whirr-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Tom White <...@cloudera.com>
Subject Re: Whirr and CDH3
Date Mon, 29 Nov 2010 18:48:12 GMT
On Thu, Nov 25, 2010 at 12:23 AM, Bo Shi <bs1984@gmail.com> wrote:
> Hello,
>
> I have created a custom AMI derived from the Canonical Ubuntu 10.04 image
> which has some large EBS volumes attached and mounted at /mnt/hadoop and
> /mnt/hadoop2 by default.  I've been trying to use whirr to set up a cluster
> using this custom AMI and using CDH3beta without much success; I've run into
> two problems which I think I can characterize but I'm at a loss about how to
> solve them.
>
>
> Issue 1
> =======
>
> I'm using the whirr source release with WHIRR-137 as the last commit and I have
> run into the same issue encountered below where whirr is not able to log into
> the nodes instantiated by jclouds:
>
> https://groups.google.com/a/cloudera.org/group/cdh-user/browse_thread/thread/73735107db0cbc05/7f3858dc7abfb80f?lnk=raot&fwc=1
>
>
>
> The root of the issue is that jclouds appears not to recognize my Ubuntu
> derived image as part of the Ubuntu OS family (see "operatingSystem=[...,
> family=unrecognized,...]" in the log sample below).
>
> DEBUG [jclouds.compute] (main) <<   matched
> image([id=us-east-1/ami-f74eb99e, name=null,
> operatingSystem=[name=null, family=unrecognized, version=,
> arch=paravirtual, is64Bit=false,
> description=098149836275/custom-ubuntu-node-small],
> description=Small/Medium Ubuntu 10.04 Cluster Node 2x200 GB EBS
> Attached, version=small, location=[id=us-east-1, scope=REGION,
> description=us-east-1, parent=ec2]])
> INFO  [org.apache.whirr.service.hadoop.HadoopService] (main) Starting
> master node
>
> ...
>
> java.util.concurrent.ExecutionException: org.jclouds.ssh.SshException:
> root@50.16.75.60:22: Error connecting to sftp.
>
>
> If I boot up the canonical Ubuntu 10.04 EC2 image, jclouds appears to recognize
> it as an Ubuntu image and knows to log into the instance as the "ubuntu" user
> and to run scripts as sudo from there (see "family=ubuntu" in the log sample
> below).  Unfortunately, thus far I haven't been able to find anything in the
> jclouds or amazon/ec2 documentation to hint at what i need to do to get jclouds
> to recognize my ubuntu derived AMI as an ubuntu image so it tries to log in
> as root and subsequently fails.

I believe this happens in jclouds code. Adrian, do you know how to control this?

>
>
> DEBUG [jclouds.compute] (main) <<   matched
> image([id=us-east-1/ami-480df921, name=null,
> operatingSystem=[name=null, family=ubuntu, version=10.04,
> arch=paravirtual, is64Bit=false,
> description=099720109477/ebs/ubuntu-images/ubuntu-lucid-10.04-i386-server-20101020],
> description=099720109477/ebs/ubuntu-images/ubuntu-lucid-10.04-i386-server-20101020,
> version=20101020, location=[id=us-east-1, scope=REGION,
> description=us-east-1, parent=ec2]])
> 2010-11-24 23:34:14,771 INFO
> [org.apache.whirr.service.hadoop.HadoopService] (main) Starting master
> node
>
>
> Issue 2
> =======
>
> For Ubuntu AMI's which can properly instantiate nodes, the installation script
> "cloudera/cdh/install" seems to fail since the hadoop user is not created,
> however I was under the impression that the scripts handled user creation::
>
>
> DEBUG [jclouds.compute] (user thread 1) << stdout from runscript as
> ubuntu@184.73.17.198
> ^M
> Setting up hadoop-0.20 (0.20.2+737-1~lucid-cdh3b3) ...^M
> update-alternatives: using /etc/hadoop-0.20/conf.empty to provide
> /etc/hadoop-0.20/conf (hadoop-0.20-conf) in auto mode.^M
> update-alternatives: using /usr/bin/hadoop-0.20 to provide
> /usr/bin/hadoop (hadoop-default) in auto mode.^M
> ^M
> Setting up hadoop-0.20-native (0.20.2+737-1~lucid-cdh3b3) ...^M
> ^M
> Processing triggers for libc-bin ...^M
> ldconfig deferred processing now taking place^M
> update-alternatives: using /etc/hadoop-0.20/conf.dist to provide
> /etc/hadoop-0.20/conf (hadoop-0.20-conf) in auto mode.^M
>
> 2010-11-24 23:36:57,352 DEBUG [jclouds.compute] (user thread 1) <<
> stderr from runscript as ubuntu@184.73.17.198
> + DFS_DATA_DIR=/mnt/hadoop/hdfs/data,/mnt/hadoop2^M
> + MAPRED_LOCAL_DIR=/mnt/hadoop/mapred/local^M
> + MAX_MAP_TASKS=2^M
> + MAX_REDUCE_TASKS=1^M
> + CHILD_OPTS=-Xmx550m^M
> + CHILD_ULIMIT=1126400^M
> + TMP_DIR='/mnt/tmp/hadoop-${user.name}'^M
> + mkdir -p /mnt/hadoop^M
> + chown hadoop:hadoop /mnt/hadoop^M
> chown: invalid user: `hadoop:hadoop'^M

I think the scripts for CDH in trunk are not working correctly. I just
noticed the same thing when working on
https://issues.apache.org/jira/browse/WHIRR-87, and will post a fix as
a part of that issue.

Cheers
Tom

>
>
>
>
>
>
>
> Here is my test configuration:
>
>
>   whirr.service-name=hadoop
>   whirr.cluster-name=testcluster
>   whirr.instance-templates=1 jt+nn,1 dn+tt
>
>   whirr.provider=ec2
>   whirr.identity=***
>   whirr.credential=***
>   whirr.private-key-file=${sys:user.home}/.ssh/id_rsa
>
>   # Uploaded to the following
>
>   # FFFFFFFffffffffffuuuuuuuuuuu
>   # Full path is the combination of the two below
>   whirr.hadoop-install-runurl=cloudera/cdh/install
>   whirr.run-url-base=http://production.tinycorp.clusters.s3.amazonaws.com/
>
>   # Confirm the Lucid (10.04) kernel is up to date wrt below:
>   # https://bugs.launchpad.net/ubuntu-on-ec2/+bug/574910
>
>   # Custom image with a bunch of EBS
>   # whirr.image-id=us-east-1/ami-f74eb99e
>
>   # Default image is Amazon Linux AMI which is based on RHEL 5
>   # Canonical Ubuntu 10.04 32-bit EBS ASMI
>   whirr.image-id=us-east-1/ami-480df921
>   # CDH Ubuntu 8.10
>   # whirr.image-id=us-east-1/ami-ed59bf84
>   # https://issues.apache.org/jira/browse/WHIRR-137
>   jclouds.ec2.ami-owners=098149836275,726089167552,099720109477
>

Mime
View raw message