I don't know if this is already answered in SO but I couldn't find a solution to my problem.
I have an IPython notebook running in a docker container in Google Container Engine, the container is based on this image jupyter/all-spark-notebook
I have also a spark cluster created with google cloud dataproc
Spark master and the notebook are running in different VMs but in the same region and zone.
My problem is that I'm trying to connect to the spark master from the IPython notebook but without success. I use this snippet of code in my python notebook
import pyspark
conf = pyspark.SparkConf()
conf.setMaster("spark://<spark-master-ip or spark-master-hostname>:7077")
I just started working with spark, so I'm sure I'm missing something (authentication, security ...),
What I found over there is connecting a local browser over an SSH tunnel
Somebody already did this kind of set up?
Thank you in advance
Dataproc runs Spark on YARN, so you need to set master to 'yarn-client'. You also need to point Spark at your YARN ResourceManager, which requires a under-documented SparkConf -> Hadoop Configuration conversion. You also have to tell Spark about HDFS on the cluster, so it can stage resources for YARN. You could use Google Cloud Storage instead of HDFS, if you baked The Google Cloud Storage Connector for Hadoop into your image.
Try:
import pyspark
conf = pyspark.SparkConf()
conf.setMaster('yarn-client')
conf.setAppName('My Jupyter Notebook')
# 'spark.hadoop.foo.bar' sets key 'foo.bar' in the Hadoop Configuaration.
conf.set('spark.hadoop.yarn.resourcemanager.address', '<spark-master-hostname>')
conf.set('spark.hadoop.fs.default.name', 'hdfs://<spark-master-hostname>/')
sc = pyspark.SparkContext(conf=conf)
For a more permanent config, you could bake these into a local file 'core-site.xml' as described here, place that in a local directory, and set HADOOP_CONF_DIR to that directory in your environment.
It's also worth noting that while being in the same Zone is important for performance, it is being in the same Network and allowing TCP between internal IP addresses in that network that allows your VMs to communicate. If you are using the default network, then the default-allow-internal firewall rule, should be sufficient.
Hope that helps.
Related
I'm trying to use the following Helm Chart for Spark on Kubernetes
https://github.com/bitnami/charts/tree/main/bitnami/spark
The documentation is of course spotty but I've muddled along. So I have it installed with custom values that assign things like resource limits etc. I'm accessing the master through a NodePort and the WebUI through a port forward. I am NOT using spark-submit, I'm writing Python code to drive the Spark Cluster as follows:
import pyspark
sc = pyspark.SparkContext(appName="Testy", master="spark://<IP>:<PORT>")
This Python code is running locally on my Windows laptop, the Kubernetes cluster is on a separate set of servers. It connects and I can see the app appear in the WebUI but the second it tries to do something I get the following:
WARN TaskSchedulerImpl: Initial job has not accepted any resources; check your cluster UI to ensure that workers are registered and have sufficient resources
The master seems to be in a cycle of removing and launching executors and the 3 workers each just fail to run a launch command. Interestingly the command has the hostname of my laptop in here:
"--driver-url" "spark://CoarseGrainedScheduler#<laptop hostname>:60557"
Got to imagine that's not right. So in this setup where should I be actually running the python code? On the kubernetes cluster? Can I run it locally on my laptop? These details are of course missing from the docs. I'm new to Spark so just looking for the absolute basics. My preferred workflow would be to develop code locally on my laptop then run it on the Kubernetes cluster I have access to.
This is probably a really silly question, but I can't find the answer with Google. I've written a simple pyspark ETL script that reads in a CSV and writes it to Parquet, something like this:
spark = SparkSession.builder.getOrCreate()
sqlContext = SQLContext(spark.sparkContext)
df = sqlContext.read.csv(input_filename)
df.write.parquet(output_path)
To run it, I start up a local Spark cluster in Docker:
$ docker run --network=host jupyter/pyspark-notebook
I run the Python script and it connects to this local Spark cluster and all works as expected.
Now I'd like to run the same script on a remote Spark cluster (AWS EMR). Can I just specify a remote IP address somewhere when initialising the Spark context? Or am I misunderstanding how Spark works?
