FastDFS分布式文件系统点滴记录6 -- download下载机制剖析

关于下载，其实和上传文件很相似。这里我们暂时先不考虑nginx httpd的方式，只通过client api 方式与FastDFS 交互。

首先，我们看fdfs_download_file.c ,这个文件下载文件的客户端例子，直接看71行：

    result = storage_download_file_to_file1( \
            pTrackerServer,  NULL, \
            file_id, local_filename, &file_size);
    if (result != 0)
    {
        printf("download file fail, " \
            "error no: %d, error info: %s\n", \
            result, STRERROR(result));
    }

我们到这个函数的内部，storage_client.c 707行：

int storage_download_file_to_file1(TrackerServerInfo *pTrackerServer, \
        TrackerServerInfo *pStorageServer, \
        const char  *file_id, \
        const char  *local_filename, int64_t  *file_size)
{
    FDFS_SPLIT_GROUP_NAME_AND_FILENAME(file_id)

    return storage_download_file_to_file(pTrackerServer, \
            pStorageServer, group_name, filename, \
            local_filename, file_size);
}

代码分析：

1. 首先从file_id ,解析出group name、file name

2. 调用storage_download_file_to_file

接着往下分析，进入storage_download_file_to_file，storage_client.c 719行：

int storage_download_file_to_file(TrackerServerInfo *pTrackerServer, \
            TrackerServerInfo *pStorageServer, \
            const char  *group_name,  const char *remote_filename, \
            const char  *local_filename, int64_t  *file_size)
{
    char *pLocalFilename;
    pLocalFilename =  (char *)local_filename;
    return storage_do_download_file(pTrackerServer, pStorageServer, \
            FDFS_DOWNLOAD_TO_FILE, group_name, remote_filename, \
            &pLocalFilename, NULL, file_size);
}

storage_download_file_to_file函数内部，会调用storage_do_download_file下载文件，注意命令FDFS_DOWNLOAD_TO_FILE。
我们通过分析storage_do_download_file，内部会和tracker服务器命令交互TRACKER_PROTO_CMD_SERVICE_QUERY_FETCH_ONE，获取合适的storage服务器。然后和storage 通过 STORAGE_PROTO_CMD_DOWNLOAD_FILE 命令下载文件。这部分和上传机制类似。我们不做详细分析。
这里，由于与tracker 交互时候发的命令是TRACKER_PROTO_CMD_SERVICE_QUERY_FETCH_ONE，这点与upload机制类似。读者可以参考此前关于upload的机制分析。

我们着重分析下与storage 交互的过程。

storage_service.c 6456行：

int storage_deal_task(struct fast_task_info *pTask)

storage_deal_task 是storage 处理任务的入口，我们之间观察STORAGE_PROTO_CMD_DOWNLOAD_FILE，storage_service.c 6467行：

    case STORAGE_PROTO_CMD_DOWNLOAD_FILE:
            result = storage_server_download_file(pTask);
            break;

接着分析 storage_server_download_file，storage_service.c 5489行：

    return storage_read_from_file(pTask, file_offset, download_bytes, \
            storage_download_file_done_callback, store_path_index);

代码分析：

1. storage_server_download_file 首先解析协议，解析出file相关信息；

2. 然后调用storage_read_from_file，注意回调函数storage_download_file_done_callback

接着往下分析，storage_service.c 5520行：

    static int storage_read_from_file(struct fast_task_info *pTask,  \
        const int64_t file_offset, const int64_t download_bytes, \
        FileDealDoneCallback done_callback, \
        const  int store_path_index)
{
    StorageClientInfo *pClientInfo;
    StorageFileContext *pFileContext;
    TrackerHeader *pHeader;
    int result;

    pClientInfo =  (StorageClientInfo *)pTask->arg;
    pFileContext =  &(pClientInfo->file_context);

    pClientInfo->deal_func = dio_read_file;
    pClientInfo->clean_func = dio_read_finish_clean_up;
    pClientInfo->total_length = sizeof(TrackerHeader) + download_bytes;
    pClientInfo->total_offset = 0;

    pFileContext->op = FDFS_STORAGE_FILE_OP_READ;
    pFileContext->open_flags = O_RDONLY | g_extra_open_file_flags;
    pFileContext->offset = file_offset;
    pFileContext->start = file_offset;
    pFileContext->end = file_offset + download_bytes;
    pFileContext->dio_thread_index = storage_dio_get_thread_index( \
        pTask, store_path_index, pFileContext->op);
    pFileContext->done_callback = done_callback;

    pTask->length = sizeof(TrackerHeader);

    pHeader =  (TrackerHeader *)pTask->data;
    pHeader->status = 0;
    pHeader->cmd = STORAGE_PROTO_CMD_RESP;
    long2buff(download_bytes, pHeader->pkg_len);

    if ((result=storage_dio_queue_push(pTask)) != 0)
    {
        if (pFileContext->fd >= 0)
        {
            close(pFileContext->fd);
        }
        pClientInfo->total_length = sizeof(TrackerHeader);
        return result;
    }

    return STORAGE_STATUE_DEAL_FILE;
}

代码分析：

1.这里和上传相似，通过队列，实现nio和dio 的转换；

2. 注意 FDFS_STORAGE_FILE_OP_READ，说明操作是读取文件；

3. pClientInfo->deal_func = dio_read_file;

