Android MediaScanner MediaProvider流程以及性能优化,音视频扫描

  • 快速扫描
  • 一、源码解析
    • github链接
    • MediaScanner时序图
    • MediaSacannerReeiver.java
    • MediaScannerService.java
    • MediaProvider.java
    • MediaScanner.java
    • MediaScanner.cpp
    • StagefrightMediaScanner.cpp
  • 二、配置修改
    • 1、修改数据库路径
    • 2、修改数据库WAL模式
  • 三、存在的问题
  • 四、性能优化
    • 1、扫描方案优化
      • 背景
      • 快表流程
    • 2、程序优化

快速扫描

传送们: 快速扫描程序.
快速扫描部分数据量大的情况下比Android原生的MediaScanner快几十倍(我车机原生两万多音视频第一次扫描十几分钟,快速扫描10秒),缺点就是没有专辑、艺术家列表。UI部分我做的很随意,主要还是想快速实现,代码也很乱,之后会整理好。
简单效果:
[video(video-fWcdsuHl-1601061165426)(type-bilibili)(url-https://player.bilibili.com/player.html?aid=839660626)(image-https://img-blog.csdnimg.cn/img_convert/749c9eca2d929b923733cc6b60275858.png)(title-快速扫描)

一、源码解析

github链接

添加了多屏操作和手势识别,向左滑动可以把视频放到副屏播放,向上滑动退出视频播放。
github,欢迎交流.

MediaScanner时序图

链接:
MediaScanner时序图.

MediaSacannerReeiver.java

主要负责service的启动,接收android.intent.action.MEDIA_MOUNTED广播启动MediaScannerService

// An highlighted blockprivate void scan(Context context, String volume) {Bundle args = new Bundle();args.putString("volume", volume);context.startService(new Intent(context, MediaScannerService.class).putExtras(args));}

MediaScannerService.java

第一次被启动走onCreate,将自己的线程启动

// A code block@Overridepublic void onCreate() {PowerManager pm = (PowerManager)getSystemService(Context.POWER_SERVICE);mWakeLock = pm.newWakeLock(PowerManager.PARTIAL_WAKE_LOCK, TAG);StorageManager storageManager = (StorageManager)getSystemService(Context.STORAGE_SERVICE);mExternalStoragePaths = storageManager.getVolumePaths();// Start up the thread running the service.  Note that we create a// separate thread because the service normally runs in the process's// main thread, which we don't want to block.Thread thr = new Thread(null, this, "MediaScannerService");thr.start();}

第二次启动走onStartCommand,从广播里面获取信息发送给mServiceHandler,通过handler通知service启动线程扫描。

// An highlighted block@Overridepublic int onStartCommand(Intent intent, int flags, int startId) {while (mServiceHandler == null) {synchronized (this) {try {wait(100);} catch (InterruptedException e) {}}}if (intent == null) {Log.e(TAG, "Intent is null in onStartCommand: ",new NullPointerException());return Service.START_NOT_STICKY;}Message msg = mServiceHandler.obtainMessage();msg.arg1 = startId;msg.obj = intent.getExtras();mServiceHandler.sendMessage(msg);// Try again later if we are killed before we can finish scanning.return Service.START_REDELIVER_INTENT;}

mServiceHandler 解析路径和volume信息然后开始扫描:

// An highlighted blockprivate final class ServiceHandler extends Handler {@Overridepublic void handleMessage(Message msg) {...scan(directories, volume);...}};

scan(String[] directories, String volumeName)方法中先 openDatabase(volumeName);发消息给MedaiProvider,让数据库先准备好,然后MediaScanner scanner = new MediaScanner(this, volumeName),scanner.scanDirectories(directories);MediaScanner.java开始扫描。

MediaProvider.java

MediaScanner.java

具体流程可以在上面提供的时序图查看,这里主要讲解几个重要的方法:
1、prescan
prescan主要是做老数据删除,先从数据库将数据读取出来,然后判断文件存不存在,不存在就删除。

