> On July 7, 2017, 5 p.m., Jonathan Hurley wrote: > > I don't know enough about HBase internals - is removing the write-ahead log > > safe - especially during a complex upgrade? Isn't it like a last line of > > defense in case something goes very wrong? > > Di Li wrote: > yes, this is indeed the last resort as it's either removing the wal dir > or having HBase master dead in the water. HBase JIRA > https://issues.apache.org/jira/browse/HBASE-14783 is for the issue, but it is > still open. IOP uses this remove wal dir approach to deal with downgrading > from 4.2.5 to 4.1. The patch here is a port of the internal logic for the > migration use case. > > It's either this patch (automated during EU) or a manual step post EU. > > Di Li wrote: > Forgot to mention that I am OK either way A) automated during EU or B) a > manual step post EU. > > Jonathan Hurley wrote: > Yeah - I'm just wondering if we should open an item for fixing this > correctly? Is there any other way aside from deleting it? Will Hbase provide > a script to run on startup which fixes it?
I don't really know any other ways, I am afraid. The feedback from our HBase team was the removal of the wal dir thus the logic shown in the patch. I would also prefer a fix on the HBase side to deal with the issue. - Di ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/60691/#review179888 ----------------------------------------------------------- On July 7, 2017, 7:34 p.m., Di Li wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/60691/ > ----------------------------------------------------------- > > (Updated July 7, 2017, 7:34 p.m.) > > > Review request for Ambari, Dmytro Grinenko, Di Li, Dmitro Lisnichenko, > Jonathan Hurley, Sumit Mohanty, Sid Wagle, and Tim Thorpe. > > > Bugs: AMBARI-21420 > https://issues.apache.org/jira/browse/AMBARI-21420 > > > Repository: ambari > > > Description > ------- > > HBase in IOP 4.2.5 is of higher version than the one shipped in HDP 2.6. > HBase master appeared as if it had started according to the Ambari action log > but later crashed with error: > 2017-07-05 09:28:08,952 FATAL [hats1:16000.activeMasterManager] > master.HMaster: Failed to become active masterjava.io.IOException: The > procedure class org.apache.hadoop.hbase.master.procedure.ServerCrashProcedure > must be accessible and have an empty constructorat > org.apache.hadoop.hbase.procedure2.Procedure.newInstance(Procedure.java:663)at > org.apache.hadoop.hbase.procedure2.Procedure.convert(Procedure.java:789)at > org.apache.hadoop.hbase.procedure2.store.wal.ProcedureWALFormatReader.read(ProcedureWALFormatReader.java:111)at > > org.apache.hadoop.hbase.procedure2.store.wal.ProcedureWALFormat.load(ProcedureWALFormat.java:81)at > > org.apache.hadoop.hbase.procedure2.store.wal.WALProcedureStore.load(WALProcedureStore.java:374)at > > org.apache.hadoop.hbase.procedure2.ProcedureExecutor.load(ProcedureExecutor.java:297)at > > org.apache.hadoop.hbase.procedure2.ProcedureExecutor.start(ProcedureExecutor.java:464)at > > org.apache.hadoop.hbase.master.HMaster.startProcedureExecutor(HMaster.java:1297)at > org.apach e.hadoop.hbase.master.HMaster.startServiceThreads(HMaster.java:1196) > Attach one or more files to this issue > Attachments > > > Diffs > ----- > > > ambari-server/src/main/resources/common-services/HBASE/0.96.0.2.0/package/scripts/hbase_service.py > a1003dc > > ambari-server/src/main/resources/common-services/HBASE/0.96.0.2.0/package/scripts/params_linux.py > 43c6036 > > ambari-server/src/main/resources/stacks/BigInsights/4.2.5/upgrades/config-upgrade.xml > 6597055 > > ambari-server/src/main/resources/stacks/BigInsights/4.2.5/upgrades/nonrolling-upgrade-to-hdp-2.6.xml > af87794 > > ambari-server/src/main/resources/stacks/BigInsights/4.2/upgrades/config-upgrade.xml > 6597055 > > ambari-server/src/main/resources/stacks/BigInsights/4.2/upgrades/nonrolling-upgrade-to-hdp-2.6.xml > af87794 > > > Diff: https://reviews.apache.org/r/60691/diff/3/ > > > Testing > ------- > > will be tested in the integration tests. > > > Thanks, > > Di Li > >
