[ https://issues.apache.org/jira/browse/HBASE-18030?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16005974#comment-16005974 ]
Abhishek Kumar commented on HBASE-18030: ---------------------------------------- My table is having two column family (weekly, daily) and using TTL. For me, these are very basic features for tables with increments. The weekly column family has small files and I don't see them compacting (total 5 HFiles). This is not the case with daily column family, which has just 1 HFile. hbase(main):016:0> get 'table-name', 'rowkey', 'daily:a' 0 row(s) in 0.0100 seconds hbase(main):017:0> get 'table-name', 'rowkey', 'weekly:a' ERROR: java.io.IOException: Invalid currTagsLen -32723. Block offset: 0, block length: 69222, position: 3447 (without header). at org.apache.hadoop.hbase.ipc.RpcServer.call(RpcServer.java:2172) A cron runs at my application service (3 of them) which flushes all the increments it accumulated, one by one to hbase. Nothing suspicious about the whole flow or any patterns I have observed. I am trying to remove TTL and moving to just 1 column family to see if it is fixing the issue. > Negative tag length for hbase block/hfile > ----------------------------------------- > > Key: HBASE-18030 > URL: https://issues.apache.org/jira/browse/HBASE-18030 > Project: HBase > Issue Type: Bug > Components: hbase, regionserver > Affects Versions: 1.1.2 > Reporter: Abhishek Kumar > > 2017-04-29 14:24:14,135 ERROR > [B.fifo.QRpcServer.handler=49,queue=1,port=16020] ipc.RpcServer: Unexpected > throwable object java.lang.IllegalStateException: Invalid currTagsLen -32712. > Block offset: 3707853, block length: 72841, position: 0 (without header). at > org.apache.hadoop.hbase.io.hfile.HFileReaderV3$ScannerV3.checkTagsLen(HFileReaderV3.java:226) > I am not not using any hbase tags feature. > The Increment operation from the application side is triggering this error. > The same is happening when scanner is run on this table. It feels that one or > more particular HFile block is corrupt (with negative tagLength). > hbase(main):007:0> scan 'table-name', {LIMIT=>1,STARTROW=>'ad:event_count:a'} > Returning the result > hbase(main):008:0> scan 'table-name', {LIMIT=>1,STARTROW=>'ad:event_count:b'} > ROW COLUMN+CELL > > > ERROR: java.io.IOException: java.lang.IllegalStateException: Invalid > currTagsLen -32701. Block offset: 272031, block length: 72441, position: > 32487 (without header). > at > org.apache.hadoop.hbase.regionserver.HRegion$RegionScannerImpl.handleException(HRegion.java:5607) > at > org.apache.hadoop.hbase.regionserver.HRegion$RegionScannerImpl.<init>(HRegion.java:5579) > at > org.apache.hadoop.hbase.regionserver.HRegion.instantiateRegionScanner(HRegion.java:2627) > at > org.apache.hadoop.hbase.regionserver.HRegion.getScanner(HRegion.java:2613) > at > org.apache.hadoop.hbase.regionserver.HRegion.getScanner(HRegion.java:2595) > at > org.apache.hadoop.hbase.regionserver.RSRpcServices.scan(RSRpcServices.java:2282) > at > org.apache.hadoop.hbase.protobuf.generated.ClientProtos$ClientService$2.callBlockingMethod(ClientProtos.java:32295) -- This message was sent by Atlassian JIRA (v6.3.15#6346)