6444,7 @@ int BlueStore::mkfs()
  out_close_fm:
   _close_fm();
  out_close_db:
-  _close_db();
+  _close_db(false);
  out_close_bdev:
   _close_bdev();
  out_close_fsid:
@@ -6493,7 +6493,7 @@ int BlueStore::_mount_for_bluefs()
 
 void BlueStore::_umount_for_bluefs()
 {
-  _close_bluefs();
+  _close_bluefs(false);
   _close_fsid();
   _close_path();
 }
@@ -7032,7 +7032,7 @@ int BlueStore::_mount(bool kv_only, bool open_db)
  out_coll:
   _flush_cache();
  out_db:
-  _close_db_and_around();
+  _close_db_and_around(false);
  out_bdev:
   _close_bdev();
  out_fsid:
@@ -7058,7 +7058,7 @@ int BlueStore::umount()
     dout(20) << __func__ << " closing" << dendl;
 
   }
-  _close_db_and_around();
+  _close_db_and_around(false);
   _close_bdev();
   _close_fsid();
   _close_path();
@@ -7110,7 +7110,7 @@ int BlueStore::cold_open()
 }
 int BlueStore::cold_close()
 {
-  _close_db_and_around();
+  _close_db_and_around(true);
   _close_bdev();
   _close_fsid();
   _close_path();
@@ -8176,7 +8176,7 @@ out_scan:
   mempool_thread.shutdown();
   _flush_cache();
 out_db:
-  _close_db_and_around();
+  _close_db_and_around(false);
 out_bdev:
   _close_bdev();
 out_fsid:
@@ -8265,8 +8265,7 @@ int BlueStore::_fsck_on_open(BlueStore::FSCKDepth depth, bool repair)
         e.get_start(), e.get_len(), fm->get_alloc_size(), used_blocks,
         [&](uint64_t pos, mempool_dynamic_bitset &bs) {
           bs.set(pos);
-        }
-       );
+        });
     }
     int r = bluefs->fsck();
     if (r < 0) {

diff --git a/src/os/bluestore/BlueStore.h b/src/os/bluestore/BlueStore.h
index cf1e994465a82b02eed34f037e202372e15d6e44..e9b71ea0b93169f4a5e404b39b09f4dc9e1a3bf8 100644 (file)
--- a/src/os/bluestore/BlueStore.h
+++ b/src/os/bluestore/BlueStore.h
@@ -2243,7 +2243,7 @@ private:
   int _minimal_open_bluefs(bool create);
   void _minimal_close_bluefs();
   int _open_bluefs(bool create);
-  void _close_bluefs();
+  void _close_bluefs(bool cold_close);
 
   // Limited (u)mount intended for BlueFS operations only
   int _mount_for_bluefs();
@@ -2256,7 +2256,7 @@ private:
   * in the proper order
   */
   int _open_db_and_around(bool read_only);
-  void _close_db_and_around();
+  void _close_db_and_around(bool read_only);
 
   // updates legacy bluefs related recs in DB to a state valid for
   // downgrades from nautilus.
@@ -2269,7 +2269,7 @@ private:
   int _open_db(bool create,
               bool to_repair_db=false,
               bool read_only = false);
-  void _close_db();
+  void _close_db(bool read_only);
   int _open_fm(KeyValueDB::Transaction t);
   void _close_fm();
   int _open_alloc();

diff --git a/src/test/objectstore/test_bluefs.cc b/src/test/objectstore/test_bluefs.cc
index ef838ab124f2f1294f3b08da025973097b27c857..f6e47950b501f4535c770500ddcfc17b39579044 100644 (file)
--- a/src/test/objectstore/test_bluefs.cc
+++ b/src/test/objectstore/test_bluefs.cc
@@ -339,7 +339,7 @@ void sync_fs(BlueFS &fs)
     while (1) {
       if (writes_done == true)
         break;
-      fs.sync_metadata();
+      fs.sync_metadata(false);
       sleep(1);
     }
 }
@@ -502,10 +502,10 @@ TEST(BlueFS, test_simple_compaction_sync) {
           string file = "file.";
          file.append(to_string(j));
           fs.unlink(dir, file);
-         fs.sync_metadata();
+         fs.sync_metadata(false);
        }
        ASSERT_EQ(0, fs.rmdir(dir));
-       fs.sync_metadata();
+       fs.sync_metadata(false);
     }
   }
   fs.compact_log();
@@ -555,10 +555,10 @@ TEST(BlueFS, test_simple_compaction_async) {
           string file = "file.";
          file.append(to_string(j));
           fs.unlink(dir, file);
-         fs.sync_metadata();
+         fs.sync_metadata(false);
        }
        ASSERT_EQ(0, fs.rmdir(dir));
-       fs.sync_metadata();
+       fs.sync_metadata(false);
     }
   }
   fs.compact_log();