os/bluestore: enable async manual compactions

Signed-off-by: Igor Fedotov <igor.fedotov@croit.io>

diff --git a/src/common/options/global.yaml.in b/src/common/options/global.yaml.in
index 52d306e4c11ca745464c35207e9e6fead97ba622..440aa3454d52bcfc0e460897e2176cc35520d4f0 100644 (file)
--- a/src/common/options/global.yaml.in
+++ b/src/common/options/global.yaml.in
@@ -5030,6 +5030,13 @@ options:
     This setting is used only when OSD is doing ``--mkfs``.
     Next runs of OSD retrieve sharding from disk.
   default: m(3) p(3,0-12) O(3,0-13)=block_cache={type=binned_lru} L=min_write_buffer_number_to_merge=32 P=min_write_buffer_number_to_merge=32
+- name: bluestore_async_db_compaction
+  type: bool
+  level: dev
+  desc: Perform DB compaction requests asynchronously
+  long_desc: 'How to perform DB compactions triggered either through async socket or
+    by OSD initialization procedure on start.'
+  default: true
 - name: bluestore_qfsck_on_mount
   type: bool
   level: dev

diff --git a/src/kv/RocksDBStore.cc b/src/kv/RocksDBStore.cc
index 28217118609ecaf2a3cdcd5b1e2ef077c299c44b..12723a8314d3ea19483e2292c1bcfa339596b7a7 100644 (file)
--- a/src/kv/RocksDBStore.cc
+++ b/src/kv/RocksDBStore.cc
@@ -1988,6 +1988,7 @@ int RocksDBStore::split_key(rocksdb::Slice in, string *prefix, string *key)
 
 void RocksDBStore::compact()
 {
+  dout(2) << __func__ << " starting" << dendl;
   logger->inc(l_rocksdb_compact);
   rocksdb::CompactRangeOptions options;
   db->CompactRange(options, default_cf, nullptr, nullptr);
@@ -1999,6 +2000,7 @@ void RocksDBStore::compact()
        nullptr, nullptr);
     }
   }
+  dout(2) << __func__ << " completed" << dendl;
 }
 
 void RocksDBStore::compact_thread_entry()

diff --git a/src/os/ObjectStore.h b/src/os/ObjectStore.h
index 3f9535337004f580d5aa3530287e17f552de9867..ddf48bfa89a4a52a8855b67881edfedb8cab5b9f 100644 (file)
--- a/src/os/ObjectStore.h
+++ b/src/os/ObjectStore.h
@@ -791,7 +791,7 @@ public:
   virtual void inject_data_error(const ghobject_t &oid) {}
   virtual void inject_mdata_error(const ghobject_t &oid) {}
 
-  virtual void compact() {}
+  virtual int compact() { return -ENOTSUP; }
   virtual bool has_builtin_csum() const {
     return false;
   }

diff --git a/src/os/bluestore/BlueStore.cc b/src/os/bluestore/BlueStore.cc
index f8d51cd0e4c6978272c5e72ec09974b189a40504..8ded55be00760233c72ce80d7ca054b991974e1f 100644 (file)
--- a/src/os/bluestore/BlueStore.cc
+++ b/src/os/bluestore/BlueStore.cc
@@ -11221,6 +11221,22 @@ void BlueStore::inject_bluefs_file(std::string_view dir, std::string_view name,
   bluefs->close_writer(p_handle);
 }
 
+int BlueStore::compact()
+{
+  int r = 0;
+  ceph_assert(db);
+  if (cct->_conf.get_val<bool>("bluestore_async_db_compaction")) {
+    dout(1) << __func__ << " starting async.." << dendl;
+    db->compact_async();
+    r = -EINPROGRESS;
+  } else {
+    dout(1) << __func__ << " starting sync.." << dendl;
+    db->compact();
+    dout(1) << __func__ << " finished." << dendl;
+  }
+  return r;
+}
+
 void BlueStore::collect_metadata(map<string,string> *pm)
 {
   dout(10) << __func__ << dendl;

diff --git a/src/os/bluestore/BlueStore.h b/src/os/bluestore/BlueStore.h
index 2932a59b89b2be8968498458eb0e187cbd55c5d6..ee1577bfa81616dcbcdcb2ee0c34575793f7fcb6 100644 (file)
--- a/src/os/bluestore/BlueStore.h
+++ b/src/os/bluestore/BlueStore.h
@@ -3379,10 +3379,7 @@ public:
                          std::string_view name,
                          size_t new_size);
 
-  void compact() override {
-    ceph_assert(db);
-    db->compact();
-  }
+  int compact() override;
   bool has_builtin_csum() const override {
     return true;
   }

diff --git a/src/os/kstore/KStore.h b/src/os/kstore/KStore.h
index 5b275dd3e011ffdcdfea07f90373084c42e121f4..9a9d413c66a82ca20967307908994ba898566d2f 100644 (file)
--- a/src/os/kstore/KStore.h
+++ b/src/os/kstore/KStore.h
@@ -580,9 +580,10 @@ public:
     TrackedOpRef op = TrackedOpRef(),
     ThreadPool::TPHandle *handle = NULL) override;
 
-  void compact () override {
+  int compact () override {
     ceph_assert(db);
     db->compact();
+    return 0;
   }
   
 private:

diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index 85207f1731166e0fbefe67f2e4630b30528d332a..cee5401156144a3df2f7e684ba3199c1ca7ddc7b 100644 (file)
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -2976,15 +2976,22 @@ will start to track new ops received afterwards.";
   } else if (prefix == "compact") {
     dout(1) << "triggering manual compaction" << dendl;
     auto start = ceph::coarse_mono_clock::now();
-    store->compact();
-    auto end = ceph::coarse_mono_clock::now();
-    double duration = std::chrono::duration<double>(end-start).count();
-    dout(1) << "finished manual compaction in "
-            << duration
-            << " seconds" << dendl;
-    f->open_object_section("compact_result");
-    f->dump_float("elapsed_time", duration);
-    f->close_section();
+    int r = store->compact();
+    if (r == 0) {
+      auto end = ceph::coarse_mono_clock::now();
+      double duration = std::chrono::duration<double>(end-start).count();
+
+      dout(1) << "finished manual compaction in "
+              << duration
+              << " seconds" << dendl;
+      f->open_object_section("compact_result");
+      f->dump_float("elapsed_time", duration);
+      f->close_section();
+    } else  if ( r == -EINPROGRESS) {
+      dout(1) << "manual compaction is being executed asynchronously" << dendl;
+    } else {
+      derr << "error starting manual compaction:" << cpp_strerror(r) << dendl;
+    }
   } else if (prefix == "get_mapped_pools") {
     f->open_array_section("mapped_pools");
     set<int64_t> poollist = get_mapped_pools();
@@ -3900,7 +3907,7 @@ int OSD::init()
   dout(2) << "superblock: I am osd." << superblock.whoami << dendl;
 
   if (cct->_conf.get_val<bool>("osd_compact_on_start")) {
-    dout(2) << "compacting object store's omap" << dendl;
+    dout(2) << "compacting object store's DB" << dendl;
     store->compact();
   }