env: Add Zoned Namespace SSD support on TerarkDB (#129)

* env: Add Zoned Namespace SSD support on TerarkDB

Support zone namespace SSD by transplanting and modifying ZenFS
from Hans Holmberg of WesternDigital Inc.

https://github.com/westerndigitalcorporation/zenfs

Since TerarkDB is based on RocksDB v5.18.3, we also made some adoption
to the codebase.

In this version, the `zenfs` is not fully functional yet but prepared the basic function & utilities for next stage integration.

Signed-off-by: Changlong Chen <chenchanglong@bytedance.com>
Signed-off-by: Kuankuan Guo <guokuankuan@bytedance.com>

Co-authored-by: Hans Holmberg <hans.holmberg@wdc.com>
Co-authored-by: Yuanliang Wang <wangyuanliang@bytedance.com>
Co-authored-by: Changlong Chen <chenchanglong@bytedance.com>

.gitignore

@@ -29,6 +29,7 @@ build/
 
 ldb
 manifest_dump
+zenfs
 sst_dump
 blob_dump
 column_aware_encoding_exp

.gitmodules

@@ -19,3 +19,6 @@
 [submodule "third-party/gflags"]
 	path = third-party/gflags
 	url = https://github.com/gflags/gflags.git
+[submodule "third-party/zenfs"]
+	path = third-party/zenfs
+	url = https://github.com/bzbd/zenfs.git

CMakeLists.txt

@@ -41,6 +41,7 @@ include(CMakeDependentOption)
 CMAKE_DEPENDENT_OPTION(WITH_TESTS "build with tests" ON "CMAKE_BUILD_TYPE STREQUAL Debug" OFF)
 option(WITH_TOOLS "build with tools" OFF)
 option(WITH_TERARK_ZIP "build with TerarkZipTable support" ON)
+option(WITH_ZENFS "build with experimental zenfs" OFF)
 option(WITH_DIAGNOSE_CACHE "build with diagnosable cache support" OFF)
 option(WITH_BOOSTLIB "build with boost, if WITH_TERARK_ZIP is ON, this will also set to ON" OFF)
 option(WITH_TERARKDB_NAMESPACE "namespace" "terarkdb")
@@ -65,6 +66,10 @@ if (WITH_WINDOWS_UTF8_FILENAMES)
   add_definitions(-DROCKSDB_WINDOWS_UTF8_FILENAMES)
 endif()
 
+if(WITH_ZENFS)
+  add_compile_definitions(ROCKSDB_NAMESPACE=${WITH_TERARKDB_NAMESPACE})
+endif()
+
 IF(FORCE_TERARKDB_RELEASE_BUILD)
   SET(CMAKE_BUILD_TYPE "Release")
 ENDIF()
@@ -374,6 +379,10 @@ if(WITH_TBB)
   list(APPEND THIRDPARTY_LIBS ${TBB_LIBRARIES})
 endif()
 
+if(WITH_ZENFS)
+  list(APPEND THIRDPARTY_LIBS zbd)
+endif()
+
 # Stall notifications eat some performance from inserts
 option(DISABLE_STALL_NOTIF "Build with stall notifications" OFF)
 if(DISABLE_STALL_NOTIF)
@@ -827,6 +836,15 @@ if(WIN32)
       port/port_posix.cc
       env/env_posix.cc
       env/io_posix.cc)
+
+    if(WITH_ZENFS)
+      list(APPEND SOURCES
+        third-party/zenfs/fs/fs_zenfs.cc
+        third-party/zenfs/fs/io_zenfs.cc
+        third-party/zenfs/fs/zbd_zenfs.cc
+        third-party/zenfs/util/zenfs.cc
+        tools/zenfs_tool.cc)
+    endif()
 endif()
 
 set(ROCKSDB_STATIC_LIB terarkdb${ARTIFACT_SUFFIX})
@@ -919,7 +937,7 @@ if(NOT WIN32 OR ROCKSDB_INSTALL_ON_WINDOWS)
   # We should also expose terarkdb's source dir
   INSTALL(DIRECTORY memtable monitoring env db cache options port table util utilities
         DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}
-        FILES_MATCHING 
+        FILES_MATCHING
         PATTERN "*.h"
         PATTERN "*.hpp"
   )
@@ -1184,6 +1202,11 @@ if(WITH_TOOLS)
       $<TARGET_OBJECTS:testharness>)
   target_link_libraries(db_bench${ARTIFACT_SUFFIX} gtest ${ROCKSDB_STATIC_LIB})
 