// An highlighted block
private void prescan(String filePath, boolean prescanFiles) throws RemoteException {Cursor c = null;String where = null;String[] selectionArgs = null;mPlayLists.clear();//清除列表,这个列表后面用来保存每个媒体问的信息:id,修改时间等if (filePath != null) {//获取单个数据// query for only one filewhere = MediaStore.Files.FileColumns._ID + ">?" +" AND " + Files.FileColumns.DATA + "=?";selectionArgs = new String[] { "", filePath };} else {//从数据库files表获取所有数据where = MediaStore.Files.FileColumns._ID + ">?";selectionArgs = new String[] { "" };}mDefaultRingtoneSet = wasRingtoneAlreadySet(Settings.System.RINGTONE);mDefaultNotificationSet = wasRingtoneAlreadySet(Settings.System.NOTIFICATION_SOUND);mDefaultAlarmSet = wasRingtoneAlreadySet(Settings.System.ALARM_ALERT);// Tell the provider to not delete the file.// If the file is truly gone the delete is unnecessary, and we want to avoid// accidentally deleting files that are really there (this may happen if the// filesystem is mounted and unmounted while the scanner is running).Uri.Builder builder = mFilesUri.buildUpon();builder.appendQueryParameter(MediaStore.PARAM_DELETE_DATA, "false");MediaBulkDeleter deleter = new MediaBulkDeleter(mMediaProvider, builder.build());// Build the list of files from the content providertry {if (prescanFiles) {// First read existing files from the files table.// Because we'll be deleting entries for missing files as we go,// we need to query the database in small batches, to avoid problems// with CursorWindow positioning.long lastId = Long.MIN_VALUE;//每次操作限制读取1000个数据Uri limitUri = mFilesUri.buildUpon().appendQueryParameter("limit", "1000").build();while (true) {selectionArgs[0] = "" + lastId;if (c != null) {c.close();c = null;}c = mMediaProvider.query(limitUri, FILES_PRESCAN_PROJECTION,where, selectionArgs, MediaStore.Files.FileColumns._ID, null);if (c == null) {break;}int num = c.getCount();//获取到的数据个数判断是否为0,空的话就不用处理了if (num == 0) {break;}//对1000个数据进行处理while (c.moveToNext()) {long rowId = c.getLong(FILES_PRESCAN_ID_COLUMN_INDEX);String path = c.getString(FILES_PRESCAN_PATH_COLUMN_INDEX);int format = c.getInt(FILES_PRESCAN_FORMAT_COLUMN_INDEX);//数据库里获取文件最后修改时间long lastModified = c.getLong(FILES_PRESCAN_DATE_MODIFIED_COLUMN_INDEX);lastId = rowId;// Only consider entries with absolute path names.// This allows storing URIs in the database without the// media scanner removing them.if (path != null && path.startsWith("/")) {boolean exists = false;try {//查询文件在系统里是否存在exists = Os.access(path, android.system.OsConstants.F_OK);} catch (ErrnoException e1) {}if (!exists && !MtpConstants.isAbstractObject(format)) {// do not delete missing playlists, since they may have been// modified by the user.// The user can delete them in the media player instead.// instead, clear the path and lastModified fields in the rowMediaFile.MediaFileType mediaFileType = MediaFile.getFileType(path);int fileType = (mediaFileType == null ? 0 : mediaFileType.fileType);//添加要删除的iddeleter.delete(rowId);//如果.nomedia文件被删除了,那么就需要重新扫描这个文件夹,因为之前没有扫描。if (path.toLowerCase(Locale.US).endsWith("/.nomedia")) {//开始删除老数据deleter.flush();String parent = new File(path).getParent();mMediaProvider.call(MediaStore.UNHIDE_CALL, parent, null);}}}}}}}}finally {if (c != null) {c.close();}//开始删除老数据deleter.flush();}// compute original size of imagesmOriginalCount = 0;c = mMediaProvider.query(mImagesUri, ID_PROJECTION, null, null, null, null);if (c != null) {mOriginalCount = c.getCount();c.close();}}

