feat(disk): add broken disk check while initialize

include/dsn/utility/filesystem.h

@@ -136,6 +136,16 @@ bool verify_file(const std::string &fname,
                  const std::string &expected_md5,
                  const int64_t &expected_fsize);
 
+// create driectory and get absolute path
+bool create_directory(const std::string &path,
+                      /*out*/ std::string &absolute_path,
+                      /*out*/ std::string &err_msg);
+
+bool write_file(const std::string &fname, std::string &buf);
+
+// check if directory is readable and writable
+bool check_dir_rw(const std::string &path, /*out*/ std::string &err_msg);
+
 } // namespace filesystem
 } // namespace utils
 } // namespace dsn

src/common/fs_manager.cpp

@@ -156,6 +156,7 @@ dsn::error_code fs_manager::initialize(const std::vector<std::string> &data_dirs
                norm_path.c_str(),
                tags[i].c_str());
     }
+    _available_data_dirs = data_dirs;
 
     if (!for_test) {
         update_disk_stat();

src/common/fs_manager.h

@@ -79,6 +79,12 @@ class fs_manager
     bool for_each_dir_node(const std::function<bool(const dir_node &)> &func) const;
     void update_disk_stat();
 
+    const std::vector<std::string> &get_available_data_dirs() const
+    {
+        zauto_read_lock l(_lock);
+        return _available_data_dirs;
+    }
+
 private:
     void reset_disk_stat()
     {
@@ -102,6 +108,7 @@ class fs_manager
     int _max_available_ratio = 0;
 
     std::vector<std::shared_ptr<dir_node>> _dir_nodes;
+    std::vector<std::string> _available_data_dirs;
 
     perf_counter_wrapper _counter_total_capacity_mb;
     perf_counter_wrapper _counter_total_available_mb;

src/replica/replica_stub.cpp

@@ -64,6 +64,12 @@
 namespace dsn {
 namespace replication {
 
+DSN_DEFINE_bool("replication",
+                ignore_broken_disk,
+                true,
+                "true means ignore broken data disk when initialize");
+DSN_TAG_VARIABLE(ignore_broken_disk, FT_MUTABLE);
+
 bool replica_stub::s_not_exit_on_log_failure = false;
 
 replica_stub::replica_stub(replica_state_subscriber subscriber /*= nullptr*/,
@@ -503,33 +509,13 @@ void replica_stub::initialize(const replication_options &opts, bool clear /* = f
     }
 
     // init dirs
-    if (!dsn::utils::filesystem::create_directory(_options.slog_dir)) {
-        dassert(false, "Fail to create directory %s.", _options.slog_dir.c_str());
-    }
     std::string cdir;
-    if (!dsn::utils::filesystem::get_absolute_path(_options.slog_dir, cdir)) {
-        dassert(false, "Fail to get absolute path from %s.", _options.slog_dir.c_str());
+    std::string err_msg;
+    if (!dsn::utils::filesystem::create_directory(_options.slog_dir, cdir, err_msg)) {
+        dassert(false, "{}", err_msg);
     }
     _options.slog_dir = cdir;
-    int count = 0;
-    for (auto &dir : _options.data_dirs) {
-        if (!dsn::utils::filesystem::create_directory(dir)) {
-            dassert(false, "Fail to create directory %s.", dir.c_str());
-        }
-        std::string cdir;
-        if (!dsn::utils::filesystem::get_absolute_path(dir, cdir)) {
-            dassert(false, "Fail to get absolute path from %s.", dir.c_str());
-        }
-        dir = cdir;
-        ddebug("data_dirs[%d] = %s", count, dir.c_str());
-        count++;
-    }
-
-    {
-        dsn::error_code err;
-        err = _fs_manager.initialize(_options.data_dirs, _options.data_dir_tags, false);
-        dassert(err == dsn::ERR_OK, "initialize fs manager failed, err(%s)", err.to_string());
-    }
+    initialize_fs_manager(_options.data_dirs, _options.data_dir_tags);
 
     _log = new mutation_log_shared(_options.slog_dir,
                                    _options.log_shared_file_size_mb,
@@ -541,7 +527,7 @@ void replica_stub::initialize(const replication_options &opts, bool clear /* = f
     ddebug("start to load replicas");
 
     std::vector<std::string> dir_list;
-    for (auto &dir : _options.data_dirs) {
+    for (auto &dir : _fs_manager.get_available_data_dirs()) {
         std::vector<std::string> tmp_list;
         if (!dsn::utils::filesystem::get_subdirectories(dir, tmp_list, false)) {
             dassert(false, "Fail to get subdirectories in %s.", dir.c_str());
@@ -789,6 +775,37 @@ void replica_stub::initialize(const replication_options &opts, bool clear /* = f
     }
 }
 