You can create a spark session by specifying the IP address of the remote master.
spark = SparkSession.builder.master("spark://<ip>:<port>").getOrCreate()
In case of AWS EMR, standalone mode is not supported. You need to use yarn in either client or cluster mode, and point HADOOP_CONF_DIR to a location on your local server where all files from /etc/hadoop/conf are present. Then setup dynamic port forwarding to connect to the EMR cluster. Create a spark session like:
spark = SparkSession.builder.master('yarn').config('spark.submit.deployMode', 'cluster').getOrCreate()
refer https://aws.amazon.com/premiumsupport/knowledge-center/emr-submit-spark-job-remote-cluster/
Currently, I can access the HDFS from inside my application, but I'd also like to - instead of running my local spark - to use Cloudera's spark as it is enabled in Cloudera Manager.
Righ now I have the HDFS defined at core-site.xml, and I run my app as (--master) YARN. Thus I don't need to set the machine address to my HDFS files. In this way, my SPARK job runs locally and not in the "cluster." I don't want that for now. When I try to set --master to [namenode]:[port] it does not connect. I wonder if I'm directing to the correct port, or if I have to map this port at docker container. Or if I'm missing something about Yarn setup.
Additionally, I've been testing SnappyData (Inc) solution as a Spark SQL in-memory database. So my goal is to run snappy JVMs locally, but redirecting spark jobs to the VM cluster. The whole idea here is to test some performance against some Hadoop implementation. This solution is not a final product (if snappy is local, and spark is "really" remote, I believe it won't be efficient - but in this scenario, I would bring snappy JVMs to the same cluster..)
Thanks in advance!
I am attempting to leverage a Hadoop Spark Cluster in order to batch load a graph into Titan using the SparkGraphComputer and BulkLoaderVertex program, as specified here. This requires setting the spark configuration in a properties file, telling Titan where Spark is located, where to read the graph input from, where to store its output, etc.
The problem is that all of the examples seem to specify a local spark cluster through the option:
spark.master=local[*]
I, however, want to run this job on a remote Spark cluster which is on the same VNet as the VM where the titan instance is hosted. From what I have read, it seems that this can be accomplished by setting
spark.master=<spark_master_IP>:7077
This is giving me the error that all Spark masters are unresponsive, which disallows me from sending the job to the spark cluster to distribute the batch loading computations.
For reference, I am using Titan 1.0.0 and a Spark 1.6.4 cluster, which are both hosted on the same VNet. Spark is being managed by yarn, which also may be contributing to this difficulty.
Any sort of help/reference would be appreciated. I am sure that I have the correct IP for the spark master, and that I am using the right gremlin commands to accomplish bulk loading through the SparkGraphComputer. What I am not sure about is how to properly configure the Hadoop properties file in order to get Titan to communicate with a remote Spark cluster over a VNet.
I have created a Spark EMR cluster. I would like to execute jobs either on my localhost or EMR cluster.
Assuming I run spark-shell on my local computer how can I tell it to connect to the Spark EMR cluster, what would be the exact configuration options and/or commands to run.
It looks like others have also failed at this and ended up running the Spark driver on EMR, but then making use of e.g. Zeppelin or Jupyter running on EMR.
Setting up our own machines as spark drivers that connected to the core nodes on EMR would have been ideal. Unfortunately, this was impossible to do and we forfeited after trying many configuration changes. The driver would start up and then keep waiting unsuccessfully, trying to connect to the slaves.
Most of our Spark development is on pyspark using Jupyter Notebook as our IDE. Since we had to run Jupyter from the master node, we couldn’t risk losing our work if the cluster were to go down. So, we created an EBS volume and attached it to the master node and placed all of our work on this volume. [...]
source
Note: If you go down this route, I would consider using S3 for storing notebooks, then you don't have to manage EBS volumes.
One way of doing this is to add your spark job as an EMR step to your EMR cluster. For this, you need AWS CLI installed on your local computer
(see here for installation guide), and your jar file on s3.
Once you have aws cli, assuming your spark class to run is com.company.my.MySparkJob and your jar file is located on s3 at s3://hadi/my-project-0.1.jar, you can run the following command from your terminal:
aws emr add-steps --cluster-id j-************* --steps Type=spark,Name=My_Spark_Job,Args=[-class,com.company.my.MySparkJob,s3://hadi/my-project-0.1.jar],ActionOnFailure=CONTINUE