crunch-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Josh Wills <jwi...@cloudera.com>
Subject Re: LeaseExpiredExceptions and temp side effect files
Date Fri, 14 Aug 2015 22:26:13 GMT
Hey Everett,

Initial thought-- there are lots of reasons for lease expired exceptions,
and their usually more symptomatic of other problems in the pipeline. Are
you sure none of the jobs in the Crunch pipeline on the non-SSD instances
are failing for some other reason? I'd be surprised if no other errors
showed up in the app master, although there are reports of some weirdness
around LeaseExpireds when writing to S3-- but you're not doing that here,
right?

J

On Fri, Aug 14, 2015 at 2:10 PM, Everett Anderson <everett@nuna.com> wrote:

> Hi,
>
> I recently started trying to run our Crunch pipeline on more data and have
> been trying out different AWS instance types in anticipation of our storage
> and compute needs.
>
> I was using EMR 3.8 (so Hadoop 2.4.0) with Crunch 0.12 (patched with the
> CRUNCH-553 <https://issues.apache.org/jira/browse/CRUNCH-553> fix).
>
> Our pipeline finishes fine in these cluster configurations:
>
>    - 50 c3.4xlarge Core, 0 Task
>    - 10 c3.8xlarge Core, 0 Task
>    - 25 c3.8xlarge Core, 0 Task
>
> However, it always fails on the same data when using 10 cc2.8xlarge Core
> instances.
>
> The biggest obvious hardware difference is that the cc2.8xlarges use hard
> disks instead of SSDs.
>
> While it's a little hard to track down the exact originating failure, I
> think it's from errors like:
>
> 2015-08-13 21:34:38,379 ERROR [IPC Server handler 24 on 45711]
> org.apache.hadoop.mapred.TaskAttemptListenerImpl: Task:
> attempt_1439499407003_0028_r_000153_1 - exited :
> org.apache.crunch.CrunchRuntimeException:
> org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.hdfs.server.namenode.LeaseExpiredException):
> No lease on
> /tmp/crunch-970849245/p662/output/_temporary/1/_temporary/attempt_1439499407003_out7_0028_r_000153_1/out7-r-00153:
> File does not exist. Holder
> DFSClient_attempt_1439499407003_0028_r_000153_1_609888542_1 does not have
> any open files.
>
> Those paths look like these side effect files
> <https://hadoop.apache.org/docs/r2.4.1/api/org/apache/hadoop/mapred/FileOutputFormat.html#getWorkOutputPath(org.apache.hadoop.mapred.JobConf)>
> .
>
> Would Crunch have generated applications that depend on side effect paths
> as input across MapReduce applications and something in HDFS is cleaning up
> those paths, unaware of the higher level dependencies? AWS configures
> Hadoop differently for each instance type, and might have more aggressive
> cleanup settings on HDs, though this is very uninformed hypothesis.
>
> A sample full log is attached.
>
> Thanks for any guidance!
>
> - Everett
>
>
> *DISCLAIMER:* The contents of this email, including any attachments, may
> contain information that is confidential, proprietary in nature, protected
> health information (PHI), or otherwise protected by law from disclosure,
> and is solely for the use of the intended recipient(s). If you are not the
> intended recipient, you are hereby notified that any use, disclosure or
> copying of this email, including any attachments, is unauthorized and
> strictly prohibited. If you have received this email in error, please
> notify the sender of this email. Please delete this and all copies of this
> email from your system. Any opinions either expressed or implied in this
> email and all attachments, are those of its author only, and do not
> necessarily reflect those of Nuna Health, Inc.




-- 
Director of Data Science
Cloudera <http://www.cloudera.com>
Twitter: @josh_wills <http://twitter.com/josh_wills>

Mime
View raw message