HLog的全部实现在包:
org.apache.hadoop.hbase.regionserver.wal 中
相关的配置为
参数名 | 默认值 | 含义 |
hbase.regionserver.hlog.enabled | true | 是否启用WAL |
hbase.regionserver.hlog.writer.impl | SequenceFileLogWriter | HLog.Writer实现类 |
hbase.regionserver.hlog.reader.impl | SequenceFileLogReader | HLog.Reader实现类 |
hbase.regionserver.hlog.keyclass | HLogKey | HLog.Entry的key实现类 |
hbase.regionserver.wal.enablecompression | false | 是否对日志压缩 |
io.file.buffer.size | 4096 | 读取文件的buffer大小 |
hbase.regionserver.hlog.replication | 1 | 复制类型 |
hbase.regionserver.hlog.blocksize | 32M | 文件系统块大小 |
hbase.regionserver.hlog.lowreplication.rolllimit | 5 | 若低于副本数,尝试几次 |
hbase.regionserver.hlog.tolerable.lowreplication | ?? | 未知 |
hbase.regionserver.maxlogs | 32 | 最大日志个数 |
hbase.regionserver.logroll.multiplier | 0.95 | 到HDFS块95%时回滚 |
hbase.regionserver.hlog.lowreplication.rolllimit
其中在Hlog.syncer方法中调用checkLowReplication方法用来判断是否hlog在hdfs上的副本数低于配置项,
若低于则requestLogRoll,最终调用logRollRequested方法,但是调用次数不超过默认5次
核心类HLog,负责创建读,写接口的实现,完成最终的写入数据,读数据等。
HLog.Reader HLog读接口,实现类:SequenceFileLogReader
HLog.Writer HLog写接口,实现类:SequenceFileLogWriter
HLog中包含了若干 HLog.Entry,Entry是一个K/V键值对
键为:HLogKey
值为:WALEdit
工具类:
HLogSplitter(./hbase hlog命令实现类)
Compressor(负责将input内容压缩到output中,或者input内容解压到output中)
WALActionsListener 日志的观察者监听类,当日志发生变化了可以触发观察者,比如replication功能就是实现了这个接口。
HLog是一个二进制格式的文件,文件开头二进制信息:
SEQ0org.apache.hadoop.hbase.regionserver.wal.HLogKey0org.apache.hadoop.hbase.regionserver.wal.WALEdit
一个HLog文件如下:
这里注明了key和value使用的分别是HLogKey和WALEdit两个类
HLog中包含若干Entry,HLog是一个顺序的文件结构,没有索引,所有的Entry都是顺序挨个读取的。
HLog格式如下图:
HLog$Entry包含一个key,value key: HLogKey value: WALEdit Entry二进制内容如下: 1.总数据长度(hlogkey+waledit) 2.key长度 3.数据
HLogKey 1.变长int 2.编码后的region名字(若启用压缩则写入压缩后的内容) 3.表名(若启用压缩则写入压缩后的内容) 4.long长度的序列号 5.long长度的时间戳 6.1字节UUID标识(是否用默认的UUID标识) 7.UUID前8个字节 8.UUID后8个字节
WALEdit 1.int长度版本号 2.int长度KeyValue个数 3.遍历写入KeyValue(若启用压缩则写入压缩后的内容) 3-1.KeyValue长度(int) 3-2.KeyValue的二进制数据(4字节的key长,4字节的value长,2字节的rowkey长,key,1字节的family长,family, qualify,8字节的timestampe,1字节的keytype,变成long的memstoreMVC 4.socpe个数(若scope为null则写入0) 5.遍历写入每个scope(int长度)
HLog读取,写入的列子
/** * 读取HLog中的内容 * @throws IOException */ public void read()throws IOException { FileSystem fs = FileSystem.get(cfg); Path path = new Path("/test/hbase/hlog.1234567890"); HLog.Reader reader = HLog.getReader(fs, path, cfg); HLog.Entry entry = reader.next(); HLogKey key = entry.getKey(); WALEdit edit = entry.getEdit(); System.out.println( Bytes.toString(key.getEncodedRegionName()) ); System.out.println( Bytes.toString(key.getTablename()) ); System.out.println( key.getLogSeqNum() ); System.out.println( key.getWriteTime() ); System.out.println( key.getClusterId().toString() ); List<KeyValue> list = edit.getKeyValues(); for(KeyValue kv:list) { System.out.println(kv.toString()); } while( (entry=reader.next()) !=null ) { System.out.println(entry); } }
/** * 将数据写入到HLog,用HLog.Writer直接写入数据 * @throws IOException */ public void write() throws IOException { FileSystem fs = FileSystem.get(cfg); Path path = new Path("/test/hbase/hlog.1234567890"); HLog.Writer writer = HLog.createWriter(fs, path, cfg); byte[] encodedRegionName = Bytes.toBytes("myregion"); byte[] tablename = Bytes.toBytes("test"); long logSeqNum = 100; long now = System.currentTimeMillis(); UUID clusterId = UUID.randomUUID(); HLogKey key1 = new HLogKey(encodedRegionName, tablename, logSeqNum, now, clusterId); WALEdit edit1 = new WALEdit(); edit1.add(generator("key111111111111111111111111", "column", "a",System.currentTimeMillis(), new byte[] { '2' })); HLog.Entry entry1 = new Entry(key1, edit1); HLogKey key2 = new HLogKey(encodedRegionName, tablename, 101L, now, UUID.randomUUID()); WALEdit edit2 = new WALEdit(); edit2.add(generator("key333333333333333333333333", "column", "b",System.currentTimeMillis(), VALUES)); HLog.Entry entry2 = new Entry(key2, edit2); writer.append(entry2); writer.append(entry1); writer.sync(); writer.close(); }
/** * 使用HLog写入数据 * @throws IOException */ public void hlogWriter() throws IOException { FileSystem fs = FileSystem.get(cfg); String regionDir = "/test/hbase/hlog"; HLog log = new HLog(fs, new Path(regionDir, "logs"), new Path(regionDir, "oldlogs"), cfg); byte[] tableName = Bytes.toBytes("test"); byte[] startKey = Bytes.toBytes("aaaaa1111111111"); byte[] endKey = Bytes.toBytes("zzzzz9999999999"); HRegionInfo info = new HRegionInfo(tableName, startKey, endKey); WALEdit edit = new WALEdit(); edit.add(generator("key333333333333333333333333", "column", "a",System.currentTimeMillis(), new byte[] { '2' })); edit.add(generator("key333333333333333333333333", "column", "b",System.currentTimeMillis(), VALUES)); UUID uuid = UUID.randomUUID(); long now = System.currentTimeMillis(); HColumnDescriptor family = new HColumnDescriptor("column"); HTableDescriptor descriptor = new HTableDescriptor(); descriptor.addFamily(family); log.append(info, tableName, edit, uuid, now, descriptor); log.sync(); log.hflush(); log.hsync(); log.close(); }
/** * 生成KeyValue * @param key * @param column * @param qualifier * @param timestamp * @param value * @return */ public KeyValue generator(String key, String column, String qualifier, long timestamp, byte[] value) { byte[] keyBytes = Bytes.toBytes(key); byte[] familyBytes = Bytes.toBytes(column); byte[] qualifierBytes = Bytes.toBytes(qualifier); Type type = Type.Put; byte[] valueBytes = value; KeyValue kv = new KeyValue(keyBytes, 0, keyBytes.length, familyBytes, 0, familyBytes.length, qualifierBytes, 0, qualifierBytes.length, timestamp, type, valueBytes, 0, valueBytes.length); return kv; }