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：

1.     result = storage_download_file_to_file1( \
   
2.             pTrackerServer, NULL, \
   
3.             file_id, local_filename, &file_size);
   
4.     if (result != 0)
   
5.     {
   
6.         printf("download file fail, " \
   
7.             "error no: %d, error info: %s\n", \
   
8.             result, STRERROR(result));
   
9.     }

我们到这个函数的内部，storage_client.c 707行：

1. int storage_download_file_to_file1(TrackerServerInfo *pTrackerServer, \
   
2.         TrackerServerInfo *pStorageServer, \
   
3.         const char *file_id, \
   
4.         const char *local_filename, int64_t *file_size)
   
5. {
   
6.     FDFS_SPLIT_GROUP_NAME_AND_FILENAME(file_id)
   
7. 
   
8.     return storage_download_file_to_file(pTrackerServer, \
   
9.             pStorageServer, group_name, filename, \
   
10.             local_filename, file_size);
    
11. }

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：

1. int storage_download_file_to_file(TrackerServerInfo *pTrackerServer, \
   
2.             TrackerServerInfo *pStorageServer, \
   
3.             const char *group_name, const char *remote_filename, \
   
4.             const char *local_filename, int64_t *file_size)
   
5. {
   
6.     char *pLocalFilename;
   
7.     pLocalFilename = (char *)local_filename;
   
8.     return storage_do_download_file(pTrackerServer, pStorageServer, \
   
9.             FDFS_DOWNLOAD_TO_FILE, group_name, remote_filename, \
   
10.             &pLocalFilename, NULL, file_size);
    
11. }

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：

1. 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：

1.     case STORAGE_PROTO_CMD_DOWNLOAD_FILE:
   
2.             result = storage_server_download_file(pTask);
   
3.             break;

接着分析 storage_server_download_file，storage_service.c 5489行：

1.     return storage_read_from_file(pTask, file_offset, download_bytes, \
   
2.             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：

1.     static int storage_read_from_file(struct fast_task_info *pTask, \
   
2.         const int64_t file_offset, const int64_t download_bytes, \
   
3.         FileDealDoneCallback done_callback, \
   
4.         const int store_path_index)
   
5. {
   
6.     StorageClientInfo *pClientInfo;
   
7.     StorageFileContext *pFileContext;
   
8.     TrackerHeader *pHeader;
   
9.     int result;
   
10. 
    
11.     pClientInfo = (StorageClientInfo *)pTask->arg;
    
12.     pFileContext = &(pClientInfo->file_context);
    
13. 
    
14.     pClientInfo->deal_func = dio_read_file;
    
15.     pClientInfo->clean_func = dio_read_finish_clean_up;
    
16.     pClientInfo->total_length = sizeof(TrackerHeader) + download_bytes;
    
17.     pClientInfo->total_offset = 0;
    
18. 
    
19.     pFileContext->op = FDFS_STORAGE_FILE_OP_READ;
    
20.     pFileContext->open_flags = O_RDONLY | g_extra_open_file_flags;
    
21.     pFileContext->offset = file_offset;
    
22.     pFileContext->start = file_offset;
    
23.     pFileContext->end = file_offset + download_bytes;
    
24.     pFileContext->dio_thread_index = storage_dio_get_thread_index( \
    
25.         pTask, store_path_index, pFileContext->op);
    
26.     pFileContext->done_callback = done_callback;
    
27. 
    
28.     pTask->length = sizeof(TrackerHeader);
    
29. 
    
30.     pHeader = (TrackerHeader *)pTask->data;
    
31.     pHeader->status = 0;
    
32.     pHeader->cmd = STORAGE_PROTO_CMD_RESP;
    
33.     long2buff(download_bytes, pHeader->pkg_len);
    
34. 
    
35.     if ((result=storage_dio_queue_push(pTask)) != 0)
    
36.     {
    
37.         if (pFileContext->fd >= 0)
    
38.         {
    
39.             close(pFileContext->fd);
    
40.         }
    
41.         pClientInfo->total_length = sizeof(TrackerHeader);
    
42.         return result;
    
43.     }
    
44. 
    
45.     return STORAGE_STATUE_DEAL_FILE;
    
46. }

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：

1. int dio_read_file(struct fast_task_info *pTask)
   
2. {
   
3.     StorageFileContext *pFileContext;
   
4.     int result;
   
5.     int64_t remain_bytes;
   
6.     int capacity_bytes;
   
7.     int read_bytes;
   
8. 
   