+  add_executable(zenfs${ARTIFACT_SUFFIX} tools/zenfs.cc
+      tools/zenfs_tool.cc
+      $<TARGET_OBJECTS:testharness>)
+  target_link_libraries(zenfs${ARTIFACT_SUFFIX} gtest ${ROCKSDB_STATIC_LIB})
+
   if(WITH_TERARK_ZIP)
     add_subdirectory(terark-tools/terark-test)
   endif()

bench.py

@@ -48,6 +48,7 @@ def bench(records, key_size, value_size, engine, db_dir, exist_db):
                        """ % BENCH_ARGS[0]
     cmd = """
            {db_bench} \
+           --fs_uri=/dev/nvme3n2
            --benchmarks={bench_type}
 	   --use_existing_db={exist_db}
            --sync=1

build_tools/build_detect_platform

@@ -332,6 +332,19 @@ EOF
         fi
     fi
 
+    if ! test $ROCKSDB_DISABLE_LZBD; then
+        # Test whether libzbd is installed
+        $CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+          #include <libzbd/zbd.h>
+          int main() {}
+EOF
+        if [ "$?" = 0 ]; then
+            COMMON_FLAGS="$COMMON_FLAGS -DLIBZBD"
+            PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lzbd"
+            JAVA_LDFLAGS="$JAVA_LDFLAGS -lzbd"
+        fi
+    fi
+
     if ! test $ROCKSDB_DISABLE_NUMA; then
         # Test whether numa is available
         $CXX $CFLAGS -x c++ - -o /dev/null -lnuma 2>/dev/null  <<EOF

env/io_posix.cc

@@ -1115,4 +1115,41 @@ Status PosixDirectory::Fsync() {
   return Status::OK();
 }
 }  // namespace TERARKDB_NAMESPACE
+
+#ifdef ZENFS_READY
+static FactoryFunc<FileSystem> zenfs_filesystem_reg =
+    ObjectLibrary::Default()->Register<FileSystem>(
+        "zenfs://.*", [](const std::string& uri, std::unique_ptr<FileSystem>* f,
+                         std::string* errmsg) {
+          std::string devID = uri;
+          FileSystem* fs = nullptr;
+          Status s;
+
+          devID.replace(0, strlen("zenfs://"), "");
+          if (devID.rfind("dev:") == 0) {
+            devID.replace(0, strlen("dev:"), "");
+            s = NewZenFS(&fs, devID);
+            if (!s.ok()) {
+              *errmsg = s.ToString();
+            }
+          } else if (devID.rfind("uuid:") == 0) {
+            std::map<std::string, std::string> zenFileSystems =
+                ListZenFileSystems();
+            devID.replace(0, strlen("uuid:"), "");
+
+            if (zenFileSystems.find(devID) == zenFileSystems.end()) {
+              *errmsg = "UUID not found";
+            } else {
+              s = NewZenFS(&fs, zenFileSystems[devID]);
+              if (!s.ok()) {
+                *errmsg = s.ToString();
+              }
+            }
+          } else {
+            *errmsg = "Malformed URI";
+          }
+          f->reset(fs);
+          return f->get();
+        });
+#endif
 #endif

include/rocksdb/env.h

@@ -1423,4 +1423,8 @@ Env* NewTimedEnv(Env* base_env);
 // This is a env forwarding method defined in env/env_io_prof.cc
 Env* NewIOProfEnv(Env* base_env);
 
+// Returns a new environment that is used for ZENFS environment.
+// This is a factory method for ZENFS declared in hdfs/env_zenfs.h
+Status NewZenEnv(Env** env, const std::string& bdevname);
+
 }  // namespace TERARKDB_NAMESPACE

include/rocksdb/terark_namespace.h.in

@@ -11,3 +11,8 @@
 #cmakedefine WITH_TERARK_ZIP
 #cmakedefine WITH_BOOSTLIB
 #cmakedefine WITH_DIAGNOSE_CACHE
+#cmakedefine WITH_ZENFS
+
+#ifdef WITH_ZENFS
+#define LIBZBD
+#endif

src.mk

@@ -269,6 +269,9 @@ MOCK_LIB_SOURCES = \
 BENCH_LIB_SOURCES =                                             \
   tools/db_bench_tool.cc                                        \
 