2、beginFile
做一些准备工作,保存文件的一些信息,如id、最后修改时间、文件是否被更改等。
3、endFile
每个文件处理都会调用一次endfile,主要是来判别文件的类型,插入到对应的表,不过并不是每次都插入,MediaIsert.java文件会对插入的数据计数,超过250条数据就一起插入数据库,调用bulkInsert。

// An highlighted blockprivate Uri endFile(FileEntry entry, boolean ringtones, boolean notifications,boolean alarms, boolean music, boolean podcasts)throws RemoteException {// update database// use album artist if artist is missingif (mArtist == null || mArtist.length() == 0) {mArtist = mAlbumArtist;}//toValues保存专辑、艺术家、标题等信息,这些信息是从navie函数handleStringTag获取的,//这个函数是通过JNI:android_media_MediaScanner.cpp调用到MediaScannerClient的//成员函数,这个函数被MediaScannerClient::addStringTag包装,其实最终还是被//StagefrightMediaScanner.cpp调用,StagefrightMediaScanner中主要是从歌曲或者视频//文件读取专辑信息,然后通过addStringTag传递给Java侧MediaScanner.javaContentValues values = toValues();String title = values.getAsString(MediaStore.MediaColumns.TITLE);if (title == null || TextUtils.isEmpty(title.trim())) {title = MediaFile.getFileTitle(values.getAsString(MediaStore.MediaColumns.DATA));values.put(MediaStore.MediaColumns.TITLE, title);}String album = values.getAsString(Audio.Media.ALBUM);if (MediaStore.UNKNOWN_STRING.equals(album)) {album = values.getAsString(MediaStore.MediaColumns.DATA);// extract last path segment before file nameint lastSlash = album.lastIndexOf('/');if (lastSlash >= 0) {int previousSlash = 0;while (true) {int idx = album.indexOf('/', previousSlash + 1);if (idx < 0 || idx >= lastSlash) {break;}previousSlash = idx;}if (previousSlash != 0) {album = album.substring(previousSlash + 1, lastSlash);values.put(Audio.Media.ALBUM, album);}}}//entry是在beginFile中就创建好,主要保存从数据库获取的该文件的信息,如果是新数据,//rowId就是0 long rowId = entry.mRowId;if (MediaFile.isAudioFileType(mFileType) && (rowId == 0 || mMtpObjectHandle != 0)) {// Only set these for new entries. For existing entries, they// may have been modified later, and we want to keep the current// values so that custom ringtones still show up in the ringtone// picker.values.put(Audio.Media.IS_RINGTONE, ringtones);values.put(Audio.Media.IS_NOTIFICATION, notifications);values.put(Audio.Media.IS_ALARM, alarms);values.put(Audio.Media.IS_MUSIC, music);values.put(Audio.Media.IS_PODCAST, podcasts);} else if ((mFileType == MediaFile.FILE_TYPE_JPEG|| MediaFile.isRawImageFileType(mFileType)) && !mNoMedia) {ExifInterface exif = null;try {exif = new ExifInterface(entry.mPath);} catch (IOException ex) {// exif is null}if (exif != null) {float[] latlng = new float[2];if (exif.getLatLong(latlng)) {values.put(Images.Media.LATITUDE, latlng[0]);values.put(Images.Media.LONGITUDE, latlng[1]);}long time = exif.getGpsDateTime();if (time != -1) {values.put(Images.Media.DATE_TAKEN, time);} else {// If no time zone information is available, we should consider using// EXIF local time as taken time if the difference between file time// and EXIF local time is not less than 1 Day, otherwise MediaProvider// will use file time as taken time.time = exif.getDateTime();if (time != -1 && Math.abs(mLastModified * 1000 - time) >= 86400000) {values.put(Images.Media.DATE_TAKEN, time);}}int orientation = exif.getAttributeInt(ExifInterface.TAG_ORIENTATION, -1);if (orientation != -1) {// We only recognize a subset of orientation tag values.int degree;switch(orientation) {case ExifInterface.ORIENTATION_ROTATE_90:degree = 90;break;case ExifInterface.ORIENTATION_ROTATE_180:degree = 180;break;case