9.     pFileContext = &(((StorageClientInfo *)pTask->arg)->file_context);
   
10. 
    
11.     do
    
12.     {
    
13.     if ((result=dio_open_file(pFileContext)) != 0)
    
14.     {
    
15.         break;
    
16.     }
    
17. 
    
18.     remain_bytes = pFileContext->end - pFileContext->offset;
    
19.     capacity_bytes = pTask->size - pTask->length;
    
20.     read_bytes = (capacity_bytes < remain_bytes) ? \
    
21.                 capacity_bytes : remain_bytes;
    
22. 
    
23.     /*
    
24.     logInfo("###before dio read bytes: %d, pTask->length=%d, file offset=%ld", \
    
25.         read_bytes, pTask->length, pFileContext->offset);
    
26.     */
    
27. 
    
28.     if (read(pFileContext->fd, pTask->data + pTask->length, \
    
29.         read_bytes) != read_bytes)
    
30.     {
    
31.         result = errno != 0 ? errno : EIO;
    
32.         logError("file: "__FILE__", line: %d, " \
    
33.             "read from file: %s fail, " \
    
34.             "errno: %d, error info: %s", \
    
35.             __LINE__, pFileContext->filename, \
    
36.             result, STRERROR(result));
    
37.     }
    
38. 
    
39.     pthread_mutex_lock(&g_dio_thread_lock);
    
40.     g_storage_stat.total_file_read_count++;
    
41.     if (result == 0)
    
42.     {
    
43.         g_storage_stat.success_file_read_count++;
    
44.     }
    
45.     pthread_mutex_unlock(&g_dio_thread_lock);
    
46. 
    
47.     if (result != 0)
    
48.     {
    
49.         break;
    
50.     }
    
51. 
    
52.     pTask->length += read_bytes;
    
53.     pFileContext->offset += read_bytes;
    
54. 
    
55.     /*
    
56.     logInfo("###after dio read bytes: %d, pTask->length=%d, file offset=%ld", \
    
57.         read_bytes, pTask->length, pFileContext->offset);
    
58.     */
    
59. 
    
60.     if (pFileContext->offset < pFileContext->end)
    
61.     {
    
62.         storage_nio_notify(pTask); //notify nio to deal
    
63.     }
    
64.     else
    
65.     {
    
66.         /* file read done, close it */
    
67.         close(pFileContext->fd);
    
68.         pFileContext->fd = -1;
    
69. 
    
70.         pFileContext->done_callback(pTask, result);
    
71.     }
    
72. 
    
73.     return 0;
    
74. 
    
75.     } while (0);
    
76. 
    
77.     /* file read error, close it */
    
78.     if (pFileContext->fd > 0)
    
79.     {
    
80.         close(pFileContext->fd);
    
81.         pFileContext->fd = -1;
    
82.     }
    
83. 
    
84.     pFileContext->done_callback(pTask, result);
    
85.     return result;
    
86. }

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：

1. static void storage_download_file_done_callback(struct fast_task_info *pTask, \
   
2.             const int err_no)
   
3. {
   
4.     StorageFileContext *pFileContext;
   
5.     TrackerHeader *pHeader;
   
6. 
   
7.     pFileContext = &(((StorageClientInfo *)pTask->arg)->file_context);
   
8.     if (err_no != 0)
   
9.     {
   
10.         pthread_mutex_lock(&stat_count_thread_lock);
    
11.         g_storage_stat.total_download_count++;
    
12.         g_storage_stat.total_download_bytes += \
    
13.                 pFileContext->offset - pFileContext->start;
    
14.         pthread_mutex_unlock(&stat_count_thread_lock);
    
15. 
    
16.         if (pTask->length == sizeof(TrackerHeader)) //never response
    
17.         {
    
18.             pHeader = (TrackerHeader *)pTask->data;
    
19.             pHeader->status = err_no;
    
20.             storage_nio_notify(pTask);
    
21.         }
    
22.         else
    
23.         {
    
24.             task_finish_clean_up(pTask);
    
25.         }
    
26.     }
    
27.     else
    
28.     {
    
29.         CHECK_AND_WRITE_TO_STAT_FILE2_WITH_BYTES( \
    
30.             g_storage_stat.total_download_count, \
    
31.             g_storage_stat.success_download_count, \
    
32.             g_storage_stat.total_download_bytes, \
    
33.             g_storage_stat.success_download_bytes, \
    
34.             pFileContext->end - pFileContext->start)
    
35. 
    
36.         storage_nio_notify(pTask);
    
37.     }
    
38. }

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