4. 上面的函数指针是dio 执行函数；

5. pFileContext->done_callback = done_callback;完成时的回调行数，也就是之前提到的storage_download_file_done_callback。

storage_dio.c 中 dio_thread_entrance 函数处理和上传机制一样。我们直接分析deal_func函数，这里是dio_read_file。
storage_dio.c 334行：

int dio_read_file(struct fast_task_info *pTask)
{
    StorageFileContext *pFileContext;
    int result;
    int64_t remain_bytes;
    int capacity_bytes;
    int read_bytes;

    pFileContext =  &(((StorageClientInfo *)pTask->arg)->file_context);

    do
    {
    if ((result=dio_open_file(pFileContext)) != 0)
    {
        break;
    }

    remain_bytes = pFileContext->end - pFileContext->offset;
    capacity_bytes = pTask->size - pTask->length;
    read_bytes =  (capacity_bytes < remain_bytes) ? \
                capacity_bytes : remain_bytes;

    /*
    logInfo("###before dio read bytes: %d, pTask->length=%d, file offset=%ld", \
        read_bytes, pTask->length, pFileContext->offset);
    */

    if (read(pFileContext->fd, pTask->data + pTask->length, \
        read_bytes)  != read_bytes)
    {
        result = errno  != 0 ? errno : EIO;
        logError("file: "__FILE__", line: %d, " \
            "read from file: %s fail, "  \
            "errno: %d, error info: %s", \
            __LINE__, pFileContext->filename, \
            result, STRERROR(result));
    }

    pthread_mutex_lock(&g_dio_thread_lock);
    g_storage_stat.total_file_read_count++;
    if (result == 0)
    {
        g_storage_stat.success_file_read_count++;
    }
    pthread_mutex_unlock(&g_dio_thread_lock);

    if (result != 0)
    {
        break;
    }

    pTask->length += read_bytes;
    pFileContext->offset += read_bytes;

    /*
    logInfo("###after dio read bytes: %d, pTask->length=%d, file offset=%ld", \
        read_bytes, pTask->length, pFileContext->offset);
    */

    if (pFileContext->offset < pFileContext->end)
    {
        storage_nio_notify(pTask); //notify nio to deal
    }
    else
    {
        /* file read done, close it */
        close(pFileContext->fd);
        pFileContext->fd = -1;

        pFileContext->done_callback(pTask, result);
    }

    return 0;

    } while (0);

    /* file read error, close it */
    if (pFileContext->fd > 0)
    {
        close(pFileContext->fd);
        pFileContext->fd = -1;
    }

    pFileContext->done_callback(pTask, result);
    return result;
}

代码分析：

1. 文件操作，打开文件，读取文件内容；

2.通过storage_nio_notify函数，通知nio 处理；

3.通过nio、dio直接的交互，直到正确读取、发送完文件数据，调用done_callback

我们接着分析 done_callback,也就是storage_download_file_done_callback。storage_service.c 596行：

static void storage_download_file_done_callback(struct fast_task_info *pTask,  \
            const  int err_no)
{
    StorageFileContext *pFileContext;
    TrackerHeader *pHeader;

    pFileContext =  &(((StorageClientInfo *)pTask->arg)->file_context);
    if (err_no != 0)
    {
        pthread_mutex_lock(&stat_count_thread_lock);
        g_storage_stat.total_download_count++;
        g_storage_stat.total_download_bytes +=  \
                pFileContext->offset - pFileContext->start;
        pthread_mutex_unlock(&stat_count_thread_lock);

        if (pTask->length == sizeof(TrackerHeader)) //never response
        {
            pHeader =  (TrackerHeader *)pTask->data;
            pHeader->status = err_no;
            storage_nio_notify(pTask);
        }
        else
        {
            task_finish_clean_up(pTask);
        }
    }
    else
    {
        CHECK_AND_WRITE_TO_STAT_FILE2_WITH_BYTES( \
            g_storage_stat.total_download_count, \
            g_storage_stat.success_download_count, \
            g_storage_stat.total_download_bytes, \
            g_storage_stat.success_download_bytes, \
            pFileContext->end - pFileContext->start)

        storage_nio_notify(pTask);
    }
}

分析发现，storage_download_file_done_callback 也即是做一些文件下载结尾的动作。

OK,这里下载就完成了。和上传很是相似。通过http方式下载，我们放到后面的章节分析。

欢迎感兴趣的朋友一起交流研究，提出意见。

FastDFS技术交流群：164684842