hbase regionserver异常宕机

原因分析：

线上hbase，在凌晨1点左右，发现某一台regionserver进行了重启(regionserver加了守护线程)

1、查看master日志：

2020-02-27 01:04:57,001 ERROR [RpcServer.FifoRWQ.default.read.handler=26,queue=10,port=16000] master.MasterRpcServices: Region server a3ster,16020,1582342923163 reported a fatal error:
ABORTING region server a3ser,16020,1582342923163: Replay of WAL required. Forcing server shutdown
Cause:
org.apache.hadoop.hbase.DroppedSnapshotException: region: T_BL,x0Ax00x00x00x00x00x00x00x00x00x00x00x00,1572576275632.069e4d877a4ff46f9964ac8bcddb09ef.
        at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushCacheAndCommit(HRegion.java:2509)
        at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushcache(HRegion.java:2186)
        at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushcache(HRegion.java:2148)
        at org.apache.hadoop.hbase.regionserver.HRegion.flushcache(HRegion.java:2039)
        at org.apache.hadoop.hbase.regionserver.HRegion.flush(HRegion.java:1965)
        at org.apache.hadoop.hbase.regionserver.MemStoreFlusher.flushRegion(MemStoreFlusher.java:505)
        at org.apache.hadoop.hbase.regionserver.MemStoreFlusher.flushRegion(MemStoreFlusher.java:475)
        at org.apache.hadoop.hbase.regionserver.MemStoreFlusher.access$900(MemStoreFlusher.java:75)
        at org.apache.hadoop.hbase.regionserver.MemStoreFlusher$FlushHandler.run(MemStoreFlusher.java:263)
        at java.lang.Thread.run(Thread.java:748)
Caused by: org.apache.hadoop.hbase.exceptions.TimeoutIOException: Failed to get sync result after 300000 ms for ringBufferSequence=101793126, WAL system stuck?
        at org.apache.hadoop.hbase.regionserver.wal.SyncFuture.get(SyncFuture.java:174)
        at org.apache.hadoop.hbase.regionserver.wal.FSHLog.blockOnSync(FSHLog.java:1406)
        at org.apache.hadoop.hbase.regionserver.wal.FSHLog.publishSyncThenBlockOnCompletion(FSHLog.java:1400)
        at org.apache.hadoop.hbase.regionserver.wal.FSHLog.sync(FSHLog.java:1512)
        at org.apache.hadoop.hbase.regionserver.wal.WALUtil.writeMarker(WALUtil.java:126)
        at org.apache.hadoop.hbase.regionserver.wal.WALUtil.writeFlushMarker(WALUtil.java:75)
        at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushCacheAndCommit(HRegion.java:2486)
        ... 9 more

2020-02-27 01:04:57,032 ERROR [RpcServer.FifoRWQ.default.read.handler=29,queue=8,port=16000] master.MasterRpcServices: Region server a3ser,16020,1582342923163 reported a fatal error:
ABORTING region server a3serz,16020,1582342923163: Replay of WAL required. Forcing server shutdown
Cause:

2、查看regioserver 日志

2020-02-27 01:04:56,813 WARN  [ResponseProcessor for block BP-1884348122-10.62.2.1-1545175191847:blk_1489206371_467735337] hdfs.DFSClient: Slow ReadProcessor read fields took 327586ms (threshold=30000ms); ack: seqno: 1 status: SUCCESS status: SUCCESS downstreamAckTimeNanos: 965211 4: "0000", targets: [11.23.3.3:9866, 11.23.3.5:9866]
2020-02-27 01:04:56,816 FATAL [MemStoreFlusher.6] regionserver.HRegionServer: ABORTING region server a3serz,16020,1582342923163: Replay of WAL required. Forcing server shutdown
org.apache.hadoop.hbase.DroppedSnapshotException: region: T_BL,x0Ax00x00x00x00x00x00x00x00x00x00x00x00,1572576275632.069e4d877a4ff46f9964ac8bcddb09ef.
        at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushCacheAndCommit(HRegion.java:2509)
        at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushcache(HRegion.java:2186)
        at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushcache(HRegion.java:2148)
        at org.apache.hadoop.hbase.regionserver.HRegion.flushcache(HRegion.java:2039)
        at org.apache.hadoop.hbase.regionserver.HRegion.flush(HRegion.java:1965)
        at org.apache.hadoop.hbase.regionserver.MemStoreFlusher.flushRegion(MemStoreFlusher.java:505)
        at org.apache.hadoop.hbase.regionserver.MemStoreFlusher.flushRegion(MemStoreFlusher.java:475)
        at org.apache.hadoop.hbase.regionserver.MemStoreFlusher.access$900(MemStoreFlusher.java:75)
        at org.apache.hadoop.hbase.regionserver.MemStoreFlusher$FlushHandler.run(MemStoreFlusher.java:263)
        at java.lang.Thread.run(Thread.java:748)
Caused by: org.apache.hadoop.hbase.exceptions.TimeoutIOException: Failed to get sync result after 300000 ms for ringBufferSequence=101793126, WAL system stuck?
        at org.apache.hadoop.hbase.regionserver.wal.SyncFuture.get(SyncFuture.java:174)
        at org.apache.hadoop.hbase.regionserver.wal.FSHLog.blockOnSync(FSHLog.java:1406)
        at org.apache.hadoop.hbase.regionserver.wal.FSHLog.publishSyncThenBlockOnCompletion(FSHLog.java:1400)
        at org.apache.hadoop.hbase.regionserver.wal.FSHLog.sync(FSHLog.java:1512)
        at org.apache.hadoop.hbase.regionserver.wal.WALUtil.writeMarker(WALUtil.java:126)
        at org.apache.hadoop.hbase.regionserver.wal.WALUtil.writeFlushMarker(WALUtil.java:75)
        at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushCacheAndCommit(HRegion.java:2486)

