The data backup and recovery functions have been added

Author: wuxianrong

.github/workflows/pika.yml

@@ -12,6 +12,7 @@ env:
   ARTIFACT_PIKA_NAME: artifact-pika
 
 jobs:
+
   build_on_ubuntu:
     # The CMake configure and build commands are platform-agnostic and should work equally well on Windows or Mac.
     # You can convert this to a matrix build if you need cross-platform coverage.
@@ -41,12 +42,14 @@ jobs:
       - name: Install Deps
         run: |
           sudo apt-get update
-          sudo apt-get install -y autoconf libprotobuf-dev protobuf-compiler clang-tidy
+          # Use GCC 10 to avoid __has_attribute compatibility issues with newer GCC and glibc
+          sudo apt-get install -y autoconf libprotobuf-dev protobuf-compiler clang-tidy gcc-10 g++-10 zlib1g-dev
 
       - name: Configure CMake
         # Configure CMake in a 'build' subdirectory. `CMAKE_BUILD_TYPE` is only required if you are using a single-configuration generator such as make.
         # See https://cmake.org/cmake/help/latest/variable/CMAKE_BUILD_TYPE.html?highlight=cmake_build_type
-        run: cmake -B build -DCMAKE_BUILD_TYPE=${{ env.BUILD_TYPE }} -DUSE_PIKA_TOOLS=ON -DCMAKE_CXX_FLAGS_DEBUG=-fsanitize=address -D CMAKE_C_COMPILER_LAUNCHER=ccache -D CMAKE_CXX_COMPILER_LAUNCHER=ccache
+        # Use GCC 10 to avoid __has_attribute compatibility issues with GCC 11+ and glibc
+        run: cmake -B build -DCMAKE_BUILD_TYPE=${{ env.BUILD_TYPE }} -DUSE_PIKA_TOOLS=ON -DCMAKE_CXX_FLAGS_DEBUG=-fsanitize=address -D CMAKE_C_COMPILER_LAUNCHER=ccache -D CMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_C_COMPILER=gcc-10 -DCMAKE_CXX_COMPILER=g++-10
 
       - name: Build
         # Build your program with the given configuration
@@ -166,7 +169,8 @@ jobs:
       - name: Install deps
         run: |
           dnf update -y
-          dnf install -y bash cmake wget git autoconf gcc perl-Digest-SHA tcl which tar g++ tar epel-release gcc-c++ libstdc++-devel gcc-toolset-13 binutils
+          # Use gcc-toolset-11 to avoid __has_attribute compatibility issues with GCC 12+ and glibc
+          dnf install -y bash cmake wget git autoconf gcc perl-Digest-SHA tcl which tar g++ tar epel-release gcc-c++ libstdc++-devel gcc-toolset-11 binutils openssl openssl-devel zlib-devel
           dnf clean all
           rm -rf /var/cache/dnf
 
@@ -181,8 +185,9 @@ jobs:
           fetch-depth: 1
 
       - name: Configure CMake
+        # Use gcc-toolset-11 to avoid __has_attribute compatibility issues with GCC 12+ and glibc
         run: |
-          source /opt/rh/gcc-toolset-13/enable
+          source /opt/rh/gcc-toolset-11/enable
           cmake -B build -DCMAKE_BUILD_TYPE=${{ env.BUILD_TYPE }} -DUSE_PIKA_TOOLS=ON -DCMAKE_CXX_FLAGS_DEBUG=-fsanitize=address .
 
