I'm trying to run a simple hello world python code on Serverless pyspark on GCP using gcloud (from local windows machine).
if __name__ == '__main__':
print("Hello")
This always results in the error
=========== Cloud Dataproc Agent Error ===========
java.lang.IllegalArgumentException: Illegal character in path at index 38: gs://my-bucket/dependencies\hello.py
at java.base/java.net.URI.create(URI.java:883)
at com.google.cloud.hadoop.services.agent.job.handler.AbstractJobHandler.registerResourceForDownload(AbstractJobHandler.java:592)
The gcloud command:
gcloud dataproc batches submit pyspark hello.py --batch=hello-batch-5 --deps-bucket=my-bucket --region=us-central1
On further analysis, I found that gcloud puts hello.py file in dependencies\hello.py under folder {deps-bucket} and Java considers backward slash '\' as illegal.
Has anyone encountered a similar situation?
As #Ronak mentioned, Can you double check the bucket name ? I have replicated your task, and simply copied your code to my Google Cloud shell. and it ran just fine. for your next run can you delete the dependencies folder and run the batch job again ?
See my replication here:
Dependencies path created after running the job:
Related
It appears that you cannot deploy a Gen2 cloud function using gcloud from a cloud source repo unless it is at the top level.
Here's a sample redacted deploy command for a gen 1 python function that works:
gcloud beta functions deploy funcname --source https://source.developers.google.com/projects/projectname/repos/reponame/moveable-aliases/main/paths/pathname --runtime python310 --trigger-http --project=projectname
if you add the -gen2 flag, it fails because it can't find main.py. Error is:
OperationError: code=3, message=Build failed with status: FAILURE and message: missing main.py and GOOGLE_FUNCTION_SOURCE not specified. Either create the function in main.py or specify GOOGLE_FUNCTION_SOURCE to point to the file that contains the function.
If you add main.py to the root of the repo and run the same command, it finds main.py, which indicates to me that it isn't honoring the paths.
There is an additional problem which doesn't matter unless the first one is fixed, which is that if pathname is below the top level (folder/subfolder) gcloud sees that as a syntax error when the gen2 flag is set, but not without it.
Is there any way around this? It is very inconvenient.
Answering as community wiki.As per above comments
There is a bug raised for this at issue tracker. Which is still open further progress can be tracked there.
I have a pyspark code stored both on the master node of an AWS EMR cluster and in an s3 bucket that fetches over 140M rows from a MySQL database and stores the sum of a column back in the log files on s3.
When I spark-submit the pyspark code on the master node, the job gets completed successfully and the output is stored in the log files on the S3 bucket.
However, when I spark-submit the pyspark code on the S3 bucket using these- (using the below commands on the terminal after SSH-ing to the master node)
spark-submit --master yarn --deploy-mode cluster --py-files s3://bucket_name/my_script.py
This returns a Error: Missing application resource. error.
spark_submit s3://bucket_name/my_script.py
This shows :
20/07/02 11:26:23 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Exception in thread "main" java.lang.RuntimeException: java.lang.ClassNotFoundException: Class com.amazon.ws.emr.hadoop.fs.EmrFileSystem not found
at org.apache.hadoop.conf.Configuration.getClass(Configuration.java:2369)
at org.apache.hadoop.fs.FileSystem.getFileSystemClass(FileSystem.java:2840)
at org.apache.hadoop.fs.FileSystem.createFileSystem(FileSystem.java:2857)
at org.apache.hadoop.fs.FileSystem.access$200(FileSystem.java:99)
at org.apache.hadoop.fs.FileSystem$Cache.getInternal(FileSystem.java:2896)
at org.apache.hadoop.fs.FileSystem$Cache.get(FileSystem.java:2878)
at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:392)
at org.apache.spark.util.Utils$.getHadoopFileSystem(Utils.scala:1911)
at org.apache.spark.util.Utils$.doFetchFile(Utils.scala:766)
at org.apache.spark.deploy.DependencyUtils$.downloadFile(DependencyUtils.scala:137)
at org.apache.spark.deploy.SparkSubmit$$anonfun$prepareSubmitEnvironment$7.apply(SparkSubmit.scala:356)
at org.apache.spark.deploy.SparkSubmit$$anonfun$prepareSubmitEnvironment$7.apply(SparkSubmit.scala:356)
at scala.Option.map(Option.scala:146)
at org.apache.spark.deploy.SparkSubmit.prepareSubmitEnvironment(SparkSubmit.scala:355)
at org.apache.spark.deploy.SparkSubmit.org$apache$spark$deploy$SparkSubmit$$runMain(SparkSubmit.scala:782)
at org.apache.spark.deploy.SparkSubmit.doRunMain$1(SparkSubmit.scala:161)
at org.apache.spark.deploy.SparkSubmit.submit(SparkSubmit.scala:184)
at org.apache.spark.deploy.SparkSubmit.doSubmit(SparkSubmit.scala:86)
at org.apache.spark.deploy.SparkSubmit$$anon$2.doSubmit(SparkSubmit.scala:928)
at org.apache.spark.deploy.SparkSubmit$.main(SparkSubmit.scala:937)
at org.apache.spark.deploy.SparkSubmit.main(SparkSubmit.scala)
Caused by: java.lang.ClassNotFoundException: Class com.amazon.ws.emr.hadoop.fs.EmrFileSystem not found
at org.apache.hadoop.conf.Configuration.getClassByName(Configuration.java:2273)
at org.apache.hadoop.conf.Configuration.getClass(Configuration.java:2367)
... 20 more
I read about having to add a Spark Step on the AWS EMR cluster to submit a pyspark code stored on the S3.
