I am running a Spark job on a two node standalone cluster (v 1.0.1).
Spark execution often gets stuck at the task mapPartitions at Exchange.scala:44.
This happens at the final stage of my job in a call to saveAsTextFile (as I expect from Spark's lazy execution).
It is hard to diagnose the problem because I never experience it in local mode with local IO paths, and occasionally the job on the cluster does complete as expected with the correct output (same output as with local mode).
This seems possibly related to reading from s3 (of a ~170MB file) immediately prior, as I see the following logging in the console:
DEBUG NativeS3FileSystem - getFileStatus returning 'file' for key '[PATH_REMOVED].avro'
INFO FileInputFormat - Total input paths to process : 1
DEBUG FileInputFormat - Total # of splits: 3
...
INFO DAGScheduler - Submitting 3 missing tasks from Stage 32 (MapPartitionsRDD[96] at mapPartitions at Exchange.scala:44)
DEBUG DAGScheduler - New pending tasks: Set(ShuffleMapTask(32, 0), ShuffleMapTask(32, 1), ShuffleMapTask(32, 2))
The last logging I see before the task apparently hangs/gets stuck is:
INFO NativeS3FileSystem: INFO NativeS3FileSystem: Opening key '[PATH_REMOVED].avro' for reading at position '67108864'
Has anyone else experience non-deterministic problems related to reading from s3 in Spark?
Related
I want to continue the DAG only if a csv file exists in S3, otherwise it should just end.
The DAG itself is being scheduled hourly.
with DAG(dag_id="my_dag",
start_date=datetime(2023, 1, 1),
schedule_interval='#hourly',
catchup=False
) as dag:
check_for_new_csv = S3KeySensor(
task_id='check_for_new_csv',
bucket_name='bucket-data',
bucket_key='*.csv',
wildcard_match=True,
soft_fail=True,
retries=1
)
start_instance = EC2StartInstanceOperator(
task_id="start_ec2_instance_task",
instance_id=INSTANCE_ID,
region_name=REGION
)
check_for_new_csv >> start_instance
But the sensor seems to run forever - in the log I can see it keeps on running:
[2023-01-10, 15:02:06 UTC] {s3.py:98} INFO - Poking for key : s3://bucket-data/*.csv
[2023-01-10, 15:03:08 UTC] {s3.py:98} INFO - Poking for key : s3://bucket-data/*.csv
Maybe the sensor in not the best choice for such logic?
A sensor is a perfect choice for this use case. I'd try setting the poke_interval and timeout to different smaller values than their default to make sure the sensor that Airflow is checking on the right intervals (by default, they are very long).
One thing to watch out for is if your sensors run on longer intervals than your schedule interval. For example, if your DAG is scheduled to run hourly, but your sensors's timeout is set for 2 hours, your next DAG run may not run as expected (depending on your concurrency and max_active_dag settings), or it may run unexpectedly because the sensor detects an older file. Ideally, you can append a timestamp in the name of your file to avoid this.
We have a Scala Spark application, that reads something like 70K records from the DB to a data frame, each record has 2 fields.
After reading the data from the DB, we make minor mapping and load this as a broadcast for later usage.
Now, in local environment, there is an exception, timeout from the RetryingBlockFetcher while running the following code:
dataframe.select("id", "mapping_id")
.rdd.map(row => row.getString(0) -> row.getLong(1))
.collectAsMap().toMap
The exception is:
2022-06-06 10:08:13.077 task-result-getter-2 ERROR
org.apache.spark.network.shuffle.RetryingBlockFetcher Exception while
beginning fetch of 1 outstanding blocks
java.io.IOException: Failed to connect to /1.1.1.1:62788
at
org.apache.spark.network.client.
TransportClientFactory.createClient(Transpor .tClientFactory.java:253)
at
org.apache.spark.network.client.
TransportClientFactory.createClient(TransportClientFactory.java:195)
at
org.apache.spark.network.netty.
NettyBlockTransferService$$anon$2.
createAndStart(NettyBlockTransferService.scala:122)
In the local environment, I simply create the spark session with local "spark.master"
When I limit the max of records to 20K, it works well.
Can you please help? maybe I need to configure something in my local environment in order that the original code will work properly?
Update:
I tried to change a lot of Spark-related configurations in my local environment, both memory, a number of executors, timeout-related settings, and more, but nothing helped! I just got the timeout after more time...
I realized that the data frame that I'm reading from the DB has 1 partition of 62K records, while trying to repartition with 2 or more partitions the process worked correctly and I managed to map and collect as needed.
Any idea why this solves the issue? Is there a configuration in the spark that can solve this instead of repartition?
Thanks!
I am having some trouble running a simple tHiveCreateTable job in Talend OS for Big Data (Print of the job where I am getting this error).
The Hive connection is fine and the job worked until Ranger was activated in the cluster.
After ranger, I started getting the following log:
[statistics] connecting to socket on port 3345
[statistics] connected
Error while processing statement: Cannot modify mapred.job.name at runtime. It is not in list of params that are allowed to be modified at runtime
[statistics] disconnected
This error occurs either using Tez or MapReduce for the job, throwing an exception in the following line of the automatically generated code:
// For MapReduce Mode
stmt_tHiveCreateTable_1.execute("set mapred.job.name=" + queryIdentifier);
Do you know any solution or workarround for this?