       - uses: actions/cache@v3
@@ -196,8 +201,9 @@ jobs:
           key: ${{ runner.os }}-rocky-buildtrees-${{ hashFiles('**/CMakeLists.txt') }}
 
       - name: Build
+        # Use gcc-toolset-11 to avoid __has_attribute compatibility issues with GCC 12+ and glibc
         run: |
-          source /opt/rh/gcc-toolset-13/enable
+          source /opt/rh/gcc-toolset-11/enable
           cmake --build build --config ${{ env.BUILD_TYPE }}
 
       - name: Cleanup
@@ -292,13 +298,14 @@ jobs:
       - name: Install Deps
         run: |
           brew list --versions cmake && brew uninstall --ignore-dependencies --force cmake || true
-          brew install gcc@13 automake cmake make binutils
+          brew install gcc@11 automake cmake make binutils
 
       - name: Configure CMake
+        # Use GCC 11 to avoid potential __has_attribute compatibility issues with GCC 13
         run: |
-          GCC_PREFIX=$(brew --prefix gcc@13)
-          export CC=$GCC_PREFIX/bin/gcc-13
-          cmake -B build -DCMAKE_C_COMPILER=$GCC_PREFIX/bin/gcc-13 -DUSE_PIKA_TOOLS=ON -DCMAKE_BUILD_TYPE=${{ env.BUILD_TYPE }} -DCMAKE_CXX_FLAGS_DEBUG=-fsanitize=address -D CMAKE_C_COMPILER_LAUNCHER=ccache -D CMAKE_CXX_COMPILER_LAUNCHER=ccache
+          GCC_PREFIX=$(brew --prefix gcc@11)
+          export CC=$GCC_PREFIX/bin/gcc-11
+          cmake -B build -DCMAKE_C_COMPILER=$GCC_PREFIX/bin/gcc-11 -DUSE_PIKA_TOOLS=ON -DCMAKE_BUILD_TYPE=${{ env.BUILD_TYPE }} -DCMAKE_CXX_FLAGS_DEBUG=-fsanitize=address -D CMAKE_C_COMPILER_LAUNCHER=ccache -D CMAKE_CXX_COMPILER_LAUNCHER=ccache
 
       - name: Build
         run: |

CMakeLists.txt

@@ -22,7 +22,15 @@ elseif (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
   endif()
 endif()
 
-link_directories("/opt/rh/gcc-toolset-13/root/lib/gcc/x86_64-redhat-linux/13")
+# Link directories for gcc-toolset on RHEL/Rocky/CentOS
+# Support gcc-toolset-11, gcc-toolset-12 and gcc-toolset-13
+if(EXISTS "/opt/rh/gcc-toolset-13/root/lib/gcc/x86_64-redhat-linux/13")
+  link_directories("/opt/rh/gcc-toolset-13/root/lib/gcc/x86_64-redhat-linux/13")
+elseif(EXISTS "/opt/rh/gcc-toolset-12/root/lib/gcc/x86_64-redhat-linux/12")
+  link_directories("/opt/rh/gcc-toolset-12/root/lib/gcc/x86_64-redhat-linux/12")
+elseif(EXISTS "/opt/rh/gcc-toolset-11/root/lib/gcc/x86_64-redhat-linux/11")
+  link_directories("/opt/rh/gcc-toolset-11/root/lib/gcc/x86_64-redhat-linux/11")
+endif()
 
 ############# You should enable sanitizer if you are developing pika #############
 # Uncomment the following two lines to enable AddressSanitizer to detect memory leaks and other memory-related bugs.
@@ -648,6 +656,10 @@ else()
 endif()
 set(LEVELDB_INCLUDE_DIR ${INSTALL_INCLUDEDIR})
 