Am I correct in saying that I would need to create a step in order to submit my pyspark job stored on the S3?
In the 'Add Step' window that pops up on the AWS Console, in the 'Application location' field, it says that I'll have to type in the location to the JAR file. What JAR file are they referring to? Does my pyspark script have to be packaged into a JAR file and how do I do that or do I mention the path to my pyspark script?
In the 'Add Step' window that pops up on the AWS Console, in the Spark-submit options, how do I know what to write for the --class parameter? Can I leave this field empty? If no, why not?
I have gone through the AWS EMR documentation. I have so many questions because I dived nose-down into the problem and only researched when an error popped up.
Your spark submit should be this.
spark-submit --master yarn --deploy-mode cluster s3://bucket_name/my_script.py
--py-files is used if you want to pass the python dependency modules, not the application code.
When you are adding step in EMR to run spark job, jar location is your python file path. i.e. s3://bucket_name/my_script.py
No its not mandatory to use STEP to submit spark job.
You can also use spark-submit
To submit a pyspark script using STEP please refer aws doc and stackoverflow
For problem 1:
By default spark will use python2.
You need to add 2 config
Go to $SPARK_HOME/conf/spark-env.sh and add
export PYSPARK_PYTHON=/usr/bin/python3
export PYSPARK_DRIVER_PYTHON=/usr/bin/python3
Note: If you have any custom bundle add that using --py-files
For problem 2:
A hadoop-assembly jar exists on /usr/share/aws/emr/emrfs/lib/. That contains com.amazon.ws.emr.hadoop.fs.EmrFileSystem.
You need to add this to your classpath.
A better option to me is to create a symbolic link of hadoop-assembly jar to HADOOP_HOME (/usr/lib/hadoop) in your bootstrap action.
I have created a basic EMR cluster in AWS, and I'm trying to use the Jupyter Notebooks provided through the AWS Console. Launching the notebooks seems to work fine, and I'm also able to run basic python code in notebooks started with the pyspark kernel. Two variables are set up in the notebook: spark is a SparkSession instance, and sc is a SparkContext instance. Displaying sc yields <SparkContext master=yarn appName=livy-session-0> (the output can of course vary slightly depending on the session).
The problem arises once I perform operations that actually hit the spark machinery. For example:
sc.parallelize(list(range(10))).map(lambda x: x**2).collect()
I am no spark expert, but I believe this code should distribute the integers from 0 to 9 across the cluster, square them, and return the results in a list. Instead, I get a lengthy stack trace, mostly from the JVM, but also some python components. I believe the central part of the stack trace is the following:
py4j.protocol.Py4JJavaError: An error occurred while calling z:org.apache.spark.api.python.PythonRDD.collectAndServe.
: org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 in stage 4.0 failed 4 times, most recent failure: Lost task 0.3 in stage 4.0 (TID 116, ip-XXXXXXXXXXXXX.eu-west-1.compute.internal, executor 17): java.lang.RuntimeException: Failed to run command: /usr/bin/virtualenv -p python3 --system-site-packages virtualenv_application_1586243436143_0002_0
The full stack trace is here.
A bit of digging in the AWS portal led me to log output from the nodes. stdout from one of the nodes includes the following:
The path python3 (from --python=python3) does not exist
I tried running the /usr/bin/virtualenv command on the master node manually (after logging in through), and that worked fine, but the error is of course still present after I did that.
While this error occurs most of the time, I was able to get this working in one session, where I could run several operations against the spark cluster as I was expecting.