分析：

出现DroppedSnapshotException错误，一般都是由于在进行刷新memstore时，出现了问题，上诉标黄的地方，说明在刷新某一个region级别的memstore时，往hdfs写入数据时间过长，导致regionserver挂掉

hbase memstore 刷新触发条件如下：

HBase会在如下几种情况下触发flush操作，需要注意的是MemStore的最小flush单元是HRegion而不是单个MemStore。可想而知，如果一个HRegion中Memstore过多，每次flush的开销必然会很大，因此我们也建议在进行表设计的时候尽量减少ColumnFamily的个数。

Memstore级别限制：当Region中任意一个MemStore的大小达到了上限（hbase.hregion.memstore.flush.size，默认128MB），会触发Memstore刷新。

Region级别限制：当Region中所有Memstore的大小总和达到了上限（hbase.hregion.memstore.block.multiplier * hbase.hregion.memstore.flush.size，默认 2* 128M = 256M），会触发memstore刷新。

Region Server级别限制：当一个Region Server中所有Memstore的大小总和达到了上限（hbase.regionserver.global.memstore.upperLimit ＊ hbase_heapsize，默认 40%的JVM内存使用量），会触发部分Memstore刷新。Flush顺序是按照Memstore由大到小执行，先Flush Memstore最大的Region，再执行次大的，直至总体Memstore内存使用量低于阈值（hbase.regionserver.global.memstore.lowerLimit ＊ hbase_heapsize，默认 38%的JVM内存使用量）。

当一个Region Server中HLog数量达到上限（可通过参数hbase.regionserver.maxlogs配置）时，系统会选取最早的一个 HLog对应的一个或多个Region进行flush

HBase定期刷新Memstore：默认周期为1小时，确保Memstore不会长时间没有持久化。为避免所有的MemStore在同一时间都进行flush导致的问题，定期的flush操作有20000左右的随机延时。

手动执行flush：用户可以通过shell命令 flush ‘tablename’或者flush ‘region name’分别对一个表或者一个Region进行flush。

上诉的这个问题不仅仅是hbase本身的问题，跟hdfs也相关。

3、查看hbase 写入数据时，datanode节点的Slow状态情况

$ egrep -o "Slow.*?(took|cost)" hadoop-hduser-datanode-a3ser.log.1 |sort |uniq -c
     36 Slow BlockReceiver write data to disk cost
   2743 Slow BlockReceiver write packet to mirror took
      2 Slow flushOrSync took
     35 Slow manageWriterOsCache took
     21 Slow PacketResponder send ack to upstream took

说明：

Slow BlockReceiver write data to disk cost : 表明在将块写入OS缓存或磁盘时存在延迟

Slow BlockReceiver write packet to mirror took ：表明在网络上写入块时有延迟

Slow manageWriterOsCache took : 表明在将块写入OS缓存或磁盘时存在延迟

Slow PacketResponder send ack to upstream took : 母鸡 。。。

Slow flushOrSync took : 表明在将块写入OS缓存或磁盘时存在延迟

4.解决方案

1.设置memstore大小；HloG数量设置；

2.check hdfs 并且修复
3、检查datanode 集群负载，网络情况

4.重启server。

借鉴：http://ddrv.cn/a/258124

a3ster