+# Workaround for __has_attribute compatibility issue with newer GCC and glibc
+# The issue is that newer glibc uses __glibc_has_attribute which conflicts with GCC's __has_attribute
+# We need to set an include path that has a fixed version of the problematic header
+
 ExternalProject_Add(brpc
   DEPENDS
   gflags
@@ -657,13 +669,14 @@ ExternalProject_Add(brpc
   snappy
   zlib
   URL
-  https://github.com/apache/brpc/archive/refs/tags/1.6.0.tar.gz
+  https://github.com/apache/brpc/archive/refs/tags/1.11.0.tar.gz
   URL_HASH
-  MD5=0d37cea25bd006e89806f461ef7e39ba
+  MD5=f55e582fb8032768f9070865b48e892d
   DOWNLOAD_NO_PROGRESS
   1
   UPDATE_COMMAND
   ""
+
   LOG_CONFIGURE
   1
   LOG_BUILD
@@ -677,6 +690,8 @@ ExternalProject_Add(brpc
   -DCMAKE_INSTALL_PREFIX=${STAGED_INSTALL_PREFIX}
   -DCMAKE_BUILD_TYPE=${LIB_BUILD_TYPE}
   -DCMAKE_PREFIX_PATH=${CMAKE_PREFIX_PATH}
+  -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
+  -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
   -DWITH_GLOG=ON
   -DWITH_SNAPPY=ON
   -DBUILD_SHARED_LIBS=OFF
@@ -724,6 +739,8 @@ ExternalProject_Add(braft
   -DCMAKE_INSTALL_PREFIX=${STAGED_INSTALL_PREFIX}
   -DCMAKE_BUILD_TYPE=${LIB_BUILD_TYPE}
   -DCMAKE_PREFIX_PATH=${CMAKE_PREFIX_PATH}
+  -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
+  -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
   -DWITH_GLOG=ON
   -DBUILD_SHARED_LIBS=OFF
   -DBUILD_UNIT_TESTS=OFF

docs/raft_implementation_review.md

@@ -0,0 +1,93 @@
+# Pika Raft 模式实现 Review 文档
+
+## 1. 概述
+
+Pika 的 Raft 模式基于 [braft](https://github.com/baidu/braft) 实现分布式一致性，使用 brpc 进行节点间通信。
+
+**核心特性**：
+- 强一致性保证
+- 自动 Leader 选举
+- 基于 RocksDB Checkpoint 的快照机制
+
+## 2. 架构概览
+
+```
+┌─────────────────────────────────────────────────────┐
+│                    Pika Server                       │
+├─────────────────────────────────────────────────────┤
+│  Client Request → RaftManager → PikaRaftNode        │
+│                                    ↓                 │
+│                              braft::Node             │
+│                                    ↓                 │
+│                           PikaStateMachine          │
+│                                    ↓                 │
+│                         Storage::OnBinlogWrite()    │
+│                                    ↓                 │
+│                              RocksDB                 │
+└─────────────────────────────────────────────────────┘
+```
+
+## 3. 核心组件
+
+| 组件 | 文件 | 职责 |
+|------|------|------|
+| `RaftManager` | [praft.h](../src/praft/include/praft/praft.h) | 管理多个 Raft 节点（每个 DB 一个） |
+| `PikaRaftNode` | [praft.h](../src/praft/include/praft/praft.h) | 封装 braft::Node，提供日志追加接口 |
+| `PikaStateMachine` | [praft.cc](../src/praft/src/praft.cc) | 实现状态机，处理日志应用和快照 |
+| `PPosixFileSystemAdaptor` | [psnapshot.cc](../src/praft/src/psnapshot.cc) | 快照文件系统适配器 |
+
+## 4. 数据流
+
+### 写入流程
+1. 客户端发送写命令
+2. Storage 层构建 Binlog（Protobuf 格式）
+3. 通过 `PikaRaftNode::AppendLog()` 提交到 Raft
+4. braft 复制日志到多数节点
+5. 提交后 `PikaStateMachine::on_apply()` 被调用
+6. 调用 `Storage::OnBinlogWrite()` 写入 RocksDB
+
+### Binlog 格式
+
+定义在 [binlog.proto](../src/praft/src/binlog.proto)：
+- 支持数据类型：Strings, Hashes, Lists, Sets, ZSets, Streams
+- 操作类型：Put, Delete
+
+## 5. 快照机制
+
+- 使用 RocksDB Checkpoint 创建快照
+- 快照恢复点基于 `GetSmallestFlushedLogIndex()`
+- 通过 `LogIndexOfColumnFamilies` 追踪各 CF 的日志应用进度
+
+## 6. 配置选项
+
+| 配置项 | 默认值 | 说明 |
+|--------|--------|------|
+| `raft_enabled` | false | 是否启用 Raft |
+| `raft_group_id` | "" | Raft Group ID |
+| `raft_election_timeout_ms` | 1000 | 选举超时(ms) |
+| `raft_snapshot_interval_s` | 3600 | 快照间隔(s) |
+
+## 7. Review 重点
+
+### 正确性
+- **日志幂等性**: `IsApplied()` 检查是否已应用，防止重复应用
+- **快照一致性**: 使用最小 flushed_log_index 确定快照点
+- **WAL 禁用**: Raft 日志提供持久性保证，RocksDB WAL 被禁用
+
+### 性能
+- 同步写入等待 Raft 提交（可优化为批量提交）
+- Binlog 使用 Protobuf 序列化
+
+### 线程安全
+- `LogIndexOfColumnFamilies` 使用 mutex 保护
+- braft 保证 `on_apply()` 顺序调用
+
+## 8. 关键代码位置
+
+| 功能 | 文件 |
+|------|------|
+| 日志应用 | [praft.cc](../src/praft/src/praft.cc) - `on_apply()` |
+| Binlog 处理 | [storage.cc](../src/storage/src/storage.cc) - `OnBinlogWrite()` |
+| Log Index 追踪 | [log_index.cc](../src/storage/src/log_index.cc) |
+| 快照创建 | [psnapshot.cc](../src/praft/src/psnapshot.cc) |
+| Raft 命令 | [pika_raft.cc](../src/pika_raft.cc) |

