Spark installation on CDH 5.2

[#1]

Did a clean installaiton of cdh 5.2 ( on AWS , 7 nodes doing a complete install , accepting all components ) , Spark would only install a history server ( and allow me to add gateway servers ) there seems to be no way to add executor nodes. In CDH 5.1 spark executors were installed on the data nodes by default , I would expect the same behaviour in 5.2. Is there a work around

[#2]

Hi,

In 5.2, Spark on YARN is now the default. This uses YARN NodeManagers for execution and does not require assigning Spark executor nodes.

Spark on YARN is strongly preferred over the old Spark, which is why we made this change. If you'd really like to use the old Spark, then you can add it to your cluster after the initial setup, just like how you add any other service to a cluster. You can't add the old standalone Spark in the cluster setup wizard.

Thanks,

Darren

[#3]

Hi Darren,

If Spark on YARN, but I can't find the job status from YARN Manager. No history in YARN manager is generated. Actually the job ran well.

URL:

HOST:8088/cluster/apps

Any ideas? how can we get the status of all Spark jobs ?

thanks,

Tod

[#4]

Hi Tod,

Did you try looking at the Spark history server? You can find a link to that URL from the status page of the Spark service in the CM UI.

Thanks,
Darren

[#5]

History Server

• Event Log Location: hdfs://lalana-490120.slc01.dev.ebayc3.com:8020/user/spark/applicationHistory

No Completed Applications Found

this is my history server page. No records there..

[#6]

How are you running your spark job? Are you sure it's configured to talk to the right spark service?

Is there any chance you have a standalone spark (as opposed to spark on yarn) set up as well, and you're accidentally submitting jobs there?

[#7]

in master machine ran the following spark submit command

/opt/cloudera/parcels/CDH/lib/spark/bin/spark-submit  {JOB_FILE}

it works in CDH5.1, but when I re-install to CDH5.2, the spark history gone..

[#8]

Did you remember to upload the spark jar after upgrading CDH?

While your job is running, does it show up in the yarn web UI?

Are there any interesting error messages in the spark history server? In the yarn logs?

[#9]

Not upgrade, I installed CDH5.2 in a clean environment.

All Applications
About Apache Hadoop
Cluster About Nodes Applications NEW NEW_SAVING SUBMITTED ACCEPTED RUNNING FINISHED FAILED KILLED Scheduler Tools	Cluster MetricsApps Submitted Apps Pending Apps Running Apps Completed Containers Running Memory Used Memory Total Memory Reserved VCores Used VCores Total VCores Reserved Active Nodes Decommissioned Nodes Lost Nodes Unhealthy Nodes Rebooted Nodes 0 0 0 0 0 0 B 9.11 GB 0 B 0 8 0 4 1 0 0 0 User Metrics for dr.whoApps Submitted Apps Pending Apps Running Apps Completed Containers Running Containers Pending Containers Reserved Memory Used Memory Pending Memory Reserved VCores Used VCores Pending VCores Reserved 0 0 0 0 0 0 0 0 B 0 B 0 B 0 0 0 Show20406080100entries Search: IDUserNameApplication TypeQueueStartTimeFinishTimeStateFinalStatusProgressTracking UI No data available in table Showing 0 to 0 of 0 entries FirstPreviousNextLast

this is YARN UI, no executed or executing jobs here..

I will check logs ..

[#10]

post image here.