Technical information on the cluster setup:
emr-6.0.0
Applications installed are "Ganglia 3.7.2, Spark 2.4.4, Zeppelin 0.9.0, Livy 0.6.0, JupyterHub 1.0.0, Hive 3.1.2". Hadoop is also included.
3 nodes (one of them as master), all r5a.2xlarge.
Any ideas what I'm doing wrong? Note that I am completely new to EMR and Spark.
Edit: Added the stdout log and information about running the virtualenv command manually on the master node through ssh.
I have switched to using emr-5.29.0, which seems to resolve the problem. Perhaps this is an issue with emr-6.0.0? In any case, I have a functional workaround.
The issue for me was that the virtualenv was being made on the executors with a python path that didn't exist. Pointing the executors to the right one did the job for me:
"spark.pyspark.python": "/usr/bin/python3.7"
Here is how I reconfiged the spark app at the beginning of the notebook:
{"conf":{"spark.pyspark.python": "/usr/bin/python3.7",
"spark.pyspark.virtualenv.enabled": "true",
"spark.pyspark.virtualenv.type": "native",
"spark.pyspark.virtualenv.bin.path":"/usr/bin/virtualenv"}
}
I'm running emr-5.12.0, with Amazon 2.8.3, Hive 2.3.2, Hue 4.1.0, Livy 0.4.0, Spark 2.2.1 and Zeppelin 0.7.3 on 1 m4.large as my master node and 1 m4.large as core node.
I am trying to execute a bootstrap action that configures some parts of the cluster. One of these includes the line:
sudo sed -i '/zeppelin.pyspark.python/c\ \"zepplin.pyspark.python\" : \"python3\",' /etc/alternatives/zeppelin-conf/interpreter.json
It makes sure that the Zeppelin uses python3.4 instead of python2.7. It works fine if I execute this in the terminal after SSH'ing to the master node, but it fails when I submit it as a Custom JAR step on the AWS Web interface. I get the following error:
ed: can't read /etc/alternatives/zeppelin-conf/interpreter.json
: No such file or directory
Command exiting with ret '2'
The same thing happens if I use
sudo sed -i '/zeppelin.pyspark.python/c\ \"zepplin.pyspark.python\" : \"python3\",' /etc/zeppelin-conf/interpreter.json
Obviously I could just change it from the Zeppelin UI, but I would like to include it in the bootstrap action.
Thanks!
It turns out that a bootstrap action submitted throug the AWS EMR web interface is submitted as a regular EMR step, so it's only run on the master node. This can be seen if you click the 'AWS CLI export' in the cluster web interface. The intended bootstrap action is listed as a regular step.
Using the command line to launch a cluster with a bootstrap action bypasses this problem, so I've just used that.
Edit: Looking back at the web interface, it's pretty clear that I was adding regular steps instead of bootstrap actions. My bad!
I'm trying to run my code of machine learning from images using tensorflow in Google CloudML. However, it seems the submitted job can't access to my files in my cloud shell or in GCS. Even though it is working fine in my local machine, I get the following error once I submit my job using the command gcloud from the cloud shell:
ERROR 2017-12-19 13:52:28 +0100 service IOError: [Errno 2] No such file or directory: '/home/user/pores-project-googleML/trainer/train.txt'
This folder can be found for sure in cloud shell, and I can check it when I type:
ls /home/user/pores-project-googleML/trainer/train.txt
I tried putting my file train.txt in GCS and access to it from my code (by specifying the path gs://my_bucket/my_path), but once the job submitted, I got a 'No such file or directory' error with the corresponding path.
To check where the job I submitted using gcloud is running, I added print(os.getcwd()) in the beginning of my python code trainer/task.py, which printed as a result in the logs: /user_dir. I couldn't find this path using the cloud shell, not even in GCS. So my question is how can I know in which machine my job is running? If it's in a certain container somewhere, how can I access from it to my files using the cloud shell and in GCS?
Before I do all of this, I succesfully completed the 'Image Classification using Flowers Dataset' tutorial.
The command I used to submit my job is:
gcloud ml-engine jobs submit training $JOB_NAME --job-dir $JOB_DIR --packages trainer-0.1.tar.gz --module-name $MAIN_TRAINER_MODULE --region us-central1
where:
TRAINER_PACKAGE_PATH=/home/use/pores-project-googleML/trainer
MAIN_TRAINER_MODULE="trainer.task"
JOB_DIR="gs://pores/AlexNet_CloudML/job_dir/"
JOB_NAME="census$(date +"%Y%m%d_%H%M%S")"
Regular Python IO library is not able to access files on GCS. Instead, you need to use GCS python client or gstuil cli to access GCS files.
Note that TensorFlow itself has native support of GCS (i.e., it can read GCS files directly).