[
https://issues.apache.org/jira/browse/MAPREDUCE-6922?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jason Lowe updated MAPREDUCE-6922:
----------------------------------
Target Version/s: 2.9.0
Looks like FrameworkCounterGroup needs to be updated to protect itself from
out-of-bounds counter indices coming from the task report. Until that has been
rolled out as a baseline we can't ever safely add new counters without breaking
this code (or using a distributed cache deploy for MapReduce, see below).
Ultimately moving to protocol buffers would be even better, but that's not
something we could accomplish until a major release like Hadoop 3.0.
I should point out that jobs would not be susceptible to this failure if they
deployed MapReduce via HDFS rather than picking it up from the nodes. See
http://hadoop.apache.org/docs/current/hadoop-mapreduce-client/hadoop-mapreduce-client-core/DistributedCacheDeploy.html.
That deployment method has the benefit that a job always runs with a
consistent version of MapReduce, and therefore changes like this _would_ be OK
during rolling upgrade. Jobs would either run entirely with the old MapReduce
version or the new one instead of a hodgepodge of both which could lead to
errors like this.
If distributed cache deploy is not sufficient and this still needs to be
reverted then normally we would simply revert MAPREDUCE-6829 from branch-2,
remove 2.9.0 from the fix version of that JIRA, and mark this as a duplicate of
that JIRA. 2.9.0 never shipped, so we don't need a tracking JIRA to note
something was removed since we never shipped a 2.x release saying it was added.
> MapReduce jobs may fail during rolling upgrade due to MAPREDUCE-6829
> --------------------------------------------------------------------
>
> Key: MAPREDUCE-6922
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-6922
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Reporter: Miklos Szegedi
> Assignee: Miklos Szegedi
> Priority: Blocker
> Attachments: YARN-6922.branch-2.000.patch
>
>
> MAPREDUCE-6829 should be reverted from branch-2 because rolling upgrade fails.
> {code}
> 2017-06-08 17:43:37,173 WARN [Socket Reader #1 for port 41187]
> org.apache.hadoop.ipc.Server: Unable to read call parameters for client
> 10.17.242.22on connection protocol
> org.apache.hadoop.mapred.TaskUmbilicalProtocol for rpcKind RPC_WRITABLE
> java.lang.ArrayIndexOutOfBoundsException: 23
> at
> org.apache.hadoop.mapreduce.counters.FrameworkCounterGroup.readFields(FrameworkCounterGroup.java:261)
> at org.apache.hadoop.mapred.Counters$Group.readFields(Counters.java:324)
> at
> org.apache.hadoop.mapreduce.counters.AbstractCounters.readFields(AbstractCounters.java:306)
> at org.apache.hadoop.mapred.TaskStatus.readFields(TaskStatus.java:489)
> at
> org.apache.hadoop.mapred.MapTaskStatus.readFields(MapTaskStatus.java:88)
> at
> org.apache.hadoop.io.ObjectWritable.readObject(ObjectWritable.java:285)
> at
> org.apache.hadoop.ipc.WritableRpcEngine$Invocation.readFields(WritableRpcEngine.java:160)
> at
> org.apache.hadoop.ipc.Server$Connection.processRpcRequest(Server.java:1909)
> at
> org.apache.hadoop.ipc.Server$Connection.processOneRpc(Server.java:1841)
> at
> org.apache.hadoop.ipc.Server$Connection.readAndProcess(Server.java:1600)
> at org.apache.hadoop.ipc.Server$Listener.doRead(Server.java:820)
> at
> org.apache.hadoop.ipc.Server$Listener$Reader.doRunLoop(Server.java:693)
> at org.apache.hadoop.ipc.Server$Listener$Reader.run(Server.java:664)
> {code}
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]