os/bluestore: multi-tier support in BlueStore

port kiizawa's patch(ceph#18211) Signed-off-by: Yang Honggang <yanghonggang@umcloud.com>
yanghonggang · Sep 15, 2019 · e91dcb9 · e91dcb9
1 parent 5609e91
commit e91dcb9
Show file tree

Hide file tree

Showing 18 changed files with 558 additions and 40 deletions.
diff --git a/src/common/ceph_strings.cc b/src/common/ceph_strings.cc
@@ -239,6 +239,8 @@ const char *ceph_osd_alloc_hint_flag_name(int f)
 		return "compressible";
 	case CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE:
 		return "incompressible";
+	case CEPH_OSD_ALLOC_HINT_FLAG_FAST_TIER:
+		return "fast_tier";
 	default:
 		return "???";
 	}

diff --git a/src/common/legacy_config_opts.h b/src/common/legacy_config_opts.h
@@ -1015,6 +1015,9 @@ OPTION(bluestore_block_db_create, OPT_BOOL)
 OPTION(bluestore_block_wal_path, OPT_STR)
 OPTION(bluestore_block_wal_size, OPT_U64) // rocksdb wal
 OPTION(bluestore_block_wal_create, OPT_BOOL)
+OPTION(bluestore_block_fast_path, OPT_STR)
+OPTION(bluestore_block_fast_size, OPT_U64) // fast tier
+OPTION(bluestore_block_fast_create, OPT_BOOL)
 OPTION(bluestore_block_preallocate_file, OPT_BOOL) //whether preallocate space if block/db_path/wal_path is file rather that block device.
 OPTION(bluestore_csum_type, OPT_STR) // none|xxhash32|xxhash64|crc32c|crc32c_16|crc32c_8
 OPTION(bluestore_csum_min_block, OPT_U32)

diff --git a/src/common/options.cc b/src/common/options.cc
@@ -3301,6 +3301,22 @@ std::vector<Option> get_global_options() {
     .add_see_also("bluestore_block_wal_path")
     .add_see_also("bluestore_block_wal_size"),
 
+    Option("bluestore_block_fast_path", Option::TYPE_STR, Option::LEVEL_DEV)
+    .set_default("")
+    .add_tag("mkfs")
+    .set_description("Path to fast block device/file"),
+
+    Option("bluestore_block_fast_size", Option::TYPE_UINT, Option::LEVEL_DEV)
+    .set_default(1_G)
+    .add_tag("mkfs")
+    .set_description("Size of file to create for backing bluestore fast tier"),
+
+    Option("bluestore_block_fast_create", Option::TYPE_BOOL, Option::LEVEL_DEV)
+    .set_default(true)
+    .add_tag("mkfs")
+    .set_description("Create bluestore_block_fast_path if it doesn't exist")
+    .add_see_also("bluestore_block_fast_path").add_see_also("bluestore_block_fast_size"),
+
     Option("bluestore_block_preallocate_file", Option::TYPE_BOOL, Option::LEVEL_DEV)
     .set_default(false)
     .add_tag("mkfs")

diff --git a/src/include/rados.h b/src/include/rados.h
@@ -525,6 +525,7 @@ enum {
 	CEPH_OSD_ALLOC_HINT_FLAG_LONGLIVED = 128,
 	CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE = 256,
 	CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE = 512,
+	CEPH_OSD_ALLOC_HINT_FLAG_FAST_TIER = 1024,
 };
 
 const char *ceph_osd_alloc_hint_flag_name(int f);

diff --git a/src/include/rados/librados.h b/src/include/rados/librados.h
@@ -150,6 +150,7 @@ enum {
   LIBRADOS_ALLOC_HINT_FLAG_LONGLIVED = 128,
   LIBRADOS_ALLOC_HINT_FLAG_COMPRESSIBLE = 256,
   LIBRADOS_ALLOC_HINT_FLAG_INCOMPRESSIBLE = 512,
+  LIBRADOS_ALLOC_HINT_FLAG_FAST_TIER = 1024,
 };
 /** @} */
 

diff --git a/src/include/rados/librados.hpp b/src/include/rados/librados.hpp
@@ -284,6 +284,7 @@ namespace librados
     ALLOC_HINT_FLAG_LONGLIVED = 128,
     ALLOC_HINT_FLAG_COMPRESSIBLE = 256,
     ALLOC_HINT_FLAG_INCOMPRESSIBLE = 512,
+    ALLOC_HINT_FLAG_FAST_TIER = 1024,
   };
 
   /*
@@ -1062,6 +1063,10 @@ namespace librados
     int aio_rmxattr(const std::string& oid, AioCompletion *c, const char *name);
     int aio_stat(const std::string& oid, AioCompletion *c, uint64_t *psize, time_t *pmtime);
     int aio_stat2(const std::string& oid, AioCompletion *c, uint64_t *psize, struct timespec *pts);
+    int aio_set_alloc_hint(const std::string& oid, AioCompletion *c,
+                           uint64_t expected_object_size,
+                           uint64_t expected_write_size,
+                           uint32_t flags);
 
     /**
      * Cancel aio operation

diff --git a/src/librados/IoCtxImpl.cc b/src/librados/IoCtxImpl.cc
@@ -1197,6 +1197,17 @@ int librados::IoCtxImpl::aio_setxattr(const object_t& oid, AioCompletionImpl *c,
   return aio_operate(oid, &op, c, snapc, 0);
 }
 
+int librados::IoCtxImpl::aio_set_alloc_hint(const object_t& oid, AioCompletionImpl *c,
+                                            uint64_t expected_object_size,
+                                            uint64_t expected_write_size,
+                                            uint32_t flags)
+{
+  ::ObjectOperation wr;
+  prepare_assert_ops(&wr);
+  wr.set_alloc_hint(expected_object_size, expected_write_size, flags);
+  return aio_operate(oid, &wr, c, snapc, 0);
+}
+
 namespace {
 struct AioGetxattrsData {
   AioGetxattrsData(librados::AioCompletionImpl *c, map<string, bufferlist>* attrset,

diff --git a/src/librados/IoCtxImpl.h b/src/librados/IoCtxImpl.h
@@ -224,6 +224,10 @@ struct librados::IoCtxImpl {
 		    map<string, bufferlist>& attrset);
   int aio_rmxattr(const object_t& oid, AioCompletionImpl *c,
 		  const char *name);
+  int aio_set_alloc_hint(const object_t& oid, AioCompletionImpl *c,
+                         uint64_t expected_object_size,
+                         uint64_t expected_write_size,
+                         uint32_t flags);
   int aio_cancel(AioCompletionImpl *c);
 
   int pool_change_auid(unsigned long long auid);

diff --git a/src/librados/librados.cc b/src/librados/librados.cc
@@ -2005,6 +2005,15 @@ int librados::IoCtx::aio_rmxattr(const std::string& oid, AioCompletion *c,
   return io_ctx_impl->aio_rmxattr(obj, c->pc, name);
 }
 
+int librados::IoCtx::aio_set_alloc_hint(const std::string& oid, AioCompletion *c,
+                                        uint64_t expected_object_size,
+                                        uint64_t expected_write_size,
+                                        uint32_t flags)
+{
+  object_t obj(oid);
+  return io_ctx_impl->aio_set_alloc_hint(obj, c->pc, expected_object_size, expected_write_size, flags);
+}
+
 int librados::IoCtx::aio_stat(const std::string& oid, librados::AioCompletion *c,
 			      uint64_t *psize, time_t *pmtime)
 {