+ZENFS_LIB_SOURCES =                                             \
+  tools/zenfs_tool.cc                                           \
+
 EXP_LIB_SOURCES = \
   utilities/col_buf_decoder.cc                                  \
   utilities/col_buf_encoder.cc                                  \

table/terark_zip_table.cc

@@ -488,11 +488,13 @@ Status TerarkZipTableFactory::SanitizeOptions(
   auto& tzto = *reinterpret_cast<const TerarkZipTableOptions*>(
       table_factory->GetOptions());
   try {
-    terark::TempFileDeleteOnClose test;
-    test.path = tzto.localTempDir + "/Terark-XXXXXX";
-    test.open_temp();
-    test.writer << "Terark";
-    test.complete_write();
+    if (tzto.terarkZipMinLevel != -2) {
+      terark::TempFileDeleteOnClose test;
+      test.path = tzto.localTempDir + "/Terark-XXXXXX";
+      test.open_temp();
+      test.writer << "Terark";
+      test.complete_write();
+    }
   } catch (...) {
     std::string msg = "ERROR: bad localTempDir : " + tzto.localTempDir;
     fprintf(stderr, "%s\n", msg.c_str());

tools/db_bench_tool.cc

@@ -82,6 +82,11 @@
 #include <io.h>  // open/close
 #endif
 
+#ifdef LIBZBD
+#include "env/env_zenfs.h"
+#include "env/zbd_zenfs.h"
+#endif
+
 using GFLAGS_NAMESPACE::ParseCommandLineFlags;
 using GFLAGS_NAMESPACE::RegisterFlagValidator;
 using GFLAGS_NAMESPACE::SetUsageMessage;
@@ -826,10 +831,14 @@ DEFINE_uint64(prepare_log_writer_num, 1, "");
 DEFINE_string(env_uri, "",
               "URI for registry Env lookup. Mutually exclusive"
               " with --hdfs.");
+DEFINE_string(fs_uri, "",
+              "URI for registry Filesystem lookup. Mutually exclusive"
+              " with --hdfs and --env_uri."
+              " Creates a default environment with the specified filesystem.");
 #endif  // ROCKSDB_LITE
 DEFINE_string(hdfs, "",
               "Name of hdfs environment. Mutually exclusive with"
-              " --env_uri.");
+              " --env_uri and --fs_uri");
 static TERARKDB_NAMESPACE::Env* FLAGS_env = TERARKDB_NAMESPACE::Env::Default();
 
 DEFINE_int64(stats_interval, 0,
@@ -5817,6 +5826,13 @@ int db_bench_tool(int argc, char** argv) {
       StringToCompressionType(FLAGS_compression_type.c_str());
 
 #ifndef ROCKSDB_LITE
+  int env_opts =
+      !FLAGS_hdfs.empty() + !FLAGS_env_uri.empty() + !FLAGS_fs_uri.empty();
+  if (env_opts > 1) {
+    fprintf(stderr,
+            "Error: --hdfs, --env_uri and --fs_uri are mutually exclusive\n");
+    exit(1);
+  }
   std::unique_ptr<Env> custom_env_guard;
   if (!FLAGS_hdfs.empty() && !FLAGS_env_uri.empty()) {
     fprintf(stderr, "Cannot provide both --hdfs and --env_uri.\n");
@@ -5827,6 +5843,14 @@ int db_bench_tool(int argc, char** argv) {
       fprintf(stderr, "No Env registered for URI: %s\n", FLAGS_env_uri.c_str());
       exit(1);
     }
+  } else if (!FLAGS_fs_uri.empty()) {
+#ifdef WITH_ZENFS
+    Status s = NewZenEnv(&FLAGS_env, FLAGS_fs_uri);
+    if (!s.ok()) {
+      fprintf(stderr, "Error: %s\n", s.ToString().c_str());
+      exit(1);
+    }
+#endif
   }
 #endif  // ROCKSDB_LITE
   if (!FLAGS_hdfs.empty()) {

tools/zenfs.cc

@@ -0,0 +1,32 @@
+// Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
+// Copyright (c) 2019-present, Western Digital Corporation
+// Copyright (c) 2021-present, Bytedance Inc.  All rights reserved.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#include <cstdio>
+#ifdef ROCKSDB_LITE
+int main() {
+  fprintf("Not supported in lite mode.\n");
+  return 1;
+}
+#else
+#ifdef GFLAGS
+#ifdef LIBZBD
+int zenfs_tool(int argc, char** argv);
+int main(int argc, char** argv) { return zenfs_tool(argc, argv); }
+#else
+int main() {
+  fprintf(stderr, "Please install libzbd to run the zenfs tool\n");
+  return 1;
+}
+#endif  // LIBZBD
+#else
+int main() {
+  fprintf(stderr, "Please install gflags to run rocksdb tools\n");
+  return 1;
+}
+#endif  // GFLAGS
+#endif  // ROCKSDB_LITE
+