ExifInterface.ORIENTATION_ROTATE_270:degree = 270;break;default:degree = 0;break;}values.put(Images.Media.ORIENTATION, degree);}}}Uri tableUri = mFilesUri;MediaInserter inserter = mMediaInserter;if (!mNoMedia) {//判断是什么类型的文件,同时创建对应的URIif (MediaFile.isVideoFileType(mFileType)) {tableUri = mVideoUri;} else if (MediaFile.isImageFileType(mFileType)) {tableUri = mImagesUri;} else if (MediaFile.isAudioFileType(mFileType)) {tableUri = mAudioUri;}}Uri result = null;boolean needToSetSettings = false;// Setting a flag in order not to use bulk insert for the file related with// notifications, ringtones, and alarms, because the rowId of the inserted file is// needed.if (notifications && !mDefaultNotificationSet) {if (TextUtils.isEmpty(mDefaultNotificationFilename) ||doesPathHaveFilename(entry.mPath, mDefaultNotificationFilename)) {needToSetSettings = true;}} else if (ringtones && !mDefaultRingtoneSet) {if (TextUtils.isEmpty(mDefaultRingtoneFilename) ||doesPathHaveFilename(entry.mPath, mDefaultRingtoneFilename)) {needToSetSettings = true;}} else if (alarms && !mDefaultAlarmSet) {if (TextUtils.isEmpty(mDefaultAlarmAlertFilename) ||doesPathHaveFilename(entry.mPath, mDefaultAlarmAlertFilename)) {needToSetSettings = true;}}if (rowId == 0) {if (mMtpObjectHandle != 0) {values.put(MediaStore.MediaColumns.MEDIA_SCANNER_NEW_OBJECT_ID, mMtpObjectHandle);}if (tableUri == mFilesUri) {int format = entry.mFormat;if (format == 0) {format = MediaFile.getFormatCode(entry.mPath, mMimeType);}values.put(Files.FileColumns.FORMAT, format);}// New file, insert it.// Directories need to be inserted before the files they contain, so they// get priority when bulk inserting.// If the rowId of the inserted file is needed, it gets inserted immediately,// bypassing the bulk inserter.if (inserter == null || needToSetSettings) {if (inserter != null) {//把数据插入数据库inserter.flushAll();}result = mMediaProvider.insert(tableUri, values);} else if (entry.mFormat == MtpConstants.FORMAT_ASSOCIATION) {inserter.insertwithPriority(tableUri, values);} else {inserter.insert(tableUri, values);}if (result != null) {rowId = ContentUris.parseId(result);entry.mRowId = rowId;}} else {// updated fileresult = ContentUris.withAppendedId(tableUri, rowId);// path should never change, and we want to avoid replacing mixed cased paths// with squashed lower case pathsvalues.remove(MediaStore.MediaColumns.DATA);int mediaType = 0;if (!MediaScanner.isNoMediaPath(entry.mPath)) {int fileType = MediaFile.getFileTypeForMimeType(mMimeType);if (MediaFile.isAudioFileType(fileType)) {mediaType = FileColumns.MEDIA_TYPE_AUDIO;} else if (MediaFile.isVideoFileType(fileType)) {mediaType = FileColumns.MEDIA_TYPE_VIDEO;} else if (MediaFile.isImageFileType(fileType)) {mediaType = FileColumns.MEDIA_TYPE_IMAGE;} else if (MediaFile.isPlayListFileType(fileType)) {mediaType = FileColumns.MEDIA_TYPE_PLAYLIST;}values.put(FileColumns.MEDIA_TYPE, mediaType);}mMediaProvider.update(result, values, null, null);}if(needToSetSettings) {if (notifications) {setRingtoneIfNotSet(Settings.System.NOTIFICATION_SOUND, tableUri, rowId);mDefaultNotificationSet = true;} else if (ringtones) {setRingtoneIfNotSet(Settings.System.RINGTONE, tableUri, rowId);mDefaultRingtoneSet = true;} else if (alarms) {setRingtoneIfNotSet(Settings.System.ALARM_ALERT, tableUri, rowId);mDefaultAlarmSet = true;}}return result;}