+void replica_stub::initialize_fs_manager(std::vector<std::string> &data_dirs,
+                                         std::vector<std::string> &data_dir_tags)
+{
+    std::string cdir;
+    std::string err_msg;
+    int count = 0;
+    std::vector<std::string> available_dirs;
+    std::vector<std::string> available_dir_tags;
+    for (int i = 0; i < data_dir_tags.size(); ++i) {
+        std::string &dir = data_dirs[i];
+        if (!utils::filesystem::create_directory(dir, cdir, err_msg) ||
+            !utils::filesystem::check_dir_rw(dir, err_msg)) {
+            if (FLAGS_ignore_broken_disk) {
+                dwarn_f("data dir[{}] is broken, ignore it, error:{}", dir, err_msg);
+            } else {
+                dassert_f(false, "{}", err_msg);
+            }
+            continue;
+        }
+        ddebug_f("data_dirs[{}] = {}", count, cdir);
+        available_dirs.emplace_back(cdir);
+        available_dir_tags.emplace_back(data_dir_tags[i]);
+        count++;
+    }
+
+    dassert_f(available_dirs.size() > 0,
+              "initialize fs manager failed, no available data directory");
+    error_code err = _fs_manager.initialize(available_dirs, available_dir_tags, false);
+    dassert_f(err == dsn::ERR_OK, "initialize fs manager failed, err({})", err);
+}
+
 void replica_stub::initialize_start()
 {
     // start timer for configuration sync
@@ -1860,7 +1877,7 @@ void replica_stub::on_disk_stat()
     uint64_t start = dsn_now_ns();
     disk_cleaning_report report{};
 
-    dsn::replication::disk_remove_useless_dirs(_options.data_dirs, report);
+    dsn::replication::disk_remove_useless_dirs(_fs_manager.get_available_data_dirs(), report);
     _fs_manager.update_disk_stat();
     update_disk_holding_replicas();
 
@@ -2555,7 +2572,7 @@ std::string replica_stub::get_replica_dir(const char *app_type, gpid id, bool cr
     std::string gpid_str = fmt::format("{}.{}", id, app_type);
     std::string replica_dir;
     bool is_dir_exist = false;
-    for (const std::string &data_dir : _options.data_dirs) {
+    for (const std::string &data_dir : _fs_manager.get_available_data_dirs()) {
         std::string dir = utils::filesystem::path_combine(data_dir, gpid_str);
         if (utils::filesystem::directory_exists(dir)) {
             if (is_dir_exist) {
@@ -2577,7 +2594,7 @@ replica_stub::get_child_dir(const char *app_type, gpid child_pid, const std::str
 {
     std::string gpid_str = fmt::format("{}.{}", child_pid.to_string(), app_type);
     std::string child_dir;
-    for (const std::string &data_dir : _options.data_dirs) {
+    for (const std::string &data_dir : _fs_manager.get_available_data_dirs()) {
         std::string dir = utils::filesystem::path_combine(data_dir, gpid_str);
         // <parent_dir> = <prefix>/<gpid>.<app_type>
         // check if <parent_dir>'s <prefix> is equal to <data_dir>

src/replica/replica_stub.h

@@ -97,6 +97,8 @@ class replica_stub : public serverlet<replica_stub>, public ref_counter
     //
     void initialize(const replication_options &opts, bool clear = false);
     void initialize(bool clear = false);
+    void initialize_fs_manager(std::vector<std::string> &data_dirs,
+                               std::vector<std::string> &data_dir_tags);
     void set_options(const replication_options &opts) { _options = opts; }
     void open_service();
     void close();

src/replica/test/replica_disk_test.cpp

@@ -198,5 +198,23 @@ TEST_F(replica_disk_test, gc_disk_useless_dir)
     ASSERT_EQ(report.error_replica_count, 2);
 }
 
+TEST_F(replica_disk_test, broken_disk_test)
+{
+    // Test cases:
+    // create: true, check_rw: true
+    // create: true, check_rw: false
+    // create: false
+    struct broken_disk_test
+    {
+        std::string mock_create_dir;
+        std::string mock_rw_flag;
+        int32_t data_dir_size;
+    } tests[]{{"true", "true", 3}, {"true", "false", 2}, {"false", "false", 2}};
+    for (const auto &test : tests) {
+        ASSERT_EQ(test.data_dir_size,
+                  ignore_broken_disk_test(test.mock_create_dir, test.mock_rw_flag));
+    }
+}
+
 } // namespace replication
 } // namespace dsn

src/replica/test/replica_disk_test_base.h

@@ -95,6 +95,21 @@ class replica_disk_test_base : public replica_test_base
         }
     }
 