include/pika_db.h

@@ -94,6 +94,8 @@ class DB : public std::enable_shared_from_this<DB>, public pstd::noncopyable {
   std::shared_ptr<storage::Storage> storage() const;
   void GetBgSaveMetaData(std::vector<std::string>* fileNames, std::string* snapshot_uuid);
   void BgSaveDB();
+  pstd::Status CreateCheckpoint(const std::string& checkpoint_dir);
+  pstd::Status LoadDBFromCheckpoint(const std::string& checkpoint_dir);
   void SetBinlogIoError();
   void SetBinlogIoErrorrelieve();
   bool IsBinlogIoError();

include/pika_server.h

@@ -71,6 +71,8 @@ enum TaskType {
   kCompactRangeSets,
   kCompactRangeZSets,
   kCompactRangeList,
+  kLoadDBFromCheckpoint,
+  kCreateCheckpoint,
 };
 
 struct TaskArg {

src/pika_db.cc

@@ -67,6 +67,75 @@ void DB::BgSaveDB() {
   g_pika_server->BGSaveTaskSchedule(&DoBgSave, static_cast<void*>(bg_task_arg));
 }
 
+pstd::Status DB::CreateCheckpoint(const std::string& checkpoint_dir) {
+  std::string checkpoint_sub_path = checkpoint_dir;
+  if (!checkpoint_sub_path.empty() && checkpoint_sub_path.back() != '/') {
+    checkpoint_sub_path.push_back('/');
+  }
+  checkpoint_sub_path += db_name_;
+
+  if (!pstd::FileExists(checkpoint_sub_path)) {
+    if (pstd::CreatePath(checkpoint_sub_path, 0755) != 0) {
+      return Status::IOError("Failed to create checkpoint path", checkpoint_sub_path);
+    }
+  }
+
+  std::shared_lock guard(dbs_rw_);
+  auto tasks = storage_->CreateCheckpoint(checkpoint_sub_path);
+  for (auto& task : tasks) {
+    auto status = task.get();
+    if (!status.ok()) {
+      return Status::Corruption("Create checkpoint failed: " + status.ToString());
+    }
+  }
+  return Status::OK();
+}
+
+pstd::Status DB::LoadDBFromCheckpoint(const std::string& checkpoint_dir) {
+  std::string checkpoint_sub_path = checkpoint_dir;
+  if (!checkpoint_sub_path.empty() && checkpoint_sub_path.back() != '/') {
+    checkpoint_sub_path.push_back('/');
+  }
+  checkpoint_sub_path += db_name_;
+
+  if (!pstd::FileExists(checkpoint_sub_path)) {
+    return Status::NotFound("Checkpoint dir does not exist: " + checkpoint_sub_path);
+  }
+
+  std::lock_guard<std::shared_mutex> guard(dbs_rw_);
+  opened_ = false;
+
+  auto old_storage = storage_;
+  storage_.reset();
+  if (old_storage) {
+    old_storage->Close();
+  }
+
+  storage_ = std::make_shared<storage::Storage>();
+  auto checkpoint_tasks = storage_->LoadCheckpoint(checkpoint_sub_path, db_path_);
+  for (auto& task : checkpoint_tasks) {
+    auto status = task.get();
+    if (!status.ok()) {
+      storage_.reset();
+      return Status::Corruption("Load checkpoint failed: " + status.ToString());
+    }
+  }
+