MediaScanner.cpp

主要负责文件的扫描,深度扫描存储设备,每扫描到文件就通过JNI调用Java侧的scanFile方法。

StagefrightMediaScanner.cpp

对文件类型进行过滤,读取媒体文件信息如标题、艺术家等。

二、配置修改

1、修改数据库路径

2、修改数据库WAL模式

  • 设置db.enableWriteAheadLogging();可以在MediaProvider写数据的时候,UI读数据不会被阻塞。读写不会阻塞,但是只允许同时只有一用户写。设置WAL模式可以提高数据库写速度,降低磁盘IO,但是读数据就会变慢。具体原理可以参考

链接: SQLite分析之WAL机制.

三、存在的问题

  • Android MediaProvider框架对于手机扫描来说是很贴切的,但是对于汽车车载系统来说就不是很友好,因为车机可能需要用USB来存储媒体数据,USB不像手机内置SD卡不用频繁插拔,如果用户更换新的USB,那么从USB读取数据并写入数据库需要耗费很多时间,从用户的角度看,我插入U盘然后查找到指定歌曲的速度太慢的话是否能接受呢。所以针对这样的问题,接下来就是车载系统的一些关于USB扫描的探讨和优化。

问题:
1、对于不同USB,MediaProvider会保留不同的数据库,占用多余磁盘空间;
2、扫描时会读取音视频文件title等信息,读取文件磁盘IO会导致扫描速度变慢,原本需要5分钟,可能就变成20分钟;
3、prescan预扫描时候可以读取数据,并且prescan后也没有明确的广播通知,如果数据被大量删除,UI会读取到已经删除的数据;
4、扫描是顺序扫描,如果一直在扫描歌曲还没扫描到视频,那么视频要等好久才能检索到;

四、性能优化

1、扫描方案优化

背景

  • 对于IO读写慢的问题是无法回避的,为什么要读取文件信息,因为需要添加歌曲视频的专辑标题等信息,在UI侧才能做成专辑等列表。但是从用户的角度分析,如果我U盘插入车机,要听音乐,我大体上用打开歌曲列表,或者文件夹,就可以快速找到自己想要播放的歌曲,或者是收藏列表。可能专辑风格艺术家等列表被打开的概率只有20%,但是这20%的概率却占用了扫描80%的时间,我觉得是不合理的,但是又不能不做,所以我觉得,可以先用很短的时间做成一个快表,这个快表能够提供歌曲视频列表,文件夹信息和收藏列表,之后再做专辑列表。我自己也尝试去做这个快表,如果整体扫描时间是20分钟的话,做快表的时间在5-15秒就可以完成,理论上可以达到5秒。等这个快表做成,就开始走正常的扫描流程,或者说两者一起并行运行也是可以的。
  • 这个方案的缺陷就是CPU占比在一瞬间会比较高,而且原来扫描流程会慢几秒(我感觉可以省略)。
  • 快速扫描功能包括预扫描删除没用的老数据、扫描数据时会判别是否存在、是否更改过,每个文件都保存上级文件夹的id,有一个歌曲表一个视频表和一个文件夹表。

快表流程

对插入的设备进行型号保存,如果是上次插入的设备,那么在扫描前要先做预扫描,删除数据库有但是存储设备中已经没有的数据,如果是新设备就直接扫描,扫描方式选择广度扫描。每扫描到一个文件就判别类型,只处理文件夹、音乐、视频,同时对媒体文件进行过滤(个数控制,支持播放类型过滤),获取父文件夹id。查看并保存文件修改时间,从数据库查询是否存在同名文件(数据库对绝对路径加索引),同名的话要对比修改时间,如果修改时间变了,就说明需要更新,如果不存在就要插入,写sql语句,保存:绝对路径、父文件夹id等信息,将对应sql语句插入到一个链表中(有两个链表,一个插入,一个存储)。当链表达到一定数量就通过事务方式写入数据库,开一个新线程来处理数据插入,主线程接着扫描,要加锁。

2、程序优化