Thanks in advance
It is possible to disable changing mapreduce.job.name and hive.query.name at runtime by Talend7 jobs.
Edit the file
{talend_install_dir}/plugins/org.talend.designer.components.localprovider_7.1.1.20181026_1147/components/templates/Hive/SetQueryName.javajet
and comment out lines 6 and 11 like that:
// stmt_<%=cid %>.execute("set mapred.job.name=" + queryIdentifier_<%=cid %>);
// stmt_<%=cid %>.execute("set hive.query.name=" + queryIdentifier_<%=cid %>);
It solved this issue for me.
I have a Dataflow pipeline, running locally. The objective is to read a JSON file using TEXTIO, make sessions and load it into BigQuery. Given the structure I have to create a temp directory in GCS and then load it into BigQuery using that. Previously I had a data schema error that prevented me to load the data, see here. That issue is resolved.
So now when I run the pipeline locally it ends with dumping a temporary JSON newline delimited file into GCS. The SDK then gives me the following:
Starting BigQuery load job beam_job_xxxx_00001-1: try 1/3
INFO [main] (BigQueryIO.java:2191) - BigQuery load job failed: beam_job_xxxx_00001-1
...
Exception in thread "main" com.google.cloud.dataflow.sdk.Pipeline$PipelineExecutionException: java.lang.RuntimeException: Failed to create the load job beam_job_xxxx_00001, reached max retries: 3
at com.google.cloud.dataflow.sdk.Pipeline.run(Pipeline.java:187)
at pedesys.Dataflow.main(Dataflow.java:148)
Caused by: java.lang.RuntimeException: Failed to create the load job beam_job_xxxx_00001, reached max retries: 3
at com.google.cloud.dataflow.sdk.io.BigQueryIO$Write$WriteTables.load(BigQueryIO.java:2198)
at com.google.cloud.dataflow.sdk.io.BigQueryIO$Write$WriteTables.processElement(BigQueryIO.java:2146)
The errors are not very descriptive and the data is still not loaded in BigQuery. What is puzzling is that if I go to the BigQuery UI and load the same temporary file from GCS that was dumped by the SDK's Dataflow pipeline manually, in the same table, it works beautifully.
The relevant code parts are as follows:
PipelineOptions options = PipelineOptionsFactory.create();
options.as(BigQueryOptions.class)
.setTempLocation("gs://test/temp");
Pipeline p = Pipeline.create(options)
...
...
session_windowed_items.apply(ParDo.of(new FormatAsTableRowFn()))
.apply(BigQueryIO.Write
.named("loadJob")
.to("myproject:db.table")
.withSchema(schema)
.withCreateDisposition(BigQueryIO.Write.CreateDisposition.CREATE_IF_NEEDED)
.withWriteDisposition(BigQueryIO.Write.WriteDisposition.WRITE_APPEND)
);
The SDK is swallowing the error/exception and not reporting it to users. It's most likely a schema problem. To get the actual error that is happening you need to fetch the job details by either:
CLI - bq show -j job beam_job_<xxxx>_00001-1
Browser/Web: use "try it" at the bottom of the page here.
#jkff has raised an issue here to improve the error reporting.
I'm building out an ETL process with Pentaho Data Integration (CE) and I'm trying to operationalize my Transformations and Jobs so that they'll be able to be monitored. Specifically, I want to be able to catch any errors and then send them to an error reporting service like Honeybadger or New Relic. I understand how to do row-level error reporting but I don't see a way to do job or transaction failure reporting.
Here is an example job.
The down path is where the transformation succeeds but has row errors. There we can just filter the results and log them.
The path to the right is the case where the transformation fails all-together (e.g. DB credentials are wrong). This is where I'm having trouble: I can't figure out how to get the error info to be sent.
How do I capture transformation failures to be logged?
You can not capture job-level errors details inside the job itself.
However there are other options for monitoring.
First option is using database logging for transformations or jobs (see the "Log" tab in the job/trans parameters dialog) - this way you always have up-to-date information about the execution status so you can, say, write a job that periodically scans the logging database and sends error reports wherever you need.
Meanwhile this option seems to be something pretty heavy-weight for development and support and not too flexible for further modifications. So in our company we ended up with monitoring on a job-execution level - i.e. when you run a job with kitchen.bat and it fails by any reason you get an "error" status of execution of the kitchen, so you can easily examine it and perform necessary actions with whenever tools you'd like - .bat commands, PowerShell or (in our case) Jenkins CI.
You could use the writeToLog("e", "Message") function in the Modified Java Script step.
Documentation:
// Writes a string to the defined Kettle Log.
//
// Usage:
// writeToLog(var);
// 1: String - The Message which should be written to
// the Kettle Debug Log
//
// writeToLog(var,var);
// 1: String - The Type of the Log
// d - Debug
// l - Detailed
// e - Error
// m - Minimal
// r - RowLevel
//
// 2: String - The Message which should be written to
// the Kettle Log