+    int32_t ignore_broken_disk_test(const std::string &mock_create_directory,
+                                    const std::string &mock_check_rw)
+    {
+        std::vector<std::string> data_dirs = {"disk1", "disk2", "disk3"};
+        std::vector<std::string> data_dir_tags = {"tag1", "tag2", "tag3"};
+        auto test_stub = make_unique<mock_replica_stub>();
+        fail::cfg("filesystem_create_directory", "return(" + mock_create_directory + ")");
+        fail::cfg("filesystem_check_dir_rw", "return(" + mock_check_rw + ")");
+        fail::cfg("update_disk_stat", "return()");
+        test_stub->initialize_fs_manager(data_dirs, data_dir_tags);
+        int32_t dir_size = test_stub->_fs_manager.get_available_data_dirs().size();
+        test_stub.reset();
+        return dir_size;
+    }
+
 public:
     int empty_dir_nodes_count = 1;
     int dir_nodes_count = 5;
@@ -131,7 +146,7 @@ class replica_disk_test_base : public replica_test_base
             dir_node *node_disk =
                 new dir_node(fmt::format("tag_empty_{}", num), fmt::format("./tag_empty_{}", num));
             stub->_fs_manager._dir_nodes.emplace_back(node_disk);
-            stub->_options.data_dirs.push_back(node_disk->full_dir);
+            stub->_fs_manager._available_data_dirs.emplace_back(node_disk->full_dir);
             utils::filesystem::create_directory(node_disk->full_dir);
             num--;
         }
@@ -160,7 +175,7 @@ class replica_disk_test_base : public replica_test_base
                                                disk_available_mb,
                                                disk_available_ratio);
 
-            stub->_options.data_dirs.push_back(
+            stub->_fs_manager._available_data_dirs.emplace_back(
                 node_disk->full_dir); // open replica need the options
             utils::filesystem::create_directory(node_disk->full_dir);
 

src/utils/filesystem.cpp

@@ -37,6 +37,7 @@
 
 #include <dsn/c/api_utilities.h>
 #include <dsn/dist/fmt_logging.h>
+#include <dsn/utility/fail_point.h>
 #include <dsn/utility/filesystem.h>
 #include <dsn/utility/utils.h>
 #include <dsn/utility/safe_strerror_posix.h>
@@ -826,6 +827,78 @@ bool verify_file(const std::string &fname,
     return true;
 }
 
+bool create_directory(const std::string &path, std::string &absolute_path, std::string &err_msg)
+{
+    FAIL_POINT_INJECT_F("filesystem_create_directory", [path](string_view str) {
+        // when str contains 'false', and path contains broken_disk_dir, mock create fail(return
+        // false)
+        std::string broken_disk_dir = "disk1";
+        return str.find("false") == string_view::npos ||
+               path.find(broken_disk_dir) == std::string::npos;
+    });
+
+    if (!create_directory(path)) {
+        err_msg = fmt::format("Fail to create directory {}.", path);
+        return false;
+    }
+    if (!get_absolute_path(path, absolute_path)) {
+        err_msg = fmt::format("Fail to get absolute path from {}.", path);
+        return false;
+    }
+    return true;
+}
+
+bool write_file(const std::string &fname, std::string &buf)
+{
+    if (!file_exists(fname)) {
+        derror_f("file({}) doesn't exist", fname);
+        return false;
+    }
+
+    std::ofstream fstream;
+    fstream.open(fname.c_str());
+    fstream << buf;
+    fstream.close();
+    return true;
+}
+
+bool check_dir_rw(const std::string &path, std::string &err_msg)
+{
+    FAIL_POINT_INJECT_F("filesystem_check_dir_rw", [path](string_view str) {
+        // when str contains 'false', and path contains broken_disk_dir, mock check fail(return
+        // false)
+        std::string broken_disk_dir = "disk1";
+        return str.find("false") == string_view::npos ||
+               path.find(broken_disk_dir) == std::string::npos;
+    });
+
+    std::string fname = "read_write_test_file";
+    std::string fpath = path_combine(path, fname);
+    if (!create_file(fpath)) {
+        err_msg = fmt::format("Fail to create test file {}.", fpath);
+        return false;
+    }
+
+    std::string value = "test_value";
+    if (!write_file(fpath, value)) {
+        err_msg = fmt::format("Fail to write file {}.", fpath);
+        return false;
+    }
+
+    std::string buf;
+    if (read_file(fpath, buf) != ERR_OK || buf != value) {
+        err_msg = fmt::format("Fail to read file {} or get wrong value({}).", fpath, buf);
+        return false;
+    }
+
+    if (!remove_path(fpath)) {
+        err_msg = fmt::format("Fail to remove test file {}.", fpath);
+        return false;
+    }
+
+    return true;
+}
+
 } // namespace filesystem
 } // namespace utils
 } // namespace dsn