+  storage::StorageOptions storage_options = g_pika_server->storage_options();
+  auto open_status = storage_->Open(storage_options, db_path_);
+  if (!open_status.ok()) {
+    storage_.reset();
+    return Status::Corruption("Storage open failed: " + open_status.ToString());
+  }
+
+  if (!g_pika_conf->raft_enabled()) {
+    storage_->DisableWal(false);
+  }
+
+  opened_ = true;
+  return Status::OK();
+}
+
 void DB::SetBinlogIoError() { return binlog_io_error_.store(true); }
 void DB::SetBinlogIoErrorrelieve() { return binlog_io_error_.store(false); }
 bool DB::IsBinlogIoError() { return binlog_io_error_.load(); }
@@ -206,6 +275,11 @@ bool DB::FlushDBWithoutLock() {
   }
 
   LOG(INFO) << db_name_ << " Delete old db...";
+  // First close the storage properly to cancel background work and wait for threads
+  // This prevents crash when RocksDB background threads access destroyed resources
+  if (storage_) {
+    storage_->Close();
+  }
   storage_.reset();
 
   std::string dbpath = db_path_;

src/pika_server.cc

@@ -497,6 +497,37 @@ Status PikaServer::DoSameThingSpecificDB(const std::set<std::string>& dbs, const
       case TaskType::kCompactRangeList:
         db_item.second->CompactRange(storage::DataType::kLists, arg.argv[0], arg.argv[1]);
         break;
+      case TaskType::kLoadDBFromCheckpoint: {
+        // arg.argv[0] should contain checkpoint_path
+        if (arg.argv.empty()) {
+          LOG(ERROR) << "LoadDBFromCheckpoint requires checkpoint_path argument";
+          return Status::InvalidArgument("Missing checkpoint_path");
+        }
+        std::string checkpoint_path = arg.argv[0];
+        auto s = db_item.second->LoadDBFromCheckpoint(checkpoint_path);
+        if (!s.ok()) {
+          LOG(ERROR) << "Failed to load DB from checkpoint: " << s.ToString();
+          return s;
+        }
+        LOG(INFO) << "Successfully loaded DB " << db_item.first << " from checkpoint: " << checkpoint_path;
+        break;
+      }
+      case TaskType::kCreateCheckpoint: {
+        // arg.argv[0] should contain checkpoint_path
+        if (arg.argv.empty()) {
+          LOG(ERROR) << "CreateCheckpoint requires checkpoint_path argument";
+          return Status::InvalidArgument("Missing checkpoint_path");
+        }
+        std::string checkpoint_path = arg.argv[0];
+        auto s = db_item.second->CreateCheckpoint(checkpoint_path);
+        if (!s.ok()) {
+          LOG(ERROR) << "Failed to create checkpoint: " << s.ToString();
+          return s;
+        }
+        LOG(INFO) << "Successfully created checkpoint for DB " << db_item.first << " at: " << checkpoint_path;
+        break;
+      }
+      
       default:
         break;
     }